Intelligent scheduling and fault prediction method, system and device for optical storage system and medium

Abstract

The invention provides an optical storage system intelligent scheduling and fault prediction method and device, equipment and a medium, and belongs to the technical field of intelligent optical storage, transportation and maintenance, and the method comprises the steps: collecting the data of an optical storage system in real time, and carrying out the preprocessing; inputting the preprocessed data into a multi-modal fusion model, and performing unified feature representation through a cross attention mechanism to obtain fusion features; inputting the fusion features into a prediction model for fault prediction and residual service life prediction; generating a state space formed by a fault type, a fault probability and a real-time electricity price according to a prediction result through a reinforcement learning algorithm, and setting a scheduling strategy which takes the charging and discharging power of the energy storage battery pack as an action space, and setting a reward function; and constructing an experience playback pool according to the actual operation data of the optical storage system, and regularly updating the strategy network of the reinforcement learning algorithm. According to the invention, the data is collected through the multi-mode sensor, the deep learning and reinforcement learning are combined, the intelligent fault prediction and dynamic scheduling of the optical storage system are realized, and the operation efficiency and reliability are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent optical storage and transportation operation and maintenance, and particularly relates to an intelligent scheduling and fault prediction method, system, device and medium for an optical storage system. BACKGROUND

[0002] With the development of new energy, the optical storage system, as an efficient renewable energy solution, can effectively solve the intermittency and instability of photovoltaic power generation, and plays an important role in improving energy utilization efficiency and power supply reliability. However, the optical storage system faces many problems in the operation process.

[0003] Firstly, the traditional scheduling method is mainly based on fixed rules or simplified models, such as using a simple strategy of charging at low price and discharging at high price, or relying on particle swarm optimization algorithm, but these methods are difficult to adapt to the randomness of new energy generation and the rapid fluctuation of load, and due to the single optimization target, they often fall into local optimum, and lack consideration of the dynamic change of equipment health status, resulting in disconnection between the scheduling strategy and the actual operation state. Secondly, the current optical storage system relies on artificial inspection, regular maintenance or threshold monitoring of a single data source, which is low in efficiency and difficult to accurately identify early faults or complex fault models. For example, the traditional fault monitoring is usually based on electrical parameters (current, voltage) or simple threshold judgment, which is difficult to predict potential faults in advance and evaluate their impact on scheduling strategy. In addition, the existing fault prediction methods often lack comprehensive utilization of multi-source information, limiting the accuracy and reliability of fault prediction. Thirdly, the current optical storage system usually takes scheduling and fault diagnosis as independent modules, and the scheduling strategy does not fully consider the potential fault risk of equipment, and the fault diagnosis information cannot be fed back to the scheduling system in real time. SUMMARY

[0004] In a first aspect, the embodiments of the present application provide an intelligent scheduling and fault prediction method for an optical storage system, the optical storage system comprising a photovoltaic array, an energy storage battery pack, an energy storage converter and a grid-connected interface, the method comprising the following steps: S1. Real-time acquisition of electrical parameters, thermal imaging data, acoustic signals, environmental data and historical operation and maintenance records through a multi-modal sensor array deployed in the photovoltaic array, the energy storage battery pack and the energy storage converter, and pre-processing of the acquired data; S2. Inputting the pre-processed data into a multi-modal fusion model based on Transformer, and obtaining fusion features through cross-attention mechanism for unified feature representation; S3. Inputting the fusion features into a prediction model to respectively perform fault prediction including fault type and fault probability for the photovoltaic array, the energy storage battery pack and the energy storage converter, and respectively perform remaining useful life prediction; S4. The prediction result is generated by a reinforcement learning algorithm to generate a scheduling strategy with a state space composed of fault type, fault probability, real-time electricity price, and an action space of energy storage battery pack charging and discharging power, and a reward function is set according to the device fault penalty of the light storage system, the SOC health of the energy storage battery, and the light storage system. The light storage system is punished for abandoning light; S5. An experience replay pool is constructed according to the actual operation data of the light storage system, and the policy network of the reinforcement learning algorithm is regularly updated based on the experience replay pool; the actual operation data includes action data of the action space at the previous time step, state data of the state space at the current step, and immediate reward value of the reward function at the previous time step.

[0005] Further, the specific steps of step S1 are as follows: S11. The string voltage and string current of the photovoltaic array are collected by the smart meter, the thermal imaging data of the photovoltaic array are collected by the infrared camera, and the acoustic signals of the photovoltaic array are collected by the microphone array; It should be noted that the hot spot fault of the photovoltaic array can be located by the thermal imaging data, and the component loosening or crack fault of the photovoltaic array can be detected by the acoustic signals; S12. The cell voltage and temperature of the energy storage battery pack are collected by the battery management controller BMS; S13. The heat sink temperature of the energy storage converter is collected by the temperature sensor; S14. The environmental data including the light intensity, temperature, humidity and wind speed are obtained by the weather station system; S15. The structured historical operation and maintenance records are obtained by the operation and maintenance system interface; S16. The string voltage and string current of the photovoltaic array, the cell voltage of the energy storage battery pack, and the environmental data are respectively processed by sliding serial normalization, and the mean and standard deviation features in the window are calculated; S17. The thermal imaging data of the photovoltaic array, the cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter are enhanced by using a convolution enhancement algorithm; S18. The acoustic signals of the photovoltaic array are converted into frequency domain acoustic features by short-time Fourier transform; S19. The historical operation and maintenance records are converted into text embedding features by using a BERT model.

