KR102815228B1 - Method, apparatus, and system for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model - Google Patents

Abstract

A hybrid battery control system based on a hydrogen fuel cell utilizing an artificial intelligence model may include a memory and a processor storing instructions. The system may perform a reversible energy conversion process of generating hydrogen by electrolyzing water, storing the generated hydrogen, and generating electricity using the stored hydrogen when necessary. The system may receive first sensing data including at least one of voltage, current, and temperature from a plurality of sensors for monitoring an operating state of a hydrogen fuel cell, and may receive second sensing data from sensors for monitoring a state of an energy storage device. The system analyzes the first sensing data and the second sensing data in real time using an artificial intelligence model implemented in the form of an AI agent capable of autonomous judgment and control to generate control parameters, and if the system's power is insufficient based on the control parameters, supplies stored hydrogen to a hydrogen fuel cell to generate power, and if the system has surplus power, controls the system to generate and store hydrogen through water electrolysis using the surplus power, and processes real-time sensor data without delay through an edge computing device installed in the system and performs power balancing between the cells of the energy storage device.

Description

METHOD, APPARATUS, AND SYSTEM FOR CONTROLLING A HYDROGEN FUEL CELL-BASED HYBRID BATTERY USING AN ARTIFICIAL INTELLIGENCE MODEL

The present invention relates to a hybrid battery system combining a hydrogen fuel cell and a battery and a control method thereof, and more particularly, to a control method and device for analyzing the output characteristics of a hydrogen fuel cell and the charge/discharge characteristics of a battery in real time using an artificial intelligence model and optimizing the efficiency of the entire system based on the analysis.

Recently, as carbon neutrality policies have been strengthened worldwide, the use of hydrogen fuel cells as a clean energy source is increasing. In particular, hydrogen fuel cells are attracting attention in various fields due to their high energy density and fast charging characteristics, but hydrogen fuel cells alone have a fundamental limitation of unstable output.

Specifically, hydrogen fuel cells have the characteristic that the output current fluctuates greatly depending on the change in hydrogen supply pressure, so a separate control device is required for stable power supply. In addition, hydrogen fuel cells have the output characteristics of low voltage and high current, making it difficult to utilize them directly.

Conventional battery management systems (BMS) simply monitor the state of charge (SOC) and state of health (SOH) of the battery and provide only basic protection functions to prevent overcharge and overdischarge. In addition, the existing passive balancing method has the problem of low energy efficiency and slow cell-to-cell balancing speed.

Against this backdrop, a new type of hybrid battery system is required that can complement the unstable output characteristics of hydrogen fuel cells and improve the efficiency of the entire system.

The present invention is intended to solve the problems of the prior art as described above, and has the purpose of providing a hybrid system that effectively combines a hydrogen fuel cell and a battery.

Another object of the present invention is to provide an intelligent control system that analyzes the output characteristics of a hydrogen fuel cell and the state of a battery in real time by utilizing an artificial intelligence model and derives optimal control parameters based on the analysis.

Another object of the present invention is to improve the performance and lifespan of the entire system by efficiently performing energy redistribution between battery cells by applying active balancing technology.

A hybrid battery control system based on a hydrogen fuel cell utilizing an artificial intelligence model may include a memory and a processor storing instructions. The system generates hydrogen by electrolyzing pure water and stores the hydrogen, receives first sensing data including at least one of voltage, current, and temperature from a plurality of sensors for monitoring an operating state of a hydrogen fuel cell, and receives second sensing data from sensors for monitoring a state of an energy storage device. The system analyzes the first sensing data and the second sensing data in real time using an artificial intelligence model implemented in the form of an AI agent capable of autonomous judgment and control to generate control parameters, and when the system has insufficient power based on the control parameters, supplies stored hydrogen to the hydrogen fuel cell to generate power, and when the system has surplus power, controls to generate and store hydrogen by electrolyzing water using the surplus power, processes real-time sensor data without delay through an edge computing device installed in the system, and performs power balancing between cells of the energy storage device.

The system may be replaced by an electronic device (e.g., a terminal, a computer). The operations of the system executed by instructions may be classified into methods.

The hydrogen fuel cell-based hybrid battery system according to the present invention can provide a stable and sustainable power supply by combining the advantages of a hydrogen fuel cell and a battery. In particular, the high energy density of the hydrogen fuel cell and the stable output characteristics of the battery work complementarily to each other, thereby improving the overall performance of the system.

The hydrogen fuel cell-based hybrid battery system according to the present invention can maximize the efficiency of the system through real-time analysis and control using an artificial intelligence model. By learning the pattern of sensor data and deriving optimal control parameters, adaptive control according to the situation can be performed.

The hydrogen fuel cell-based hybrid battery system according to the present invention can improve the life and performance of the battery by applying active balancing technology. In addition, by efficiently performing energy redistribution between cells, the overall performance of the battery pack can be optimized.

The hydrogen fuel cell-based hybrid battery system according to the present invention can provide stable power supply in various usage environments. It can continuously provide power even in environments where power supply is limited, such as during a power outage, outdoor activities, or emergency situations.

FIG. 1 is a diagram for explaining a hydrogen fuel cell-based hybrid battery control system utilizing an artificial intelligence model according to an embodiment.
Figure 2 is a diagram for explaining learning of a neural network according to an embodiment.
Figure 3 is a diagram illustrating the configuration of an artificial intelligence model according to an embodiment.
Figure 4 is a block diagram showing the configuration of a hydrogen fuel cell-based hybrid battery control system using an artificial intelligence model according to an embodiment.
FIG. 5 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.
FIG. 6 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.
Fig. 7 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.

Hereinafter, various embodiments will be described in detail with reference to the attached drawings. However, these embodiments can be modified in various ways, and therefore the scope of rights of the patent application is not limited or restricted by these embodiments.

