CN121008176A

CN121008176A - A method and system for predicting battery static endurance time

Info

Publication number: CN121008176A
Application number: CN202511527087.2A
Authority: CN
Inventors: 周培; 翟霄雁; 李延红; 颜文静; 孔令兴
Original assignee: China National Heavy Duty Truck Group Jinan Power Co Ltd
Current assignee: China National Heavy Duty Truck Group Jinan Power Co Ltd
Priority date: 2025-10-24
Filing date: 2025-10-24
Publication date: 2025-11-25

Abstract

The invention discloses a battery static maintenance time prediction method and system, and mainly relates to the technical field of vehicle storage battery state monitoring and prediction. The method comprises the steps of collecting battery parameters of a vehicle at multiple power-down time and power-up time, classifying and collecting data according to the state of an electric switch of a battery power supply 30, training a self-feedback radial basis function neural network RBFNN model based on the collected battery parameter data, timely reading the battery parameters at the last power-down time during power-down of the vehicle, selecting a corresponding trained RBFNN model according to the state of the electric switch 30, inputting the battery parameters to predict static maintenance time, calculating the time difference from the current time to the last power-down time and remaining maintenance time, comparing the remaining maintenance time with a preset time threshold, and sending charge reminding information to a user through a cloud system. The invention has the beneficial effects that the problems that the battery is damaged and the vehicle cannot be started due to long-term standstill of the vehicle can be effectively avoided.

Description

Battery static maintenance time prediction method and system

Technical Field

The invention relates to the technical field of vehicle storage battery state monitoring and prediction, in particular to a method and a system for predicting static sustainable time of a commercial vehicle storage battery based on a self-feedback radial basis function neural network.

Background

Currently, monitoring and prediction techniques for commercial vehicle storage batteries rely primarily on Intelligent Battery Sensors (IBS). The sensor can monitor key parameters such as the state of charge (SOC), the state of health (SOH), the voltage, the current and the temperature of the battery in real time during the power-on operation of the vehicle. In terms of battery life prediction, prior art solutions are mostly based on vehicle operation data or charging process data. For example, patent CN112731154A proposes to predict battery life by analyzing actual running data of a vehicle under different working conditions, patent CN113779750a uses cooling parameters and charging time of a power battery during charging to evaluate remaining life of the battery, and patent CN110873841B adopts a method of combining data driving and battery characteristics to estimate state of health (SOH) and predict life.

However, the prior art has the following problems:

1. the existing IBS sensor and the prediction method matched with the existing IBS sensor are mainly activated and operated only when the vehicle is powered up and operated. After the vehicle is powered off after flameout, the system enters a dormant state, and the attenuation trend of the battery state cannot be continuously tracked and predicted in real time, so that the time fault of monitoring data is caused;

2. the battery power shortage risk early warning of a long-term stationary vehicle is insufficient, because continuous data of a battery state cannot be acquired during power-down, the existing system is difficult to accurately evaluate and predict the sustainable time (namely the time required by the fact that the electric quantity is reduced to a threshold value of the vehicle cannot be started) of the battery in the long-term stationary state, so that effective charging early warning cannot be timely and actively sent to a user;

3. The predictive model has poor adaptability to static discharge scenes, and the existing life predictive model is mostly aimed at the aging behavior of the battery in charge-discharge cycles or depends on rich data when the vehicle runs. For a complex static discharge process influenced by multiple factors such as ambient temperature, the self-health state of a battery, whether the battery is completely powered off (such as a 30-electric switch state) and the like during the stationary period of a vehicle, a specialized and high-precision prediction model is lacked, so that a large deviation exists between a prediction result and an actual attenuation condition.

Therefore, a method and a system for predicting the static sustainable time of a commercial vehicle storage battery based on a self-feedback radial basis function neural network are needed to solve the above problems.

Disclosure of Invention

The invention aims to provide a prediction method and a prediction system for static sustainable time of a commercial vehicle storage battery based on a self-feedback radial basis function neural network, which can effectively avoid the problems that the vehicle is at rest for a long time and the battery is damaged and the vehicle cannot be started.