[0006] Further, the specific steps of step S2 are as follows: S21. The mean and standard deviation features of the string voltage and string current of the photovoltaic array, the cell voltage of the energy storage battery pack, and the environmental data are encoded by using an LSTM model to calculate the hidden state. S22. Use the CNN algorithm to perform image feature encoding on the enhanced features of the thermal imaging data of the photovoltaic array, the battery cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter; S23. Use a cross-modal attention fusion algorithm to fuse the temporal feature encoding with the image feature encoding to obtain temporal image features; S24. Modally align the acoustic features of the photovoltaic array with the text embedding features of the historical operation and maintenance records, and then combine them with the time series image features to form a global fusion feature.

[0007] Furthermore, the specific steps of step S3 are as follows: S31. Pre-build fault prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using a deep neural network (DNN). Train the models using historical operating data, fault status, and actual fault types. S32. Pre-build life prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using the LSTM model framework, and train them using historical runtime data and actual lifespan. S33. Fusion features Input the fault prediction model to output the fault type and corresponding fault probability;

[0008] in, is the Sigmoid function, is the fault type, is the failure probability, is the weight coefficient of the deep convolutional network, is the bias term; S34. Input the time series data of the fusion features into the life prediction model to predict the remaining life of the photovoltaic array, energy storage battery pack, and energy storage converter;

[0009] in, is the time series data of fusion features, is the remaining life prediction value vector of the photovoltaic array, energy storage battery group, and energy storage converter.

[0010] Furthermore, the specific steps of step S4 are as follows: S41. Constructing the state space of reinforcement learning:

[0011] in, is the real-time electricity price, is the failure probability of the PV array, is the remaining life of the PV array, a failure probability of the energy storage battery pack, a remaining life of the energy storage battery pack, a failure probability of the energy storage converter, a remaining life of the energy storage battery; S42. defining the action space as the charge-discharge power of the energy storage battery pack;

[0012] wherein, a maximum charging power of the energy storage battery pack, a maximum discharging power of the energy storage battery pack, a charge-discharge power control value of the energy storage battery pack; S43. constructing a multi-objective reward function:

[0013] wherein, an interaction power of the energy storage system with the power grid, a real-time electricity price, a photovoltaic failure penalty coefficient, a battery failure penalty coefficient, a converter failure penalty coefficient, a storage battery SOC health weight coefficient, an optimal SOC of the energy storage battery, a real-time SOC of the energy storage battery, an auxiliary service reward coefficient, a power grid auxiliary service item, a curtailed photovoltaic penalty coefficient, a photovoltaic power generation power at time t, an actual photovoltaic power consumed by the energy storage or the power grid.

[0014] Further, the step S5 specifically comprises the following steps: S51. in the actual operation process of the energy storage system, storing the state , action , immediate reward and next state of each time step t as an experience data into an experience replay pool D, and preferentially placing experience samples with high reward value or large state change;

[0015]

[0016] wherein P represents an experienced placement priority; S52. A deep deterministic policy gradient algorithm DDPG is used for policy optimization, and a loss function is constructed for the value evaluation network Critic:

[0017] wherein, is the current parameter of the value evaluation network Critic, is the target parameter of the value evaluation network Critic; is the current parameter of the policy network Actor, is the target parameter of the policy network Actor, is a discount factor, and N is the number of sampled experience samples in the experience replay pool; is calculated according to the multi-objective function ; S53. The policy network Actor is updated based on the gradient direction of the loss function of the value evaluation network Critic:

[0018] wherein, represents the gradient of the policy network parameter θ is the objective function of the policy network Actor, E is the mathematical expectation, i.e. the batch average operation, is the gradient of the action , represents the state-action value function output by the value evaluation network Critic, i.e. the long-term return prediction value of taking the action in the state , represents the action output by the policy network Actor network in the state ; S54. Actual operation data of the optical storage system is obtained, the actual operation data includes a state vector and an actual fault label , and the state vector is input into the trained fault prediction model to obtain a predicted fault label, and a cross-entropy loss function is constructed to measure the difference between the predicted fault label and the actual fault label:

[0019] wherein, is the total number of fault types, is the encoding of the actual fault label, is the probability of the predicted cth fault; S55. Update the model parameters of the failure prediction model based on the cross-entropy loss function and through a back propagation algorithm.

[0020] Further, step S55 specifically includes the following steps: S551. Derive the gradient of the model parameters based on the cross-entropy loss function and the failure prediction model parameters. S552. