CN120409810A

CN120409810A - An intelligent production task scheduling method based on time series knowledge base

Info

Publication number: CN120409810A
Application number: CN202510530557.4A
Authority: CN
Inventors: 战腾; 赵立刚; 邵永刚; 刘成明
Original assignee: Shandong Inspur Intelligent Production Technology Co ltd
Current assignee: Shandong Inspur Intelligent Production Technology Co ltd
Priority date: 2025-04-25
Filing date: 2025-04-25
Publication date: 2025-08-01

Abstract

The present invention discloses an intelligent production task scheduling method based on a time series knowledge base, comprising: obtaining production laws based on equipment operation data, production task data, and environmental data collected in real time through time series analysis; querying the knowledge base based on the production laws to obtain a production task scheduling plan; and adjusting the time series knowledge base based on the execution of the production task scheduling plan. The present invention significantly improves the dynamic adaptability of production task scheduling through a closed-loop process of real-time data collection, time series analysis, dynamic query of the knowledge base, and feedback iteration. Multi-dimensional data fusion and time series law mining generate accurate scheduling plans. The knowledge base continuously learns new data, adapts to long-term changes in the production environment, reduces manual intervention, and improves decision-making accuracy.

Description

Intelligent production task scheduling method based on time sequence knowledge base

Technical Field

The invention belongs to the technical field of intelligent production task scheduling, and particularly relates to an intelligent production task scheduling method based on a time sequence knowledge base.

Background

With the rapid development of industry 4.0 and intelligent manufacturing technology, the intellectualization and dynamic adaptability of production scheduling become the core demands for improving the competitiveness of the manufacturing industry. Conventional production task scheduling methods rely mostly on manually set static rules or empirical models to perform tasks through predefined priorities and resource allocation policies. However, such methods have significant limitations in facing complex and diverse production environments.

Conventional scheduling systems typically perform task allocation based on fixed rules and historical experience, and are difficult to respond in real time to equipment failures, environmental parameter fluctuations, or dynamic changes such as sudden task demands. For example, when an abnormality occurs in the production equipment or the environmental conditions (such as temperature, humidity and pressure) exceed a preset range, the system cannot quickly adjust the scheduling policy, resulting in production interruption or resource waste.

In addition, the prior art uses multiple stay single point or segment analysis of production data, failing to fully exploit the time-series dependency and potential rules in time-series data. For example, key information such as periodic fluctuation of the running state of the equipment, bottleneck rules of task execution and the like is not systematically extracted, so that the scheduling decision lacks data-driven accuracy.

In summary, the existing production scheduling technology has significant shortcomings in dynamic adaptability, data driving decision capability and the like, and an intelligent production task scheduling method is needed to realize dynamic adaptation of production task scheduling.

Disclosure of Invention

The invention provides an intelligent production task scheduling method based on a time sequence knowledge base, which aims to solve the problems of insufficient dynamic adaptability of the method caused by difficulty in real-time response to dynamic change and insufficient data driving decision accuracy caused by neglecting time sequence dependency and potential rules.

The technical scheme adopted by the invention is as follows:

An intelligent production task scheduling method based on a time sequence knowledge base comprises the following steps:

according to the equipment operation data, the production task data and the environment data acquired in real time, analyzing by a time sequence method to obtain a production rule;

inquiring a knowledge base according to the production rule to obtain a production task scheduling scheme;

And adjusting the time sequence knowledge base according to the execution condition of the production task scheduling scheme.

The intelligent production task scheduling method based on the time sequence knowledge base disclosed by the invention also has the following additional technical characteristics:

inquiring a knowledge base according to the production rule to obtain a production task scheduling scheme, wherein the production task scheduling scheme specifically comprises the following steps:

according to the data acquired in real time, a plurality of primary selection scheduling schemes corresponding to the production rule are obtained through the similarity of the real-time data and cluster classification of a plurality of scheduling schemes in a knowledge base;

And according to the multiple primary selection scheduling schemes, combining corresponding execution condition data in the knowledge base to obtain a production task scheduling scheme.

The knowledge base specifically comprises:

the knowledge base comprises a plurality of production rules and corresponding execution conditions of a production task scheduling scheme;

the production rule comprises a device operation mode, a task execution mode and a production flow relation.

The knowledge base is constructed specifically as follows:

obtaining a time sequence embedded vector through data conversion according to the historical data;

And according to the time sequence embedded vector, obtaining the production rule cluster classification of a plurality of scheduling schemes through cluster analysis according to Euclidean distance between data points in the cluster and the center of the cluster.

The equipment operation data, the production task data and the environment data are specifically as follows:

The equipment operation data at least comprises any one of the working state, temperature, load and fault record of the equipment,

The production task data at least comprises any one of execution requirements, priority and required resources of the production task,

The environmental data at least comprises any one of temperature, humidity, pressure and air flow.

According to the execution condition of the production task scheduling scheme, the time sequence knowledge base is adjusted, specifically:

according to the equipment operation data, the production task data and the environment data which are acquired in real time and the execution feedback data of the production task scheduling scheme, the execution feedback data at least comprise any one of task completion time, resource utilization rate, abnormal event records and manual intervention records;

obtaining a production task scheduling scheme according to the equipment operation data, the production task data and the environment data;

and obtaining the execution condition of the production task scheduling scheme according to the execution feedback data so as to update and optimize the production task scheduling scheme.

When a plurality of production tasks exist, the production task scheduling method specifically comprises the following steps:

obtaining a plurality of production task scheduling schemes according to the corresponding production tasks;

according to the production task scheduling schemes, combining priorities of the production tasks to obtain a production task scheduling strategy, wherein the production task scheduling strategy comprises a task execution sequence, equipment resource allocation and a time window;

The priority of the production task is determined according to the task emergency degree, the deadline constraint and the resource demand weight.

The intelligent production task scheduling method based on the time sequence knowledge base further comprises the following steps:

preprocessing the acquired data, wherein the preprocessing comprises denoising, missing value filling and normalization;

The denoising processing at least comprises a wavelet transformation method, the missing value filling at least comprises an interpolation method, and the normalization processing at least comprises a linear normalization method.

The present invention also discloses a storage medium,

The storage medium stores a computer program which when executed implements the steps of the intelligent production task scheduling method based on the time sequence knowledge base.

The invention discloses a processing device again, comprising:

A memory for storing a computer program;

And the processor is used for realizing the intelligent production task scheduling method based on the time sequence knowledge base when executing the computer program.

By adopting the technical scheme, the invention has the following beneficial effects:

1. In the invention, the dynamic scheduling strategy is generated by collecting the equipment operation data, the production task data and the environment data in real time, analyzing and mining the production rule by utilizing the time sequence and combining the historical scheduling cases and the execution feedback in the knowledge base. The real-time response mechanism remarkably reduces the production interruption risk caused by static rules in the traditional method, reduces the dependence on manual experience through the automatic decision of a production task scheduling scheme, and ensures the continuity of the production flow.

And adjusting the time sequence knowledge base according to the execution condition of the production task scheduling scheme. After each production period is finished, new data are added into the knowledge base, and an iteration strategy is learned through online. The dynamic update of the historical case library gradually adapts to long-term change, and the rule maintenance cost is reduced. The closed loop feedback mechanism continuously improves the intelligent level of the method, and a rule base does not need to be updated manually and regularly.

In addition, the method supports multidimensional data fusion (equipment operation data, production task data and environment data) and builds a knowledge base. When the equipment or task type is newly added, the corresponding data is only needed to be supplemented to the knowledge base, and the production line with different scales can be adapted. The expandability of the knowledge base is suitable for the differentiated requirements of complex multitasking scenes.

In summary, the invention significantly improves the dynamic adaptability of the production task scheduling through the closed loop flow of real-time data acquisition, time sequence analysis, knowledge base dynamic query and feedback iteration. And the multidimensional data fusion and the time sequence rule mining generate an accurate scheduling scheme, the knowledge base continuously learns new data, the method is suitable for long-term changes of production environments, manual intervention is reduced, and decision accuracy is improved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and do not constitute a limitation on the invention. In the drawings:

Fig. 1 is a flow chart of the intelligent production task scheduling method based on the time sequence knowledge base according to an embodiment of the invention.

Detailed Description

In order to more clearly illustrate the general inventive concept, a detailed description is given below by way of example with reference to the accompanying drawings.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those described herein, and therefore the scope of the present invention is not limited to the specific embodiments disclosed below.

As shown in fig. 1, an intelligent production task scheduling method based on a time sequence knowledge base includes:

And S100, analyzing by a time sequence method according to the equipment operation data, the production task data and the environment data acquired in real time to obtain a production rule.

The main purpose of the step is to identify potential rules and modes in the production process through real-time monitoring and analysis of the running state of production equipment, the execution condition of production tasks and environmental parameters. These rules help to predict future production needs, optimize resource allocation, and provide scientific basis for subsequent task scheduling. By the method, the system can be better adapted to the dynamically-changed production environment, and the production efficiency and the resource utilization rate are improved.

The time series method is a statistical analysis method and is used for researching patterns and trends in the time series data. Through time sequence analysis, the system can identify potential rules such as a periodic fault mode of equipment, a task execution time rule and the like, predict potential problems in production in advance, and optimize a scheduling strategy.