The invention aims to achieve the aim, and the aim is achieved by the following technical scheme:

in one aspect, the invention provides a method for predicting static maintenance time of a battery, comprising the following steps:

step S1, acquiring battery parameters of a vehicle at a plurality of times of power-down time and power-up time through an IBS battery sensor and an intelligent communication platform carried by the vehicle, wherein the battery parameters comprise battery temperature, ambient temperature, battery health state SOH, power-down time battery SOC value, power-up time battery SOC value and system time, and acquiring data according to a battery power supply 30 electric switch state in a classified manner, wherein the 30 electric switch state comprises a natural attenuation state and a 30 electric switch state;

Step S2, training a self-feedback radial basis function neural network RBFNN model based on acquired battery parameter data, wherein the training process of the model comprises data preprocessing, model parameter initialization, forward propagation calculation, error calculation and parameter optimization;

step S3, during the power-down period of the vehicle, the battery parameters at the last power-down time are read at fixed time, the corresponding trained RBFNN model is selected according to the state of the 30 electric switch, and the battery parameters are input to predict the static maintenance time ;

Step S4, calculating the time difference from the current time to the last power-down timeCalculating remaining sustain timeComparison ofAnd a preset time thresholdWhen (when)And sending charging reminding information to the user through the cloud system.

Preferably, the step S1 includes:

step S11, after the vehicle is flameout, acquiring battery parameters at the power-down time, including a battery SOC value, SOH, battery temperature, environment temperature and system time;

Step S12, after the key power is turned on next time, battery parameters at the power-on time are collected, wherein the battery parameters comprise a battery SOC value, SOH, battery temperature, environment temperature and system time;

step S13, calculating the actual use time difference of the battery based on the system time of the power-down time and the power-up time I.e., from the power down time to the power up time,;

And step S14, respectively acquiring data of the battery in a natural attenuation state and a 30 electric opening state according to the 30 electric switching state, wherein the natural attenuation state corresponds to 30 electric closing and the 30 electric opening state corresponds to 30 electric opening.

Preferably, the preprocessing of data in the training process of the model in step S2 includes:

Identifying and eliminating abnormal values of the collected battery parameter data by adopting a sliding window filtering method;

average filtering is carried out on the data with the abnormal values removed to smooth the data;

Normalizing the data, and mapping the data to the data by adopting a Max_Min normalization method The interval, normalized formula is: Wherein As the raw data is to be processed,AndRespectively minimum and maximum values in the parameter data set;

The preprocessed data are divided into a training data set and a test data set according to the proportion of 8:2, wherein the training data set is used for model training, and the test data set is used for model verification and performance evaluation.

Preferably, the initializing of the model parameters in the training process of the model in step S2 includes:

initializing an RBFNN model structure, and setting the number of layers of the neural network to be 3, wherein the RBFNN model structure comprises an input layer, an hidden layer and an output layer;

The number of nodes of the input layer is 5, corresponding to the input vector WhereinThe temperature of the battery is indicated and,Indicating the temperature of the environment and,Indicating the state of health SOH of the battery,The SOC value of the battery at the last power-down time is indicated,Representing the SOC value of the battery at the next power-on time;

The number of output layer nodes is 1, and the output is the battery use time difference Namely, the time difference predicted by the model represents the predicted time difference from the power-down time to the power-up time;

The number of hidden layer nodes is 5, the hidden layer activation function is a gaussian function, for each hidden layer neuron Hidden layer neuronsOutput of (2)The calculation is as follows: Wherein Is the firstCenter vector of each neuron, and input vectorThe dimensions are the same asFirst, theThe width of the Gaussian function of each neuron controls the diffusion degree of the function;

Output of output layer I.e. the actual time difference is calculated as: Wherein For outputting layer weight, correspond toWeights of the implicit neurons;

at initialization, implicit layer center vector Random initialization, obeying uniform distribution, width of Gaussian functionInitialized to 1, output layer weightsThe random normal distribution with standard deviation of 1 is initialized, and the initial value of the bias parameter is 0.5.