The terms "first" or "second" may be used to describe various components, but this is simply to distinguish each component. For example, a first component may be referred to as a second component, and vice versa. When a component is referred to as being "connected" to another component, this means that it may be directly connected or there may be other components intervening. The terms used are for descriptive purposes only and are not intended to be limiting. The singular includes the plural unless the context clearly indicates otherwise. It should be understood that the terms "comprises" or "has" herein indicate the presence of a feature or element described in the specification, but do not exclude the presence of other features or elements.

Unless otherwise specifically defined, all terms used herein have the meaning generally understood by a person of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted in the context of the relevant technology and shall not be interpreted in an overly formal sense unless explicitly defined in this application. In addition, when describing drawings, the same reference numerals are used for identical components and duplicate descriptions are omitted. When describing embodiments, if a detailed description of a related technology becomes unnecessarily unclear, the detailed description may be omitted.

The embodiment can be implemented in various forms such as personal computers, laptops, tablets, smartphones, televisions, smart home appliances, intelligent cars, kiosks, wearable devices, etc. An artificial intelligence (AI) system is a computer system that implements human-level intelligence, and unlike existing rule-based smart systems, it is a system in which machines learn and make judgments on their own. As the use of AI systems increases, their recognition rate improves and they understand user preferences more accurately, so existing systems are gradually being replaced by deep learning-based AI.

Artificial intelligence technology consists of machine learning and various element technologies that utilize it. Machine learning refers to an algorithm that classifies and learns the characteristics of input data on its own, and element technologies are technologies that mimic human cognitive and judgment functions through machine learning algorithms such as deep learning. These technologies include linguistic understanding, visual understanding, inference and prediction, knowledge representation, and motion control.

The following are several fields in which AI technology is applied. Linguistic understanding is a technology that recognizes and processes human language, including natural language processing, machine translation, conversational systems, question-answering, and speech recognition and synthesis. Visual understanding is a technology that recognizes and processes objects, including object recognition, tracking, image search, person recognition, and scene understanding. Inference prediction is a technology that judges and predicts information, including knowledge-based inference, optimization prediction, and recommendation systems. Knowledge representation is a technology that automatically processes human experience information, including knowledge construction and management. Motion control is a technology that controls the movement of autonomous vehicles or robots, including navigation, collision avoidance, and driving control. In general, in order to apply machine learning algorithms to real life, they are learned through the Trial and Error method. In particular, deep learning requires hundreds of thousands of repeated executions. If it is difficult to implement this in a real environment, a virtual environment is implemented on a computer and learning is conducted through simulation.

In the present invention, artificial intelligence (AI) refers to a technology that implements human learning, reasoning, and perception capabilities into a computer, and includes concepts such as machine learning and symbolic logic. Machine learning (ML) is an algorithm technology that learns the characteristics of input data on its own. AI technology can analyze input data through such machine learning algorithms, learn the results, and make judgments or predictions based on them. In addition, technologies that mimic human cognitive and judgment functions using machine learning algorithms can also be included in the category of artificial intelligence. For example, this includes fields such as linguistic understanding, visual understanding, reasoning and prediction, knowledge representation, and motion control. Machine learning refers to the process of training a neural network model through experience in processing data. Through this, computer software can improve its data processing capabilities on its own. A neural network model models the correlation between data and can be expressed by multiple parameters. A neural network model extracts and analyzes features from given data to derive relationships between data, and repeating this process to optimize the parameters of the model is the core of machine learning. For example, a neural network model can learn relationships between inputs and outputs, or learn relationships by deriving regularities from input data alone.

Artificial intelligence learning models or neural network models are designed to implement the human brain structure on a computer, and include multiple network nodes that mimic neurons. These nodes are connected to each other to send and receive signals, and process data through layers of various depths in the artificial intelligence learning model. These models may include artificial neural networks, convolutional neural networks (CNNs), etc. For example, artificial intelligence learning models can be machine-learned in supervised learning, unsupervised learning, reinforcement learning, etc. Machine learning algorithms that can be used include decision trees, Bayesian networks, support vector machines, artificial neural networks, AdaBoost, perceptrons, genetic programming, and clustering.

CNN is a type of multilayer perceptron designed with minimal preprocessing, consisting of one or more convolutional layers and general artificial neural network layers. This structure allows CNN to effectively utilize two-dimensional input data, and it shows excellent performance in the fields of images and speech. CNN is trained through standard backpropagation, and has the advantage of being easier to train than other feedforward artificial neural network techniques and using a small number of parameters. Convolutional networks are neural networks that contain a set of nodes with bound parameters, and many computer vision tasks have been improved as the amount of training data increases and computational power improves. Overfitting is not important in today's large-scale data sets, and increasing the network size improves test accuracy. As optimal utilization of computing resources becomes a limiting factor, distributed and scalable implementations of deep neural networks are required.

FIG. 1 is a diagram for explaining a hydrogen fuel cell-based hybrid battery control system utilizing an artificial intelligence model according to one embodiment.

As illustrated in FIG. 1, a hydrogen fuel cell-based hybrid battery control system (100) utilizing an artificial intelligence model may include multiple user terminals (110-1, ??), a server (120), and a database (130). According to one embodiment, the database (130) is configured separately from the server (120), but it is not necessarily configured in this manner, and the database (130) may be located within the server (120). For example, the server (120) may include multiple artificial intelligences for performing a machine learning algorithm. According to another embodiment, multiple user terminals (110-1, ??), the server (120), and the database (130) may be connected to each other so as to be able to communicate with each other through a network (N).

The network (N) can perform wireless or wired communication among multiple user terminals (110-1, ??), a server (120), and a database (130). For example, the network can support wireless communication in a manner such as LTE (long-term evolution), LTE-A (LTE Advanced), CDMA (code division multiple access), WCDMA (wideband CDMA), WiBro (Wireless BroadBand), WiFi (wireless fidelity), Bluetooth (Bluetooth), NFC (near field communication), GPS (Global Positioning System), or GNSS (global navigation satellite system). In addition, the network (N) can perform wired communication in a manner such as USB (universal serial bus), HDMI (high definition multimedia interface), RS-232 (recommended standard 232), or POTS (plain old telephone service).