In this step, the collected data includes equipment operation data, production task data, and environmental data. The equipment operation data can monitor the equipment state in real time, thereby being beneficial to finding out equipment abnormality in time, reducing downtime and improving production efficiency. The production task data provides accurate task data, supports more reasonable resource allocation and task arrangement, and avoids resource waste and bottleneck problems. The environmental data provides accurate environmental monitoring data, is helpful to maintain optimal production conditions, and ensures product quality and consistency.

And mining a production rule from the historical data by applying a time sequence analysis method (such as Dynamic Time Warping (DTW), K mean value clustering and the like) according to the acquired data. For example, the time series data is converted into a high-dimensional vector to be embedded through ACGE-Text-Embedding, and then clustering analysis is performed to identify the periodic maintenance requirement or task execution bottleneck of the equipment.

Based on a time series model (e.g., LSTM network), production trends over a period of time in the future are predicted. For example, historical data is used to predict likely failures of equipment within a few days of the future, and maintenance schedules are scheduled in advance. Accurate trend prediction helps to address potential problems in advance, reducing the risk of unexpected outages and production breaks.

In one embodiment, in an automotive manufacturing facility, operation data (e.g., operating status, temperature, load, etc.), production task data (e.g., current lot of welding tasks and their priorities), and environmental data (e.g., temperature and humidity within a plant) of welding robot equipment are collected in real time. The phenomenon that the temperature fluctuation of the equipment is large in a specific time period and the load is too high is identified, the data are further analyzed, and the periodic rule of the temperature rise of the equipment in the afternoon every week is found, which is probably caused by the production task concentration in the time period.

Based on the rule, the corresponding production task scheduling scheme is convenient to take, the manager is reminded to check the equipment cooling system in advance in the morning on Tuesday, task allocation is adjusted, and overload of equipment in peak time is avoided. In addition, bottleneck links under certain production task combinations can be found through the production rule, and optimization suggestions are provided, such as adjusting the priority of part of the tasks or reallocating resources, so that the efficiency and stability of the whole production line are improved.

In summary, the steps not only realize comprehensive management and intelligent analysis of production data, but also provide a solid foundation for subsequent task scheduling.

And S200, inquiring a knowledge base according to the production rule to obtain a production task scheduling scheme.

The main purpose of the step is to quickly generate an optimal production task scheduling scheme by utilizing the extracted production rule and combining the constructed time sequence knowledge base. By the mode, the system can make full use of historical data and real-time data to make more accurate and efficient scheduling decisions, so that the production efficiency and the resource utilization rate are improved, and the production bottleneck is reduced.

According to the potential rules in the production process obtained in the previous step. These laws may include periodic failure modes of the device, time laws of task execution, production bottlenecks, etc. By mining these rules, the system can predict and cope with potential problems in advance, optimize resource allocation and improve production efficiency.

It will be appreciated that the knowledge base is a collection of data that stores historical production data and scheduling policies for supporting intelligent decisions.

The knowledge base contains a plurality of production rules and the history records of the corresponding production task scheduling schemes. For example, information about different device operating modes, task execution modes, and production flow relationships is stored in a knowledge base.

The system uses the current production rule to compare with the history cases in the knowledge base to find the most similar history scheduling scheme. A common method includes a Dynamic Time Warping (DTW) algorithm that measures the similarity between two time series.

Wherein Dynamic Time Warping (DTW) refers to an algorithm for calculating an optimal alignment path between two time sequences, particularly for sequences of different lengths.

The scheduling scheme closest to the current situation can be quickly found from a large amount of historical data by the system through similarity matching, so that the time and complexity for redesigning the scheduling strategy are reduced.

Based on the similarity matching result, the system combines the specific requirements of the current production task to generate an optimal task scheduling scheme. This includes the order of execution of tasks, allocation of device resources, and the necessary time window. The generated scheduling scheme not only considers historical experience, but also combines the current actual situation, thereby ensuring the accuracy and adaptability of scheduling decision.

In one embodiment, a semiconductor manufacturing facility identifies the existence of equipment load peaks during a specific time period during the execution of a task for a batch of chip production. This phenomenon may be due to multiple high priority tasks being started simultaneously during this period.

The system queries the knowledge base for the most similar historical case to the current production law. Under the similar situation, by adjusting the time window of some non-critical tasks, the equipment load can be effectively dispersed, and the overload problem of peak time can be avoided. The system calculates the similarity between the current production rule and the historical case through a DTW algorithm, and confirms that the current production rule is the closest reference case.

Based on this historical case, the system generates a new production task scheduling scheme.

The production task scheduling scheme can comprise the steps of adjusting a time window of part of non-critical tasks to a time period with lower load, arranging equipment resources required by the critical tasks in advance to ensure the completion of the critical tasks on time, dynamically monitoring the task execution progress and preparing an adjustment scheduling strategy at any time to cope with emergency.

Through the intelligent scheduling mode, the risk of overload of equipment is successfully avoided in a factory, and the overall efficiency and stability of a production line are improved. In addition, as the system can quickly find the optimal scheduling scheme, the requirement of manual intervention is reduced, and the continuity and automation level of production are further improved.