Preferably, the parameter optimization in the training process of the model in step S2 includes:

setting an error index function ;

Setting the maximum iteration number of model training to 1000 times, and setting an error threshold to beWhen the number of iterations reaches 1000 or the errorStopping training when the training is stopped;

the weight parameters are updated by adopting a self-feedback adjustment algorithm, and the weight updating process comprises the following steps:

For each iteration Calculating an errorWeight pairGradient of (2)Due toThus, it is;

Calculating weight update itemWhereinFor the learning rate, the initial value is set to 0.01;

adding a self-feedback motion term to calculate WhereinAs a momentum factor, an initial value is set to 0.9;

the final weight is updated to ;

Momentum factor in self-feedback momentum termAccording to the adaptive updating of the error index function, the updating formula is as follows: Wherein As first order components of the motion term, calculated as,As a second order component, calculated as,AndThe attenuation factor is 0.9,The learning rate of the dynamic term is given as the value,To correct the term constant, take the value as。

Preferably, in step S2, the first RBFNN model and the second RBFNN model are trained according to the 30 electrical switch states, respectively:

The first RBFNN model is used for predicting static maintenance time of the battery in a natural attenuation state, and training data is from historical data when 30 electricity is shut down;

the second RBFNN model is used for predicting static maintenance time of the battery in a 30-electricity open state, and training data are from historical data of the battery in the 30-electricity open state;

The first RBFNN model and the second RBFNN model have the same network structure and comprise an input layer 5 node, an hidden layer 5 node and an output layer 1 node, wherein different weight parameters and biases are obtained through independent training;

During prediction, a model is dynamically selected according to the 30 electric state, so that the prediction accuracy and adaptability are improved.

Preferably, the step S3 specifically includes:

during the power-down period of the vehicle, reading battery parameters including battery temperature, ambient temperature, battery SOH and battery SOC value at the power-down time at the last power-down time through an intelligent communication platform at every 12 hours;

According to the 30 electric switch state stored by the intelligent communication platform, selecting and calling a corresponding RBFNN prediction model, if the 30 electric switch is closed, using a natural attenuation state model, and if the 30 electric switch is opened, using the 30 electric switch state model;

inputting battery parameters into a selected RBFNN model, wherein the model is input as WhereinIn order to be the temperature of the battery,In order to be at the temperature of the environment,In the case of a battery SOH,For the battery SOC value at the time of power down,Is a minimum startup value SOCm;

model output is the static hold time of the battery from the time of power down to the time when the battery charge drops to the minimum start-up value SOCm Wherein SOCm is a preset minimum battery start value, which represents the minimum battery SOC at which the vehicle can start.

Preferably, the step S4 specifically includes:

Recording the current time in real time through the upper computer system, and calculating the time difference from the current time to the last power-down time ,;

Calculating remaining sustain timeWhereinStatic hold time for model prediction;

comparison of And a preset time threshold,A time threshold value for reminding a user of charging is represented, and a typical value is 24 hours according to the requirement of the user;

When (when) Triggering early warning reminding, and sending the early warning state to an intelligent diagnosis and repair platform through an upper computer cloud system;

the intelligent diagnosis and repair platform sends charging reminding information to a set user through a short message to prompt the user to timely start the vehicle to charge the storage battery, and the problem that the battery loss affects the vehicle start is avoided.