The database (130) has a function of storing various data. The data stored in the database (130) is data acquired, processed, or used by at least one component of multiple user terminals (110-1, ??) and the server (120), and may include software (e.g., a program). The database (130) may include volatile and/or nonvolatile memory.

Figure 2 is a diagram for explaining learning of a neural network according to an embodiment.

As shown in Fig. 2, the learning device can train a neural network (123) to process review responses received from multiple user terminals (110-1, ??) for each item. In addition, the learning device can train a neural network (123) to extract a user's stay record based on the user's movement path information. According to one embodiment, the learning device can be a separate entity from the server (120), but this is not limited.

The neural network (123) includes an input layer (121) and an output layer (125), and a training sample is input to generate a training output, and learning is performed based on the difference between the output and the label. The label can be defined in the user's stay history according to the review response-related items and movement path information, and the neural network (123) is composed of several node groups defined by weights and activation functions between the nodes.

The learning device can learn the neural network (123) using the Gradient Descent (GD) or Stochastic Gradient Descent (SGD) technique. GD is a method that adjusts the parameters of the model using the entire data set, and SGD is a method that enables faster learning using only some randomly selected data. The learning device can calculate the training error through a loss function designed in the form of Mean Square Error (MSE) or entropy, and find and optimize the weights that affect the training error using the Backpropagation technique.

In one embodiment, the learning device extracts a first object from the review response to obtain a first label, and performs learning with the first training output generated by applying the first label to the first neural network. In addition, the learning device extracts a second object from the movement path information to obtain a second label, and performs learning with the second training output generated by applying the second label to the second neural network.

The learning device can generate a first training feature vector based on the composition, location, and pattern features of the review response, and a second training feature vector based on the composition, length, and pattern features of the movement path information. These feature vectors are applied to each neural network to generate training outputs, and are used for learning the review item extraction algorithm and the user stay history acquisition algorithm together with the corresponding labels. At this time, backpropagation refers to a learning method that adjusts the weights of each layer while propagating the error generated in the output layer toward the input layer.

FIG. 3 is a diagram illustrating the configuration of an artificial intelligence model according to one embodiment.

According to one embodiment, the artificial intelligence model may consist of an input layer, a hidden layer, and an output layer.

The input layer is a layer related to values input to the artificial intelligence model. In the hidden layer, a feature map can be generated by performing a MAC operation (multiply-accumulate) and an activation operation on the input values. The MAC operation is a process of multiplying the input values by the corresponding weights and then adding the multiplied values. The activation operation is a process of inputting the result of the MAC operation into the activation function and outputting the final result, and various forms of activation functions can be used. For example, activation functions may include, but are not limited to, the sigmoid function, the tangent function, the Relu function, the Riki-Relu function, the Maxout function, and the Elu function.

The hidden layer may be composed of at least one layer, and for example, when divided into a first hidden layer and a second hidden layer, the first hidden layer performs a MAC operation and an activation operation based on an input value of an input layer to generate a feature map, and this feature map can be used as an input value of the second hidden layer. The second hidden layer can perform a MAC operation and an activation operation based on a feature map generated as a result of the first hidden layer.

The output layer may be a layer related to the results of operations performed in the hidden layer.

In one embodiment, a learning model learns syllable (character) patterns that are frequently used together in a given corpus to automatically identify boundaries between compound words and named entities, and integrates object information of a first UI source and object information rendered in a browser to generate a learning object information file. Using this learning object information file, data for deep learning network training is generated, and data received from various domains is standardized into a unified format according to one or more standardization methods suitable for each domain. Thereafter, data of a specific domain can be learned and inferred, information required for standardization in the domain can be determined, and post-processing can be performed on data received from various domains.

The first UI source includes an XML file, and the learning object information file includes an input JSON file for feature learning and an output JSON file used as correct answer (Label) data during learning. This output JSON file is a file containing DOM Tree information of HTML implemented in compliance with web standards. Various domains include at least one of RAN (radio access network), transport, or core, and the post-processing process may include a correlation function.

Figure 4 is a block diagram showing the configuration of a hydrogen fuel cell-based hybrid battery control system using an artificial intelligence model according to an embodiment.

The system (400) according to one embodiment may include a processor (420) and a memory (430), and some of the illustrated configurations may be omitted or replaced. The system (400) according to one embodiment may be a server or a terminal. According to one embodiment, the processor (420) is a configuration capable of performing operations or data processing related to control and/or communication of each component of the system (400), and may be configured with one or more processors. The memory (430) may store information related to the above-described method or store a program in which the above-described method is implemented. The memory (430) may be a volatile memory or a nonvolatile memory. The memory (430) may store various file data, and the stored file data may be updated according to the operation of the processor (120).

According to one embodiment, the processor (420) can execute a program and control the device (400). The code of the program executed by the processor (120) can be stored in the memory (430). The operations of the processor (420) can be performed by loading instructions stored in the memory (430). The system (400) can be connected to an external device (e.g., a personal computer or a network) through an input/output device (not shown in the drawing) and exchange data.

A system (400) according to an embodiment includes a processor (420) and a memory (430). The system (400) according to an embodiment may be the server or terminal described above. The processor (420) may include at least one of the devices described above through FIGS. 1 to 3, or may perform at least one method described above through FIGS. 1 to 3. The memory (430) may store information related to the method described above, or may store a program in which the method described above is implemented. The memory (430) may be a volatile memory or a nonvolatile memory.

A system (400) according to an embodiment includes a processor (420) and a memory (430). The system (400) according to an embodiment may be the server or terminal described above. The processor (420) may include at least one of the devices described above through FIGS. 1 to 3, or may perform at least one method described above through FIGS. 1 to 3. The memory (430) may store information related to the method described above, or may store a program in which the method described above is implemented. The memory (430) may be a volatile memory or a nonvolatile memory.