S300, adjusting the time sequence knowledge base according to the execution condition of the production task scheduling scheme.

The main purpose of this step is to dynamically update and optimize the time series knowledge base by analyzing the actual execution of the production task scheduling scheme. Through continuous learning and adjustment, the system can better adapt to the changed production environment, continuously improve the scheduling strategy and improve the production efficiency and the resource utilization rate. This process ensures long term performance improvement of the system and reduces the need for human intervention.

In this step, the execution condition of the task (such as task completion time, resource utilization rate, abnormal event record, manual intervention record, etc.) is monitored in real time during the production process. These data are the key basis for evaluating the effectiveness of the scheduling scheme. Real-time monitoring is helpful for finding and solving the problems in production in time, and reducing downtime and resource waste.

The system compares the actual execution situation with the expected target, and evaluates the effect of the scheduling scheme. For example, checking whether a task is completed on time, whether resource utilization is efficient, whether there is an abnormal event, etc. If the task is successfully completed according to the plan and the resource utilization rate is high, the scheduling scheme is considered to be effective, and if delay or resource waste occurs, further analysis of reasons is needed. By continuously evaluating the effect of the scheduling scheme, the system can identify which policies are valid and which need improvement, thereby gradually optimizing the scheduling policies.

The execution feedback data is incorporated into a time series knowledge base. This includes information such as task completion time, resource utilization, exception event records, etc. By continuously updating the knowledge base, the system can better adapt to long-term changes, such as equipment aging, process improvement and the like, and the effectiveness of the scheduling strategy is maintained.

In one embodiment, in an automotive manufacturing facility, the system generates a production task scheduling scheme that schedules welding robots to complete a welding task for a lot of product in a particular time period. In the execution process, the system monitors the working state, task progress and resource use condition of each robot in real time, and records all relevant execution feedback data (such as task completion time, resource utilization rate, abnormal event record and the like).

These execution feedback data are first analyzed and evaluated. Suppose that a robot is found to have overheat phenomenon when executing a task, which results in prolonged task completion time and affects overall production progress. The system records this anomaly and incorporates it into a time series knowledge base.

Based on the anomaly data, a load spike phenomenon of the robot within a specific time period is identified. To prevent similar problems from reoccurring, the system adjusts the scheduling strategy to decide to schedule cooling time in advance in future similar tasks or to assign some tasks to other idle robots.

In summary, the method not only realizes comprehensive evaluation and feedback of the execution condition of the production task scheduling scheme, but also ensures long-term performance improvement of the system by dynamically updating the time sequence knowledge base to continuously optimize the scheduling strategy. The self-adaptive learning mechanism enables the system to better cope with complex and changeable production environments, reduces manual intervention, and improves production efficiency and resource utilization rate.

As a preferred embodiment of the present invention, the device operation data, the production task data, the environment data specifically include:

The primary purpose of this embodiment is to determine which types of equipment operation data, production task data, and environmental data are specifically collected. By specifying the types of data which have a large influence on the production scheduling, various factors in the production process can be captured comprehensively and accurately, so that a solid data base is provided for subsequent time sequence analysis and intelligent scheduling.

The equipment operation data at least comprises any one of the working state, temperature, load and fault record of the equipment. The working state refers to whether the equipment is in a state of starting up, stopping, waiting and the like at present. The temperature refers to the temperature change condition of the equipment when the equipment is operated, and is used for monitoring whether the equipment has overheat risk. The load refers to the current workload of the device, helping to assess the efficiency of use and potential bottlenecks of the device. The fault record refers to the fault of the equipment and the repair record thereof, and is used for predicting future maintenance requirements.

The working state of the real-time monitoring equipment is favorable for finding out abnormal conditions in time, and the downtime is reduced. The temperature and load data may help identify periodic maintenance needs of the equipment, scheduling maintenance plans in advance. Fault logging helps to accumulate historical data and optimize future maintenance strategies.

The production task data at least comprises any one of execution requirements, priority and required resources of the production task. Wherein, the execution requirement refers to the specific operation steps and standards of each production task, and ensures that the task is completed according to the specification. The priority refers to determining the priority order of the tasks according to the urgency degree and importance of the tasks, and ensuring that the critical tasks are completed preferentially. The required resources refer to raw materials, equipment and other resources required for completing tasks, and are used for reasonably distributing the resources.

Explicit execution requirements help to ensure consistency and stability of product quality. The reasonable task priority ordering can avoid the delay of the key tasks and improve the overall production efficiency. Accurate resource demand information supports more reasonable resource allocation, and reduces resource waste and bottleneck problems.