On the other hand, the invention also provides a battery static maintenance time prediction system, which is used for realizing the battery static maintenance time prediction method, and comprises the following steps:

The data acquisition module is used for acquiring battery parameters of the vehicle at the time of multiple power-down and power-up through an IBS battery sensor and an intelligent communication platform carried by the vehicle, wherein the battery parameters comprise battery temperature, ambient temperature, battery health state SOH, battery SOC value at the time of power-down, battery SOC value at the time of power-up and system time, and acquiring data in a classified manner according to the state of the battery power switch 30;

the model training module is used for training a self-feedback radial basis function neural network RBFNN model based on the acquired data, wherein the RBFNN model takes battery parameters as input and outputs the static maintenance time from the power-down time to the time when the battery electric quantity is reduced to a minimum starting value SOCm ;

The prediction module is used for regularly reading the battery parameters at the last power-down time in the power-down period of the vehicle, selecting the corresponding trained RBFNN model according to the 30 electric switch state, inputting the battery parameters and predicting the static maintenance time;

The reminding module is used for calculating the time difference from the current time to the last power-down timeCalculating remaining sustain timeComparison ofAnd a preset time thresholdWhen (when)And sending charging reminding information to the user through the cloud system.

Preferably, the data acquisition module comprises an IBS battery sensor, a power domain controller PCU and an intelligent communication platform, wherein the IBS battery sensor monitors the SOC, SOH, voltage, current and temperature parameters of the battery, the SOC, SOH, voltage, current and temperature parameters are transmitted to the PCU through a LIN line at the power-on moment, the PCU controller forwards the parameters to a CAN line of the whole vehicle, and the intelligent communication platform records and stores the parameters of the battery from the CAN line in real time and reads the system time and the parameters of the battery at the power-off moment;

the model training module is deployed on the upper computer and is used for performing data preprocessing, model parameter initialization, RBFNN model training and parameter optimization, and the trained model parameters are stored in the upper computer;

The prediction module is deployed in the cloud system, reads battery parameters and 30 electric switch states of the last power-down time stored on the intelligent communication platform every 12 hours at regular time, selects and calls a corresponding RBFNN model, and predicts static maintenance time ;

The reminding module is deployed in the cloud system, calculates the time difference delta T0 between the current time and the power-down time, and calculates the remaining maintenance timeComparison ofAnd threshold valueWhen (when)And sending a short message to remind the user through the intelligent diagnosis and repair platform.

Compared with the prior art, the invention has the beneficial effects that:

1. The battery state continuous monitoring and early warning during the power-down period of the vehicle are realized:

By combining the IBS sensor historical data with an upper computer timing prediction mechanism, the limitation that the prior art cannot work during power-down is overcome, continuous evaluation of static maintenance time of a storage battery is realized, and the battery safety management capability of the vehicle in a long-term static state is remarkably improved.

2. Accuracy and reliability of static maintenance time prediction are improved:

the self-feedback RBF neural network is adopted for modeling, so that nonlinear characteristics of the battery, which are affected by multiple factors such as temperature, SOH, SOC and the like in the static attenuation process, can be effectively captured. Through the self-adaptive adjustment and classification training of the dynamic terms (distinguishing 30 electric switch states), the model is more attached to the actual attenuation law, and the prediction precision and scene adaptability are improved.

3. The timeliness of user reminding and the reliability of vehicle use are enhanced:

Through calculating remaining maintenance time and comparing with the set threshold value, can be before the battery electric quantity is insufficient leads to unable start, initiatively send the warning of charging to the user, make the user intervene in advance, effectively avoid the unable start problem of vehicle because of battery deficiency causes, promoted user experience and the reliability of vehicle trip.

4. Battery life and use cost are optimized:

through accurate prediction and timely reminding of the static maintenance time of the battery, a user can avoid overdischarge of the battery, effectively delay the aging speed of the battery, prolong the service life of the battery and reduce the replacement cost and use interruption loss caused by early failure of the battery.

5. The adaptability and the robustness of the system under different scenes are improved:

By respectively training the prediction models by distinguishing 30 electric switch states, the system can adapt to different static power consumption scenes (such as a normal sleep state and an unexpected non-power-off state), and the generalization capability of the prediction models in practical application and the overall robustness of the system are enhanced. .

Drawings

FIG. 1 is a flow chart of the method of the present invention;

Fig. 2 is a schematic diagram of the system architecture of the present invention.