FIG. 5 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.

Although the process steps, method steps, algorithms, etc., are illustrated in a sequential order in the flowchart of FIG. 5, such processes, methods, and algorithms may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention need not be performed in the order described herein. Furthermore, even if some steps are illustrated as being performed asynchronously, in other embodiments such some of the steps may be performed concurrently. Furthermore, the illustration of a process by depiction in the drawings is not intended to exclude other changes and modifications thereto, nor is it intended to imply that the illustrated process or any of its steps is essential to one or more of the various embodiments of the present invention, nor is it intended to imply that the illustrated process is preferred.

In operation 510, the system (400) can receive first sensing data including voltage, current, and temperature from a plurality of sensors for monitoring the operating status of the hydrogen fuel cell and second sensing data from sensors for monitoring the status of the energy storage device.

The system (400) can monitor and control the status of each component in real time in a hybrid battery system combining a hydrogen fuel cell and an energy storage device. In order to monitor the operating status of the hydrogen fuel cell, first sensing data including stack voltage, output current, stack temperature, hydrogen injection pressure, oxygen concentration, catalyst deterioration, and electrolyte membrane resistance can be collected through a plurality of sensors. At this time, the output current can fluctuate significantly depending on the hydrogen injection pressure due to the characteristics of the hydrogen fuel cell, and the system (400) can detect and compensate for such output fluctuation in real time.

In the case of an energy storage device, the system (400) can collect second sensing data including cell voltage, cell current, state of charge (SOC), state of health (SOH), cell temperature, internal resistance, and self-discharge rate of a lithium ion battery. In addition, third sensing data including voltage, current, temperature, and equivalent series resistance of a supercapacitor can be collected. Here, the supercapacitor can play a role in buffering rapid output fluctuations of a hydrogen fuel cell, because the output fluctuations due to pressure changes are large due to the characteristics of a hydrogen fuel cell.

In operation 520, the system (400) can analyze the first sensing data and the second sensing data in real time using an artificial intelligence model to generate control parameters.

The system (400) can analyze the collected sensing data in real time using an artificial intelligence model based on a large language model (LLM) to predict the maximum power point of a hydrogen fuel cell, the optimal charge/discharge pattern of a battery, and the power compensation time of a supercapacitor. In particular, by analyzing weather information and the user's past 24-hour power usage data, it can predict the power demand by time zone for the next 24 hours, and based on this, it can optimize the distribution ratio of each power source.

In operation 530, the system (400) can control power distribution between the hydrogen fuel cell and the energy storage device based on the generated control parameters and perform power balancing between cells of the energy storage device.

For cell balancing of an energy storage device, an active balancing circuit based on a bidirectional DC-DC converter with a switching frequency of 100 kHz can be applied. At this time, a zero voltage switching method using an LLC resonant topology can be used to achieve a power conversion efficiency of 95% or more. The system (400) can perform precise balancing control to maintain the voltage deviation between cells to 10 mV or less by analyzing the state of each cell at a cycle of 500 ms.

In addition, the system (400) can control the coolant flow rate and cooling fan speed by monitoring the difference between the stack temperature and the ambient temperature in real time to prevent the performance of the hydrogen fuel cell from deteriorating due to temperature. When the temperature difference exceeds 15 degrees Celsius, the load sharing ratio is reduced in steps of 10% to control the output of the hydrogen fuel cell, and the humidifier temperature is maintained at 5 degrees Celsius lower than the stack temperature to maintain a humidifying condition of 95% or higher relative humidity.

The system (400) can drive a life prediction model based on a recurrent neural network for battery life management and calculate a prediction confidence interval using a particle filtering technique. Through this, the remaining life of each cell group can be quantitatively predicted, and the charge/discharge allowable current can be differentially applied between 0.1C and 1C according to the predicted life. Through an abnormality detection model based on a long short-term memory (LSTM) neural network, an abnormal pattern of a cell can be detected early, and a dangerous situation such as thermal runaway can be prevented in advance.

According to one embodiment, the system (400) can generate hydrogen by electrolyzing pure water and store it. Here, the electrolysis of water can be performed by applying a voltage between the anode and the cathode to decompose pure water (H2O) into hydrogen (H2) and oxygen (O2). At this time, the voltage and current used for electrolysis can be controlled by the AI agent to an optimized value considering the power status of the system and the amount of hydrogen stored.

According to one embodiment, the system (400) can utilize multiple sensors to monitor the operating status of the hydrogen fuel cell in real time. Specifically, the stack voltage is a key indicator of the output performance of the fuel cell, the current density is a key indicator of the amount of current generated per unit area, the stack temperature is a key indicator of the thermal management status of the system, and the hydrogen injection pressure and oxygen concentration are key indicators of the reaction efficiency, and these data can be collected in real time at a sampling frequency of 200 kHz or higher. In addition, the alternating current impedance spectroscopy can be applied in a wide frequency range from 1 Hz to 100 kHz to measure the impedance change inside the fuel cell.

According to one embodiment, the system (400) can collect voltage, current, SOC, SOH, temperature, and internal resistance data of each cell at a cycle of 500 ms to monitor the state of the energy storage device. Here, SOC can be used as an indicator of the current state of charge of the battery, and SOH can be used as an indicator of the lifespan and performance degradation of the battery. In particular, cell voltage and current data can be precisely measured through a bidirectional DC-DC converter capable of high-frequency switching of 100 kHz.

According to one embodiment, the system (400) can analyze sensor data collected through an AI agent in real time to generate parameters necessary for system control. At this time, the AI agent is implemented as a hybrid structure combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network, so as to learn time series patterns of sensor data and predict future states. In particular, the remaining life of each cell can be quantitatively predicted through a battery life prediction model based on a recurrent neural network and a particle filtering technique.