The environmental data at least comprises any one of temperature, humidity, pressure and air flow. Temperature and humidity refer to the temperature and humidity levels in the production environment, and are particularly important in environmentally sensitive production processes. Pressure refers to the air pressure level within the production plant, affecting air flow and contaminant control. The air flow refers to the air flow condition in the workshop, ensuring good ventilation and pollution control.

Accurate temperature and humidity control is helpful for maintaining optimal production conditions and ensuring product quality and consistency. Proper pressure and air flow management can effectively control air quality and pollutant level in a workshop, and reduce accidents in production. The comprehensive environment monitoring data supports dynamic adjustment of the production environment, and improves the production efficiency and the safety.

According to the method and the device, various factors in the production process can be comprehensively known through detailed collection and processing of the equipment operation data, the production task data and the environment data, and a solid foundation is provided for subsequent time sequence analysis and intelligent scheduling. The multidimensional data fusion not only improves the intelligent level, but also remarkably improves the production efficiency and the resource utilization rate, and ensures the continuity and the stability of production.

As a preferred embodiment of the present invention, the method for scheduling an intelligent production task based on a time-series knowledge base further includes:

The main purpose of this embodiment is to ensure the accuracy of the subsequent analysis and scheduling decisions by preprocessing the acquired data. The preprocessing comprises denoising, missing value filling and normalization processing, and the operations can improve the quality of data, reduce noise interference, fill the missing value, and unify the data to the same scale, so that the prediction capability of a model and the overall performance of a system are improved.

Wavelet transformation is a signal processing technique for decomposing a signal and extracting useful information while removing noise. And denoising the data by adopting a wavelet transformation method. The wavelet transform method effectively removes noise interference by decomposing and reconstructing signals.

Noise is removed in the method, so that the purity of the data can be improved, and misjudgment in subsequent analysis is reduced. The wavelet transform can effectively separate high-frequency noise components in the signal, and retain useful information.

Interpolation is a data filling method that estimates missing values from known data points. The interpolation method fills in missing data according to the front and back trends of the time sequence.

The missing value filling in the step can avoid analysis deviation caused by incomplete data. The interpolation method can reasonably estimate the missing value and maintain the time continuity of the data.

The linear normalization scales the data to a uniform range (typically between 0 and 1) for subsequent analysis and processing. And carrying out normalization processing on the data by adopting a linear normalization method.

In the step, the normalization processing can eliminate the influence of different dimensions, so that the data are compared on the same scale. The linear normalization can simplify the subsequent calculation and analysis processes and improve the training efficiency of the model.

As a preferred implementation mode under the invention, according to the production rule, inquiring a knowledge base to obtain a production task scheduling scheme, specifically:

The main purpose of this embodiment is to utilize the data collected in real time to match with the historical scheduling scheme in the time sequence knowledge base, and identify the primary selected scheduling scheme which most accords with the current production rule through similarity analysis. And then combining the historical execution condition data of the primary options to finally generate an optimal production task scheduling scheme. This step aims at improving the accuracy and adaptability of the scheduling decision and ensuring the efficient operation of the production process.

And (3) processing and primarily screening the real-time data, and analyzing and obtaining a production rule by using a time sequence method according to the equipment operation data, the production task data and the environment data acquired in real time. The system is ensured to be capable of capturing new trend or change in the production process in time, and the latest basis is provided for subsequent scheduling.

The real-time data is compared for similarity to a plurality of scheduling schemes stored in a knowledge base. A Dynamic Time Warping (DTW) algorithm is typically used to measure the similarity between two time series and find therefrom several primary scheduling schemes closest to the current production law.

Where X and U are two time series, phi (i) represents the index of the ith element in X in Y, such that the total distance between the two sequences is minimized.

Through similarity matching, the scheduling scheme which is most suitable for the current situation can be rapidly positioned from a large amount of historical data, and the time cost of redesigning the scheduling strategy is reduced.

Based on the initially selected scheduling schemes, the corresponding execution condition data (such as task completion rate, resource utilization rate, abnormal event record and the like) in the knowledge base are further combined to comprehensively evaluate each scheme, and the optimal one is selected as the final production task scheduling scheme.

A number of factors are considered in the evaluation, including but not limited to task priority, resources required, expected completion time, and potential risk. By comprehensively evaluating the historical performance of each primary option, the actual effect of each primary option under the current condition can be predicted more accurately, so that more scientific and reasonable scheduling decisions can be made.

Through the steps, not only is the production rule effectively utilized, but also the historical experience in the knowledge base is combined, and the accurate and high-adaptability production task scheduling scheme is generated.

As a preferred example in this embodiment, the knowledge base is specifically:

The main purpose of this embodiment is to construct and maintain a time-series knowledge base containing various production laws and their corresponding execution conditions of the production task scheduling schemes. The knowledge base records key information such as equipment operation modes, task execution modes, production flow relationships and the like, and continuously optimizes the scheduling strategy through historical data and real-time feedback so as to improve production efficiency and resource utilization rate.