Detailed Description

The application will be further illustrated with reference to specific examples. It is to be understood that these examples are illustrative of the present application and are not intended to limit the scope of the present application. Further, it will be understood that various changes and modifications may be made by those skilled in the art after reading the teachings of the application, and equivalents thereof fall within the scope of the application as defined by the claims.

In the present invention, terms such as "upper", "lower", "left", "right", "front", "rear", "vertical", "horizontal", "side", "bottom", etc. refer to an orientation or a positional relationship based on that shown in the drawings, and are merely relational terms, which are used for convenience in describing structural relationships of various components or elements of the present invention, and do not denote any one of the components or elements of the present invention, and are not to be construed as limiting the present invention.

Example 1:

As shown in fig. 1, the embodiment provides a battery static maintenance time prediction method, which comprises the following implementation steps:

1. and a data acquisition stage:

Through an IBS battery sensor carried on a commercial vehicle, after the vehicle is flameout every time, acquiring battery parameters at the power-down time, including a battery SOC value of 75%, an SOH value of 92%, a battery temperature of 25 ℃, an ambient temperature of 20 ℃ and a system time of 2024-01-1518:30:00';

when the vehicle is started by opening a key for the next time, acquiring battery parameters at the power-on time, including a battery SOC value 68%, an SOH value 92%, a battery temperature 18 ℃, an ambient temperature 15 ℃ and a system time of 2024-01-1809:15:00';

In order to ensure that the training target is consistent with the predicted target, the embodiment additionally records the actual time from the power-down time to the battery SOC falling to the preset minimum starting value SOCm (set to 50%) as the training tag in the data acquisition stage. Obtaining the data through a standard discharge test under a laboratory constant temperature environment;

and respectively acquiring data according to the state of the 30 electric switch, wherein the data are recorded as natural attenuation state data when the 30 electric switch is closed, and the data are recorded as 30 electric open state data when the 30 electric switch is opened.

2. Model training stage:

preprocessing the collected 300 groups of historical data, namely dividing the collected 300 groups of historical data into a training set (front 240 groups) and a testing set (rear 60 groups) according to time sequence, so as to avoid the problem of data leakage;

The average value filtering is carried out on the residual data, and Max-Min normalization is used for mapping the data to a [ -1,1] interval;

initializing two RBFNN models specific to static hold time prediction, input vector The output layer is adjusted to directly predict the static maintenance time from power-down to SOCm corresponding to the battery temperature, the ambient temperature, SOH, power-down SOC and SOCm respectively;

model training is carried out by using 240 groups of training data, the maximum iteration number is set to 1000, and the error threshold value is set The weight is updated by adopting a self-feedback regulation algorithm, and the learning rate is improvedMomentum factorInitial value 0.9, by the formulaSelf-adaptive updating;

the prediction error on the test set is 0.015, and the precision requirement is met.

3. Prediction execution stage:

during the power-down period of the vehicle, the last power-down time parameter is read through an intelligent communication platform every 12 hours, wherein the battery temperature is 10 ℃, the ambient temperature is 8℃, SOH & lt 90 & gt, and the power-down SOC is 80%;

detecting 30 electric closing states, and selecting a natural attenuation state prediction model;

Input vector Wherein the fifth parameter 50 isSet to 50% (vehicle minimum start value);

model output prediction static hold time Hours.

4. A reminding triggering stage:

Calculating the time difference between the current time and the power-down time Hours;

calculating remaining sustain time Hours;

comparison of And a preset threshold valueFor an hour, as 35 is greater than 24, the reminding is not triggered temporarily;

the prediction was re-made after 12 hours, The time is less than 24 hours, the early warning is triggered, and a short message is sent to a user through the intelligent diagnosis platform, wherein the electric quantity of a vehicle storage battery is low, the vehicle storage battery cannot be started after 23 hours, and the vehicle storage battery is charged in time. ".