According to one embodiment, the system (400) can perform energy conversion in both directions according to the power demand and supply status. In the case of a power shortage, stored hydrogen is supplied to a fuel cell to generate electricity, and at this time, a power conversion efficiency of 95% or more can be achieved by a zero voltage switching method using an LLC resonant topology. On the other hand, in the case of surplus power, it can be used to generate and store hydrogen by electrolyzing water.

According to one embodiment, the system (400) can perform real-time data processing and decision-making through an edge computing device. The edge device processes sensor data with a response speed of less than 5 ms, and in particular, can perform balancing control in real time to maintain a voltage difference between cells to less than 10 mV. In addition, energy can be efficiently redistributed between cell groups through an active balancing circuit based on a DC-DC converter.

FIG. 6 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.

Although the process steps, method steps, algorithms, etc., are illustrated in the flowchart of FIG. 6 in a sequential order, such processes, methods, and algorithms may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention need not be performed in the order described herein. Furthermore, even if some steps are illustrated as being performed asynchronously, in other embodiments such some of the steps may be performed concurrently. Furthermore, the illustration of a process by depiction in the drawings does not imply that the illustrated process is exclusive of other changes and modifications thereto, nor does it imply that the illustrated process or any of its steps is essential to one or more of the various embodiments of the present invention, nor does it imply that the illustrated process is preferred.

In operation 610, the system (400) can analyze a variation pattern of output current according to a pressure change of the hydrogen fuel cell from the first sensing data using a machine learning-based model. The system (400) can monitor and control the state of the hydrogen fuel cell using the machine learning-based model. Specifically, the system (400) can analyze a variation pattern of output current according to a pressure change of the hydrogen fuel cell, because the output current of the hydrogen fuel cell can greatly vary depending on the hydrogen injection pressure due to the characteristics of the hydrogen fuel cell. A hydrogen fuel cell can generally generate a high current of about 20 A even at a voltage of 3.5 V, and thus it can be important to detect and control such rapid output variation in real time.

In operation 620, the system (400) analyzes the state of charge (SOC) and the state of health (SOH) of the energy storage device from the second sensing data, and controls the output of the hydrogen fuel cell and the charging and discharging of the energy storage device based on the analyzed data. The system (400) can analyze the state of charge (SOC) and the state of health (SOH) to monitor the state of the energy storage device. Here, the SOC is an indicator representing the current charge amount of the battery, and the SOH is an indicator representing the lifespan and the degree of performance degradation of the battery. The system (400) can adjust the output of the hydrogen fuel cell and optimize the charging and discharging of the energy storage device based on the analyzed data. In particular, since charging and discharging in a state where the SOC is too high or too low can shorten the lifespan of the battery, it can be controlled to operate within an appropriate SOC range.

In operation 630, the system (400) can control the power distribution ratio between the lithium ion battery and the supercapacitor, and perform power balancing between cells in an active balancing manner using a DC-DC converter. The system (400) can control power distribution of a hybrid energy storage device composed of a lithium ion battery and a supercapacitor. The supercapacitor has the characteristic of being capable of instantaneous high current charging and discharging, and thus can play a role in buffering rapid output fluctuations of a hydrogen fuel cell. The system (400) can perform power balancing between each cell in an active balancing manner using a DC-DC converter, which has the advantages of high energy efficiency and fast balancing speed compared to a passive balancing manner.

In operation 640, the system (400) monitors the first sensing data and the second sensing data in real time using an integrated management unit linked with a smartphone application to detect abnormal signs exceeding preset thresholds, and provides a notification according to a predefined warning stage for the detected abnormal signs through the smartphone application.

According to one embodiment, the system (400) can provide an integrated management function that is linked with a smartphone application. The user can monitor the real-time operating status of the system through the smartphone application, and can receive an immediate notification when an abnormality exceeding a threshold value of various sensor data is detected. For example, when the cell voltage is out of a safe range, the temperature rises excessively, or the hydrogen pressure fluctuates abnormally, a notification can be provided to the user according to a predefined warning stage. This can improve the safety of the system and increase the convenience of the user.

According to one embodiment, the system (400) can monitor the state of the energy storage device in real time through an artificial intelligence model based on a large language model (LLM). At this time, the charge/discharge current value of the cell can be collected in real time from the current sensor, and the cell voltage value can be collected from the voltage sensor, respectively, to calculate the state of charge (SOC) of the battery. The system (400) can dynamically control the charge/discharge current of the battery between 0.1C and 3C according to the calculated SOC value. The system (400) can manage the ratio of the charge/discharge current to the rated capacity of the battery. In addition, the system (400) can collect real-time power generation data from a power meter that monitors the power generation amount of a solar power generation facility, and determine whether surplus power is generated compared to the current power demand.

According to one embodiment, the system (400) can perform learning using a deep reinforcement learning algorithm by combining past power usage data and weather forecast data received from the Korea Meteorological Administration. Through this, power demand by time zone can be predicted, and the optimal time for energy conversion can be determined by comprehensively considering the predicted demand and the current system status (e.g., battery SOC, stored hydrogen amount, etc.). The system (400) can process sensor data in real time by utilizing an edge computing device having a fast response speed of, for example, less than 5 milliseconds, thereby distributing the computational load of the central server and minimizing network delay. In particular, the edge device can operate an anomaly detection model based on a long short-term memory (LSTM) neural network to detect abnormal patterns of cells early and prevent dangerous situations such as thermal runaway in advance.

According to one embodiment, the system (400) can be utilized for various purposes such as a medium-sized ESS for home use or an auxiliary battery for vehicles. In particular, when operated in conjunction with a solar power generation facility, it can directly supply solar power during the day when the weather is good or generate hydrogen using surplus power. The generated hydrogen can be utilized to supply power during times when solar power generation is not possible, such as at night or during rainy weather. In addition, the stability of the power system can be improved by generating power using the stored hydrogen during peak times when power demand is concentrated. This energy conversion and storage process can be automatically controlled in units of a specific time (e.g., 15 minutes) through a multi-objective optimization based on a genetic algorithm, and at this time, the real-time generation cost and carbon dioxide emissions for each power source can be utilized as major input variables for optimization.