The embodiment builds a comprehensive knowledge base, which includes different types of production rules (such as equipment operation mode, task execution mode and production flow relation) and historical execution conditions of production task scheduling schemes corresponding to the rules. A comprehensive data base is provided so that the system can make accurate matching and scheduling decisions based on real-time data in the current production environment.

And storing the production rules in a classified manner according to the equipment operation mode, the task execution mode and the production flow relation. For example, the operation mode of the equipment can comprise normal operation, high-temperature operation, low-load state and the like, the task execution mode can be continuous operation, intermittent operation and the like, and the production flow relation relates to the dependency relation among different procedures.

Wherein the device operating mode is used to describe the operating state or performance characteristics of the device over a particular period of time. The task execution mode is used to describe the operational flow or manner in which the production task is followed in the execution process. The production flow relation is used for referring to the dependency relation of different production tasks in time sequence, so that the next process can be started after the previous process is completed.

In summary, this step enables efficient management and utilization of production laws.

Specifically, the knowledge base is constructed by:

The main purpose of this embodiment is to construct a structured knowledge base from historical data, convert the original data into high-dimensional feature vectors by using a time-series embedding technique, and classify similar production rules into different clusters by cluster analysis. The process can systematically organize the modes in the historical data, and provides efficient and accurate classification basis for the matching of the follow-up scheduling scheme, so that the intelligent level of the production task scheduling is improved.

The method for converting the preprocessed data into the time sequence embedded vector comprises the following steps:

Position sensitive embedded vectors are generated by sine/cosine functions based on the position coding of the transducer.

Where pos is the time step position and d is the embedded vector dimension.

The delay embedding of the Havok-Method reconstructs the phase space of the time series by selecting the appropriate embedding dimensions and delay times.

Deep learning models (e.g., LSTM/GRU) extract time-dependent features through sequential modeling.

The SEANet model combines the hole convolution and SOS constraints to generate a compact embedded vector.

This step captures long-term dependencies (e.g., device periodic failure modes) in the time series. And the high-dimensional time sequence is compressed into a low-dimensional vector, so that the computational complexity is reduced.

The embedded vectors are grouped using a clustering algorithm (e.g., K-means, hierarchical clustering, or DBSCAN), and cluster boundaries are partitioned according to Euclidean distance between data points within a cluster and the center of the cluster.

Wherein, the K-means algorithm is that,

Where v _i is the ith embedded vector and c _k is the kth cluster center.

By iterative optimization, the total distance within the cluster is minimized,

Hierarchical clustering is a bottom-up or top-down merging/decomposing of clusters through a tree graph (Dendrogram), suitable for scenes of unknown clusters

It will be appreciated that, during clustering, the optimal cluster number is selected by a contour coefficient (Silhouette Score),

The closer S is to 1, the better the clustering effect.

The step classifies similar production rules (such as equipment high-temperature operation mode and task burstiness peak) into the same cluster, so that a scheduling scheme can be matched quickly. The search range in real-time calculation is reduced by the representativeness of the cluster center.

And associating each cluster with the execution effect (such as the task completion rate and the resource utilization rate) of the historical scheduling scheme to form a mapping relation of a production rule cluster-scheduling scheme. After the real-time data is embedded, the optimal scheduling scheme is rapidly positioned by calculating the minimum distance between the real-time data and the cluster center. And combining historical execution data to improve the reliability of the scheduling scheme.

In one embodiment, for equipment scheduling optimization of a semiconductor manufacturing plant, SEANet models are employed in conjunction with cavity convolution to extract features of equipment operating modes. The embedding dimension d=64, and the hole convolution step increases exponentially (e.g., by a power of 2). The loss function is

Wherein lambda ₁,λ₂ is a weight coefficient, balancing compression and reconstruction accuracy.

Compression Error is the number of compression errors,Where V is the embedded vector and X is the original data.

Reconstruction Error the reconstruction of the error is performed,Where X' is the reconstructed data and X is the original data.

The model output generates a 64-dimensional embedded vector for each device operating mode (e.g., "high temperature high load" and "low load standby").

For the embedded vector, the cluster number k=3 is determined by the contour coefficients by K-means clustering. Cluster 1 is obtained as a high temperature high load mode (temperature >80 ℃, load > 90%), cluster 2 as an intermittent task mode (task interval >2 hours), and cluster 3 as a continuous high priority task mode (task priorities are all "emergency").

Cluster 1 corresponds to scheduling scheme a (increasing cooling intervals, decreasing failure risk). Cluster 3 corresponds to scheduling scheme C (preferentially allocates resources, ensuring that the urgent task is completed).