Example 2:

as shown in fig. 2, the present embodiment provides a battery static maintenance time prediction system, including:

the data acquisition module comprises an IBS battery sensor, a power domain controller PCU and an enhanced intelligent communication platform;

The intelligent communication platform is added with SOCm parameter storage functions, the actual discharge time corresponding to different SOCm is recorded in the data acquisition stage, and accurate label data are provided for model training;

the PCU controller receives IBS sensor data through the LIN bus and forwards the IBS sensor data to the whole vehicle CAN bus, and the data transmission frequency is dynamically adjusted according to the vehicle state, wherein the data transmission frequency is 1Hz in the power-on period and 0.1Hz in the power-off period so as to reduce the power consumption.

The model training module is deployed on a high-performance upper computer provided with an Intel i7 processor and a 32GB memory;

A specially designed training algorithm is used to ensure that the training target (time to SOCm) is exactly the same as the predicted target. Training data are strictly divided into time sequences, wherein the first 80% is used for training, and the second 20% is used for testing;

two training modes, namely batch training based on historical data and incremental training based on real-time data, are supported, and model accuracy is continuously optimized.

The prediction module is deployed on an Ariyan cloud end server and adopts a micro-service architecture;

the SOCm parameters input during prediction are consistent with the training stage, so that the consistency of the model input space is ensured;

And calling an intelligent communication platform API every 12 hours regularly, and acquiring the latest data by adopting a sliding time window mechanism to ensure the continuity of a time sequence.

The data acquisition module comprises an IBS battery sensor, a power domain controller PCU and an intelligent communication platform, wherein the IBS battery sensor monitors the SOC, SOH, voltage, current and temperature parameters of the battery, the SOC, SOH, voltage, current and temperature parameters are transmitted to the PCU through a LIN wire at the power-on moment, the PCU controller forwards the parameters to a CAN wire of the whole vehicle, the intelligent communication platform records and stores the parameters of the battery from the CAN wire in real time, and the system time and the parameters of the battery are read at the power-off moment;

The model training module is deployed on the upper computer and used for performing data preprocessing, model parameter initialization, RBFNN model training and parameter optimization, and the trained model parameters are stored in the upper computer;

While the preferred embodiment of the present application has been described in detail, the present application is not limited to the embodiments described above, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the present application, and these equivalent modifications and substitutions are intended to be included in the scope of the present application as defined in the appended claims.

Claims

1. A method for predicting the static lifespan of a battery, characterized by comprising the following steps:

Step S1: Using the vehicle's IBS battery sensor and intelligent communication platform, collect battery parameters at multiple power-off and power-on times. The battery parameters include battery temperature, ambient temperature, battery health status (SOH), battery SOC value at power-off time, battery SOC value at power-on time, and system time. Collect data according to the battery power 30 switch status, which includes natural degradation state and 30 switch on state, where 30 switch is the constant power supply.

Step S2: Based on the collected battery parameter data, train the self-feedback radial basis function neural network (RBFNN) model. The training process of the model includes: data preprocessing, model parameter initialization, forward propagation calculation, error calculation, and parameter optimization.

Step S3: During vehicle power-off, periodically read the battery parameters at the most recent power-off moment, select the corresponding trained RBFNN model based on the 30-volt switch status, and input the battery parameters to predict the static maintenance time. ;

Step S4: Calculate the time difference between the current time and the most recent power-off time. Calculate the remaining duration ,Compare With preset time threshold ,when At that time, charging reminders are sent to users via a cloud system.

2. The battery static sustaining time prediction method according to claim 1, characterized in that step S1 includes:

Step S11: After the vehicle is turned off, collect the battery parameters at the moment of power-off, including battery SOC value, SOH, battery temperature, ambient temperature and system time.

Step S12: After the vehicle is turned on again, collect the battery parameters at the time of power-on, including battery SOC value, SOH, battery temperature, ambient temperature and system time.