Fig. 7 is a flowchart illustrating a method for controlling a hydrogen fuel cell-based hybrid battery using an artificial intelligence model according to one embodiment.

Although the process steps, method steps, algorithms, etc., are illustrated in a sequential order in the flowchart of FIG. 7, such processes, methods, and algorithms may be configured to operate in any suitable order. In other words, the steps of the processes, methods, and algorithms described in various embodiments of the present invention need not be performed in the order described herein. Furthermore, even if some steps are illustrated as being performed asynchronously, in other embodiments such some of the steps may be performed concurrently. Furthermore, the illustration of a process by depiction in the drawings is not intended to exclude other changes and modifications thereto, nor is it intended to imply that the illustrated process or any of its steps is essential to one or more of the various embodiments of the present invention, nor is it intended to imply that the illustrated process is preferred.

In operation 710, the system (400) can collect sensing data including output voltage, output current, temperature, etc. through each sensor group from the hydrogen fuel cell, the battery, and the supercapacitor. The system (400) can collect various sensor data from each component of the hybrid battery. In the hydrogen fuel cell, data of a first sensor group including stack voltage, output current, stack temperature, hydrogen injection pressure, oxygen concentration, catalyst deterioration degree, electrolyte membrane resistance, etc. can be collected. In the battery, data of a second sensor group including cell voltage, cell current, state of charge (SOC), state of health (SOH), cell temperature, internal resistance, and self-discharge rate can be collected, and in the supercapacitor, data of a third sensor group including voltage, current, temperature, and equivalent series resistance can be collected.

In operation 720, the system (400) can analyze the collected sensing data in real time through a large-scale language model (LLM) based on an artificial neural network to predict the maximum power point of the hydrogen fuel cell, the optimal charge/discharge pattern of the battery, and the power compensation time of the supercapacitor. The system (400) can analyze the collected sensor data through a large-scale language model (LLM) based on an artificial neural network to predict the optimal operating point of each component. In the case of the hydrogen fuel cell, efficient operation is possible by finding the maximum power point, and the battery can derive the optimal charge/discharge pattern considering the lifespan and efficiency. In addition, the supercapacitor can predict the optimal intervention time to compensate for the rapid output fluctuation of the hydrogen fuel cell.

In operation 730, the system (400) can adjust the power load sharing ratio in real time and supply the surplus power of the hydrogen fuel cell to the electrolysis device to produce and store hydrogen. The system (400) can optimize the load sharing between each power source in real time. When the output of the hydrogen fuel cell decreases below a set voltage threshold, the discharge current of the battery can be increased to maintain a stable power supply. In addition, when the SOC of the battery reaches the upper threshold, the surplus power of the hydrogen fuel cell can be supplied to the electrolysis device to produce and store hydrogen, thereby maximizing energy utilization.

In operation 740, the system (400) performs active balancing using a DC-DC converter when the voltage deviation between battery cells exceeds a preset reference value, and performs fault diagnosis based on deep learning to classify the fault and execute a response measure. The system (400) can apply an active balancing technology based on a DC-DC converter for cell balancing of the battery. When the voltage deviation between cells exceeds a preset reference value, balancing can be performed by directly transferring energy of a high-voltage cell to a low-voltage cell. In addition, an abnormal state can be detected early through a deep-learning-based fault diagnosis system, and a fault type can be classified to execute an appropriate response measure.

In operation 750, the system (400) can learn an optimal operation pattern considering life and efficiency based on past operation data of the hydrogen fuel cell, battery, and supercapacitor, thereby maximizing the energy efficiency of the system. The system (400) can learn an operation pattern optimizing life and efficiency by analyzing past operation data of each component. Through this, an integrated control strategy can be established to minimize catalyst deterioration of the hydrogen fuel cell, extend the life of the battery, and maximize the energy storage efficiency of the supercapacitor. In addition, the overall energy efficiency of the system can be improved and operating costs can be reduced.

According to one embodiment, the system (400) implements the artificial intelligence model as a Large Language Model, extracts first performance indicators including stack voltage, current density, oxygen concentration, hydrogen concentration, stack temperature, coolant temperature, and humidifier temperature of a hydrogen fuel cell from the first sensing data, extracts second performance indicators including cell voltage, cell temperature, charge/discharge current, internal resistance, and self-discharge rate of an energy storage device from the second sensing data, constructs time series data of the first performance indicators and the second performance indicators in real time, and uses the time series data as input so that the Large Language Model can interpret the operating state of the system as a numerical parameter and calculate a control parameter value, and converts the calculated control parameter value into an electrical control signal.

According to one embodiment, the system (400) can optimize the performance of a hybrid battery system through an artificial intelligence model based on a Large Language Model. From a hydrogen fuel cell, first performance indicators such as stack voltage, current density, oxygen and hydrogen concentration, stack temperature, coolant temperature, and humidifier temperature can be collected. Here, the stack voltage can be a key indicator representing the output performance of the fuel cell, the current density can be a key indicator representing the amount of current generated per unit area, the gas concentration can be a key indicator representing the reaction efficiency, and various temperatures can be a key indicator representing the thermal management status of the system.

According to one embodiment, the energy storage device can extract second performance indicators such as cell voltage, cell temperature, charge/discharge current, internal resistance, and self-discharge rate. Time series data of these performance indicators can be constructed in real time and used as input to a large language model, and the model can interpret the operating state of the system as numerical parameters based on this and calculate optimal control parameter values. The calculated parameter values can be converted into electrical control signals for controlling actual hardware.