Device a had a current temperature of 85 ℃, a load of 95%, and the embedded vector was the smallest distance from the center of cluster 1 (d=0.12). And the scheduling scheme A is automatically triggered, a cooling system is adjusted, and the overheat shutdown of equipment is avoided.

According to the embodiment, the complex production rules in the historical data are converted into the structured knowledge base through time sequence embedding and cluster analysis, so that the matching efficiency and accuracy of the scheduling scheme are remarkably improved.

As a preferred embodiment of the present invention, the time-series knowledge base is adjusted according to the execution condition of the production task scheduling scheme, specifically:

The main purpose of the embodiment is to dynamically update the production rule cluster classification and scheduling scheme template in the time sequence knowledge base through the production data collected in real time and the execution feedback data, thereby realizing continuous optimization of production task scheduling. The process can improve the adaptability of the scheduling scheme, reduce abnormal events and improve the resource utilization rate.

The method comprises the steps of collecting equipment operation data in real time, wherein the equipment operation data comprise equipment states (such as temperature, load and fault codes), operation modes (such as high speed/low speed), energy consumption and the like, production task data comprise task types, priorities, required resources (such as equipment and raw materials), task completion time and the like, and environment data comprise workshop temperature, humidity, power supply stability and the like.

Feedback data is executed, including task completion time, resource utilization (e.g., equipment idle rate), abnormal event records (e.g., equipment failure, material shortage), and manual intervention records (e.g., manual adjustment times).

According to the current equipment, task and environment data, the current equipment, task and environment data are converted into time series embedded vectors so as to carry out knowledge base retrieval. And determining the best matched production rule cluster by calculating Euclidean distance between the embedded vector and the center of each cluster in the knowledge base. And selecting a scheduling scheme template with the best history performance from the matching clusters. The cluster classification based on real-time data ensures a high adaptation of the scheduling scheme to the current production law to adjust the scheduling scheme according to the current execution feedback data.

Based on the execution feedback data, the difference between the planned time and the actual completion time is compared, e.g

ΔT=T_{Actual practice is that of}-T_Planning

Computing resource utilization

And the frequency of the abnormal events, and counting the occurrence times (such as the times of equipment faults) of the abnormal events in unit time. The manual intervention degree is used for recording the times and reasons of manual adjustment (such as the automatic switching of the system to the manual mode).

The step is to quickly locate the defects of the scheduling scheme through the association analysis of the abnormal event and the equipment/environment data. The quantization of the key index provides a quantifiable basis for subsequent optimization.

And according to the execution feedback data, realizing the dynamic updating of the knowledge base. The current production data and the execution feedback data are added into a knowledge base, and the time series embedded vector base is updated. Cluster centers and boundaries are recalculated by online learning or periodic offline training. And adjusting the weight of the scheduling scheme in the cluster (such as reducing the priority of the high-failure rate scheme) according to the execution feedback data, and optimizing the scheduling scheme. Based on the anomaly event record, an anomaly threshold value for the device's operational mode is updated.

By continuously updating, the knowledge base can adapt to the aging of equipment, environmental change and the like. Through long-term optimization, the problem of failure of a scheduling scheme caused by the overtime of historical data is avoided.

According to the method, closed-loop optimization of production task scheduling is achieved through real-time data acquisition, feedback analysis execution and dynamic knowledge base updating.

As a preferred embodiment of the present invention, when there are a plurality of production tasks, the production task scheduling method specifically includes:

The main purpose of this embodiment is to generate a globally optimal production task scheduling policy by integrating task priorities, resource constraints and feasibility of a scheduling scheme when there are multiple production tasks. The strategy requires defining task execution sequence, equipment resource allocation and time window to maximize resource utilization, reduce task delay and meet priority requirements of urgent tasks.

In this embodiment, the detailed information (task type, required resources, processing time, priority) of the plurality of production tasks, equipment resource status (available equipment, capacity, maintenance schedule), and environmental constraints (e.g., plant capacity, energy limitations) are based on the plurality of production tasks. A plurality of candidate scheduling schemes are generated, each scheme including a task order, a resource allocation, and a time window.

The priority is determined according to the task urgency, deadline constraints, and resource demand weights. A weight is given according to the urgency level of the task (e.g., urgent task weight=0.5). From the deadlines, the remaining time of the task deadline and the current time (remaining time=t _Cut-off-T_{Currently, the method is that}) is calculated, and the shorter the remaining time is, the higher the priority is.

According to the weight of the resource requirement

The higher the resource demand, the lower the task priority (avoiding resource contention).

Determining a composite score

Where α, β, γ are weight coefficients (e.g., α=0.4, β=0.3, γ=0.3).

And descending order of the tasks is carried out according to P _i, so as to form a priority queue.

The step avoids subjective judgment deviation through quantitative decision making, and ensures objectivity of priority allocation. The emergency task insertion or resource change can be flexibly dealt with through weight coefficient adjustment.