Step S13: Calculate the difference in actual battery usage time based on the system time at the power-off and power-on times. That is, from the moment of power-off to the moment of power-on. ;

Step S14: Based on the 30-watt switch status, collect data on the battery in the natural degradation state and the 30-watt on state respectively, where the natural degradation state corresponds to the 30-watt off state and the 30-watt on state corresponds to the 30-watt on state.

3. The battery static sustaining time prediction method according to claim 1, characterized in that, the data preprocessing during model training in step S2 includes:

Outliers were identified and removed from the collected battery parameter data using a sliding window filtering method.

Mean filtering is applied to the data after outlier removal to smooth the data.

The data is normalized using the Max_Min normalization method, mapping the data to... The normalization formula for the interval is: ,in This is the original data. and These are the minimum and maximum values in the parameter dataset, respectively;

The preprocessed data was divided into training and testing datasets in an 8:2 ratio. The training dataset was used for model training, and the testing dataset was used for model validation and performance evaluation.

4. The battery static sustaining time prediction method according to claim 3, characterized in that, the model parameter initialization during model training in step S2 includes:

Initialize the RBFNN model structure, setting the number of neural network layers to 3, including an input layer, hidden layers, and an output layer;

The input layer has 5 nodes, corresponding to the input vector. ,in Indicates battery temperature. Indicates ambient temperature. This indicates the battery's state of health (SOH). This indicates the battery's SOC value at the most recent power-off moment. Indicates the battery's SOC value at the next power-on time;

The output layer has 1 node, and the output is the battery usage time difference. That is, the time difference predicted by the model, which represents the predicted time difference from the moment of power-off to the moment of power-on;

The hidden layer has 5 nodes, and the activation function is a Gaussian function. For each hidden layer neuron... Hidden layer neurons Output The calculation is as follows: ,in For the first The center vector of each neuron, and the input vector Same dimension, for No. The width of the Gaussian function for each neuron controls the degree of function diffusion;

Output of the output layer The actual time difference is calculated as follows: ,in These are the output layer weights, corresponding to the [number]th layer. The weights of each hidden neuron;

During initialization, the hidden layer center vector Random initialization, following a uniform distribution, with a Gaussian function width. Initialize to 1, output layer weights It is initialized as a random normal distribution with a standard deviation of 1, and the initial value of the bias parameter is 0.5.

5. The battery static sustaining time prediction method according to claim 4, characterized in that, parameter optimization during model training in step S2 includes:

Set error index function ;

The maximum number of iterations for model training is set to 1000, and the error threshold is set to... When the number of iterations reaches 1000 or the error... Training should be stopped immediately.

The weight parameters are updated using a self-feedback adjustment algorithm. The weight update process includes:

For each iteration Calculation error weight gradient ,because ,therefore ;

Calculate weight update term ,in The learning rate is initially set to 0.01.

Add a self-feedback momentum term and calculate ,in The momentum factor is initially set to 0.9.

The final weights are updated to ;

Momentum factor in self-feedback momentum term The update is based on the adaptive update of the error index function, and the update formula is as follows: ,in The first component of the momentum term is calculated as follows: , For second-order components, the calculation is as follows: , and This is the attenuation factor, with a value of 0.9. The learning rate for the momentum term takes a value of , The constant for the correction term has a value of [value missing]. .

6. The battery static sustaining time prediction method according to claim 1, characterized in that, in step S2, the first RBFNN model and the second RBFNN model are trained according to the 30-phase switch state respectively:

The first RBFNN model is used to predict the static sustaining time of the battery under natural degradation conditions, and the training data comes from historical data when the battery is turned off at 30°C.

The second RBFNN model is used to predict the static sustaining time of the battery in the 30% charge-on state, and the training data comes from historical data when the battery is in the 30% charge-on state;

The first RBFNN model and the second RBFNN model have the same network structure, including 5 nodes in the input layer, 5 nodes in the hidden layer, and 1 node in the output layer, but they obtain different weight parameters and biases through independent training;

During prediction, a model is dynamically selected based on the 30 electrical states to improve prediction accuracy and adaptability.