According to one embodiment, the system (400) controls the instantaneous output of the hydrogen fuel cell and the charge/discharge power of the energy storage device in real time based on the control signal, analyzes weather information and the user's past 24-hour power usage data to predict the power demand by time zone for the next 24 hours, compares the predicted power demand with the current available capacity of the hydrogen fuel cell and the energy storage device to determine the power distribution ratio by time zone, and performs Gaussian process regression analysis based on the state of charge (SOC) and state of health (SOH) for each cell group of the energy storage device to quantitatively predict the remaining life of the cells.

According to one embodiment, the system (400) can perform real-time power management of hydrogen fuel cells and energy storage devices based on the generated control signal. In particular, by analyzing weather information and the user's past 24-hour power usage pattern, the power demand by time zone for the next 24 hours can be predicted. The optimal power distribution ratio can be determined by comparing this predicted information with the current available capacity of each power source. In addition, in order to manage the life of the energy storage device, the remaining life of the cell can be quantitatively predicted by performing Gaussian process regression analysis based on the remaining capacity (SOC) and the state of aging (SOH) of each cell group.

According to one embodiment, the system (400) differentially applies the charge/discharge allowable current of each cell group between 0.1 C and 1 C according to the predicted remaining lifespan, maintains the voltage deviation between the cell groups to 50 mV or less through active balancing using a supercapacitor, monitors the difference between the stack temperature and the ambient temperature in real time to control the coolant flow rate and the cooling fan speed to prevent the performance of the hydrogen fuel cell from deteriorating due to temperature, and controls the output of the hydrogen fuel cell by gradually reducing the load sharing ratio in units of 10% when the temperature difference exceeds 15 degrees Celsius, and maintains the humidifier temperature 5 degrees Celsius lower than the stack temperature to maintain a humidifying condition of 95% or more relative humidity.

According to one embodiment, the system (400) can differentially apply the charge and discharge current amount according to the predicted remaining life of the cells. A low current of 0.1C is allowed to a group of cells whose life has decreased more, and a current of up to 1C is allowed to a group of cells in good condition to balance the overall life. Here, the C-rate is an index indicating the ratio of the charge and discharge current to the rated capacity of the battery. In addition, the voltage deviation between the cell groups can be precisely maintained at 50 mV or less through active balancing using a supercapacitor. In order to manage the temperature of the hydrogen fuel cell, the difference between the stack temperature and the ambient temperature can be monitored in real time to control the coolant flow rate and the cooling fan speed. When the temperature difference exceeds 15 degrees Celsius, the load sharing ratio can be gradually reduced in units of 10% to prevent overheating. In addition, by maintaining the humidifier temperature 5 degrees Celsius lower than the stack temperature, the optimal relative humidity of 95% or more can be maintained, thereby optimizing the ion conductivity of the electrolyte membrane.

According to one embodiment, the system (400) monitors in real time the output voltage and current fluctuations of the hydrogen fuel cell at a sampling frequency of 200 kHz or higher, measures the internal impedance change of the hydrogen fuel cell by AC impedance spectroscopy in a frequency band of 1 Hz to 100 kHz, and analyzes the frequency spectrum and time domain characteristics of the measured voltage and current waveforms using an artificial intelligence model having a convolutional neural network (CNN) structure, thereby predicting the output reduction rate of the fuel cell.

The system (400) can precisely monitor and analyze the performance of a hydrogen fuel cell. With a high sampling frequency of 200 kHz or more, minute fluctuations in output voltage and current can be captured in real time, and impedance changes inside the fuel cell can be measured by applying AC impedance spectroscopy in a wide frequency range from 1 Hz to 100 kHz. Through such high-frequency sampling and impedance analysis, the state of each component of the fuel cell, such as the hydration state of the electrolyte membrane, the activity of the catalyst layer, and the mass transfer characteristics of the gas diffusion layer, can be precisely diagnosed.

According to one embodiment, the system (400) can analyze the frequency spectrum and time domain characteristics of measured voltage and current waveforms by utilizing an artificial intelligence model of a convolutional neural network (CNN) structure. Through this, the deterioration pattern of the fuel cell can be learned and the rate of decline in output performance can be predicted in advance. For cell balancing of the energy storage device, a bidirectional DC-DC converter capable of high-frequency switching of 100 kHz can be used. In particular, by applying the LLC resonant topology and the zero voltage switching method, a high power conversion efficiency of 95% or more can be achieved.

According to one embodiment, the system (400) applies an active balancing circuit based on a bidirectional DC-DC converter having a switching frequency of 100 kHz for cell balancing of the energy storage device, and achieves a power conversion efficiency of 95% or more by using a zero-voltage switching method using an LLC resonant topology, and the artificial intelligence model analyzes voltage, current, temperature, internal resistance, and state of charge (SOC) and state of health (SOH) data of each cell at a cycle of 500 ms to perform balancing control with a voltage deviation of 10 mV or less between cells, and drives a battery life prediction model based on a recurrent neural network through real-time data synchronization between distributed edge computing nodes and a cloud server having a response speed of less than 5 ms, and can calculate a prediction confidence interval using a particle filtering technique.

According to one embodiment, the artificial intelligence model of the system (400) can analyze the status data of each cell at a 500 ms cycle to perform precise voltage balancing of 10 mV or less. Distributed edge computing nodes can process data at a fast response speed of less than 5 ms and synchronize it with a cloud server in real time to drive a battery life prediction model based on a recurrent neural network. The confidence interval of the life prediction can be quantitatively calculated through a particle filtering technique.

According to one embodiment, the system (400) controls the charge and discharge current variably within the range of 0.1C to 3C according to the remaining capacity and load demand of the battery by using a deep reinforcement learning algorithm, determines the load distribution ratio between power sources every 15 minutes through a genetic algorithm-based multi-objective optimization that uses real-time generation cost and carbon dioxide emissions of each power source as input variables, detects abnormal patterns of cell voltage, current, and temperature through an anomaly detection model based on a long short-term memory (LSTM) neural network, and generates an alarm 10 minutes before the occurrence of thermal runaway, and implements a digital twin based on MATLAB and Simulink to simulate the real-time state with a time synchronization of less than 1 ms with an actual battery.