Candidate solutions meeting the priority requirements are screened from the generated scheduling solutions (e.g., high priority tasks must be completed within the first 3 time windows). If the high-priority task and the low-priority task compete for the same equipment, the low-priority task is forcedly interrupted. A minimum resource quota is reserved for high priority tasks.

The Gantt chart is used to visualize task scheduling, and the time window is adjusted to eliminate resource bottlenecks.

The higher the priority, the larger the adjustment amplitude.

Finally, the task execution sequence (such as A, B, C) is realized. Device resource allocation (e.g., task a to device X, task B to device Y). A time window is determined (e.g., task A:08:00-10:00, task B: 10:30-12:00).

In one embodiment, a certain automotive parts factory needs to handle the following three tasks simultaneously:

Task a, emergency order (100 brake pads produced, 24 hours off time, equipment X required). Task B, regular order (500 filters produced, after a cut-off time of 48 hours, equipment Y is required). Task C, maintenance task (maintenance of device Z, 1 hour, no deadline).

And (3) generating a scheme 1, namely preferentially completing the tasks A and C and delaying the task B. Task order A→C→B, device X/Y/Z allocation, time window 08:00-18:00.

Generating scheme 2-balancing resource requirements of task B, task a using idle periods of device X. Task order A→B→C, device X/Y/Z allocation, time window 08:00-19:00.

For task priority quantitative evaluation, task a, urgency (class a, 0.5), remaining time 24 hours (weight 0.3→0.3/24=0.0125), resource demand weight (device X occupancy 100% →0.3/100=0.003).

P_A=0.4·0.5+0.3·0.0125+0.3·0.003=0.203

Task B, degree of urgency (class B, 0.3), remaining time 48 hours (0.3/48=0.00625), resource demand weight (device Y occupancy 80% → 0.3/80= 0.00375).

P_B=0.4·0.3+0.3·0.00625+0.3·0.00375=0.123

Task C, degree of urgency (level D, 0.1), resource demand weight (device Z occupancy 10% → 0.3/10=0.03).

P_C=0.4·0.1+0.3·0+0.3·0.03=0.049

The prioritization is determined as a > B > C.

Thus, the task order, A→B→C, is determined. Device allocation, task A→device X, task B→device Y, task C→device Z. Time window, task A:08:00-12:00 (emergency complete), task B:12:30-17:00 (with device Y idle period), task C:17:30-18:30 (low priority task).

Tasks a and B are parallel on different devices, avoiding device X/Y contention. Task a was completed within 24 hours and task B was completed within 48 hours, meeting the deadline.

According to the method, efficient scheduling in a complex production environment is achieved through multi-task scheduling scheme generation, priority quantitative evaluation and dynamic resource optimization.

The present invention also provides a storage medium,

Therefore, any effect of the intelligent production task scheduling method based on the time sequence knowledge base can be achieved, and details are omitted herein.

The present invention again provides a processing apparatus comprising:

A memory for storing a computer program;

The invention can be realized by adopting or referring to the prior art at the places which are not described in the invention.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.

The foregoing is merely exemplary of the present invention and is not intended to limit the present invention. Various modifications and variations of the present invention will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are to be included in the scope of the claims of the present invention.

Claims

1. An intelligent production task scheduling method based on a time sequence knowledge base is characterized by comprising the following steps:

2. The intelligent production task scheduling method based on the time sequence knowledge base according to claim 1, wherein the knowledge base is queried according to the production rule to obtain a production task scheduling scheme, which specifically comprises the following steps:

3. The intelligent production task scheduling method based on a time series knowledge base according to claim 2, wherein the knowledge base specifically comprises:

4. The intelligent production task scheduling method based on the time sequence knowledge base according to claim 3, wherein the construction of the knowledge base is specifically as follows:

5. The intelligent production task scheduling method based on the time sequence knowledge base according to claim 1, wherein the equipment operation data, the production task data and the environment data are specifically:

6. The intelligent production task scheduling method based on the time sequence knowledge base according to claim 1, wherein the time sequence knowledge base is adjusted according to the execution condition of the production task scheduling scheme, specifically:

7. The intelligent production task scheduling method based on a time series knowledge base according to claim 1, wherein when a plurality of production tasks exist, the production task scheduling method specifically comprises:

8. The intelligent production task scheduling method based on a time series knowledge base according to claim 1, further comprising:

9. A storage medium, characterized in that,

The storage medium has stored thereon a computer program which, when executed, implements the steps of the intelligent production task scheduling method based on a time-series knowledge base as claimed in any one of claims 1 to 8.

10. A processing apparatus, comprising:

A memory for storing a computer program;

A processor for implementing the steps of the intelligent production task scheduling method based on a time-series knowledge base according to any one of claims 1 to 8 when executing said computer program.