7. The method for predicting battery static maintenance time according to claim 1, wherein step S3 specifically includes:

During the vehicle's power-off period, the battery parameters at the most recent power-off time are read through the intelligent communication platform every 12 hours, including battery temperature, ambient temperature, battery SOH, and battery SOC value at the time of power-off.

Based on the 30-voltage switch status stored in the Smart Connect platform, the corresponding RBFNN prediction model is selected and called. If the 30-voltage switch is off, the natural decay state model is used; if the 30-voltage switch is on, the 30-voltage on state model is used.

Input the battery parameters into the selected RBFNN model. The model input is... ,in For battery temperature, For ambient temperature, For battery SOH, This represents the battery's SOC value at the moment of power-off. The minimum startup value is SOCm;

The model output is the static sustaining time of the battery from the moment of power-off until the battery charge drops to the minimum starting value SOCm. Where SOCm is the preset minimum battery starting value, representing the minimum battery SOC at which the vehicle can start.

8. The method for predicting battery static sustaining time according to claim 1, wherein step S4 specifically includes:

The host computer system records the current time in real time and calculates the time difference between the current time and the most recent power-off time. , ;

Calculate the remaining duration ,in The static duration predicted by the model;

Compare With preset time threshold , This indicates the time threshold for reminding users to charge, which can be set according to user needs, with a typical value of 24 hours.

when When an early warning is triggered, the warning status is sent to the intelligent diagnosis and repair platform via the host computer cloud system.

The intelligent diagnostic and repair platform sends charging reminders to designated users via SMS, prompting them to start the vehicle in time to charge the battery and avoid battery depletion affecting vehicle starting.

9. A battery static sustaining time prediction system, used to implement the battery static sustaining time prediction method as described in any one of claims 1-8, characterized in that it comprises:

The data acquisition module is used to collect battery parameters of the vehicle at multiple power-off and power-on times through the vehicle's IBS battery sensor and intelligent communication platform. The battery parameters include battery temperature, ambient temperature, battery health status (SOH), battery SOC value at power-off time, battery SOC value at power-on time, and system time. The data is collected according to the battery power switch status.

The model training module is used to train a self-feedback radial basis function neural network (RBFNN) model based on the collected data. The RBFNN model takes battery parameters as input and outputs the static maintenance time of the battery from the moment of power-off until the battery charge drops to the minimum starting value SOCm. ;

The prediction module is used to periodically read the battery parameters at the most recent power-off time during vehicle power-off periods, select the corresponding trained RBFNN model based on the 30-volt switch state, and input the battery parameters to predict the static maintenance time. ;

The reminder module is used to calculate the time difference between the current time and the most recent power-off time. Calculate the remaining duration ,Compare With preset time threshold ,when At that time, charging reminders are sent to users via a cloud system.

10. A battery static sustaining time prediction system according to claim 9, characterized in that the data acquisition module includes an IBS battery sensor, a power domain controller (PCU), and an intelligent communication platform. The IBS battery sensor monitors the battery's SOC, SOH, voltage, current, and temperature parameters, and transmits them to the PCU via a LIN bus at power-on. The PCU controller forwards these parameters to the vehicle's CAN bus. The intelligent communication platform records and saves the battery parameters in real time from the CAN bus, and reads the system time and battery parameters at power-off.

The model training module is deployed on the host computer and is used to perform data preprocessing, model parameter initialization, RBFNN model training and parameter optimization. The trained model parameters are saved to the host computer.

The prediction module is deployed in a cloud system. Every 12 hours, it reads the battery parameters and 30V power switch status stored on the smart communication platform at the time of the most recent power-off, selects and calls the corresponding RBFNN model, and predicts the static maintenance time. ;

The reminder module is deployed in a cloud system, calculates the time difference ΔT0 between the current time and the power-off time, and calculates the remaining duration. ,Compare With threshold ,when At that time, the system will send a text message to remind the user through the intelligent diagnosis and repair platform.