According to one embodiment, the system (400) can optimize the charging and discharging of the battery through a deep reinforcement learning algorithm. The charging and discharging current can be dynamically adjusted in a range from 0.1C to a maximum of 3C depending on the remaining capacity and the load demand, and the optimal load distribution ratio between each power source can be determined in 15-minute units through multi-objective optimization using a genetic algorithm. At this time, the real-time generation cost and carbon dioxide emissions of each power source can be considered to simultaneously optimize economic efficiency and environmental friendliness.

According to one embodiment, the system (400) can secure the safety of the battery through an abnormality detection model based on a long short-term memory (LSTM) neural network. It can detect abnormal patterns of cell voltage, current, and temperature early and generate an alarm 10 minutes before thermal runaway occurs. By implementing a digital twin system using MATLAB and Simulink, it is possible to simulate the state of an actual battery in real time with precise time synchronization within 1 ms, and through this, the performance of the system can be predicted and optimized.

The examples described above may be implemented by hardware and software components or a combination thereof. For example, the devices, methods, and components mentioned may be implemented by a general-purpose or special-purpose computer, such as a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor (DSP), a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding. The processing unit may execute an operating system (OS) and one or more software applications running on the operating system. In addition, the device may perform functions to access, store, manipulate, process, and generate data in accordance with the execution of the software. For ease of understanding, the processing unit may be described as one, but one of ordinary skill in the art will recognize that multiple processing elements or different types of processing elements may be included. For example, the processing unit may consist of multiple processors, or a processor and a controller. In addition, other processing configurations, such as parallel processors, may be used.

The software may include computer programs, codes, instructions, or a combination thereof, which may be used to configure or independently or collectively command a processing device in a desired manner. The software and data may be interpreted by the processing device, or may be permanently or temporarily embodied in various machines, components, physical or virtual devices, computer storage media, or transmission signals to provide instructions and data. The software may be distributed and stored or executed on a network-connected computer system, and may be stored on one or more computer-readable recording media.

Although the above examples are illustrated by specific drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on them. For example, appropriate results can be obtained even if the described techniques are performed in a different order than presented, or if the components of the described systems, structures, devices, and circuits are combined in different forms, or if other components or equivalents are substituted.

Claims (3)

A hydrogen fuel cell-based hybrid battery control system using an artificial intelligence model
Memory for storing instructions, and
Contains a processor,
The above instructions, when executed by the processor, cause the system to
It performs a reversible energy conversion process that generates hydrogen by electrolyzing water, stores the generated hydrogen, and generates electricity using the stored hydrogen when needed.
Receive first sensing data including at least one of voltage, current, and temperature from a plurality of sensors for monitoring the operating status of a hydrogen fuel cell,
Receive second sensing data from sensors for monitoring the status of the energy storage device,
By using an artificial intelligence model implemented in the form of an AI agent capable of autonomous judgment and control, the first sensing data and the second sensing data are analyzed in real time to generate control parameters,
Based on the above control parameters, when the system's power is insufficient, the stored hydrogen is supplied to the hydrogen fuel cell to generate power, and when the system has surplus power, the surplus power is used to generate and store hydrogen through water electrolysis.
Processes real-time sensor data without delay through edge computing devices installed in the system and controls power balancing between cells of the energy storage device.
Using the above artificial intelligence model
The state of charge (SOC) of the energy storage device is monitored based on real-time data collected from current sensors and voltage sensors, and the charge/discharge current is variably controlled within a set range according to the state of charge.
Monitor the amount of power supplied from solar power generation facilities connected to the system in real time and determine whether there is surplus power compared to demand power.
By learning past power usage patterns and weather information, it predicts power demand by time zone for a set period, and based on this, it determines the optimal time for hydrogen generation or power generation.
By distributing real-time sensor data from edge computing devices installed in the system, the load on the central server is reduced and real-time local decision-making is performed.
A system that controls power management linked to solar power generation facilities, energy storage through two-way conversion between hydrogen and electricity, and autonomous load response during peak electricity demand hours when used as a medium-sized ESS for home use or as an auxiliary battery for vehicles.
delete In paragraph 1,
The above energy storage device
Including lithium-ion batteries and supercapacitors,
The above lithium ion battery stores power supplied from the hydrogen fuel cell,
The above supercapacitor buffers the output fluctuations of the hydrogen fuel cell,
The above lithium ion battery and the above supercapacitor are characterized in that they are connected in parallel and operate,
The above system analyzes the variation pattern of the output current according to the pressure change of the hydrogen fuel cell from the first sensing data using the artificial intelligence model,
The state of charge (SOC) and state of health (SOH) of the energy storage device are analyzed from the second sensing data.
Based on the above analyzed data, the output of the hydrogen fuel cell and the charging/discharging of the energy storage device are controlled,
Controlling the power distribution ratio between the lithium ion battery and the supercapacitor,
Using the above artificial intelligence model, the operating parameters including voltage, current, temperature and internal resistance of each cell of the energy storage device are analyzed in real time to perform power balancing between cells.
The above cell-to-cell power balancing is characterized by being performed using an active balancing method using a DC-DC converter.
By using an integrated management unit linked to a smartphone application, the first sensing data and the second sensing data are monitored in real time to detect abnormal signs exceeding a preset threshold value,
Notifications are provided through the above smartphone application according to predefined warning levels for detected abnormal signs.
Control the power distribution and balancing of the system based on the control parameters provided from the above artificial intelligence model,
A system that controls the efficiency optimization data generated by analyzing the operation data of the above system to provide it as learning data for the above artificial intelligence model.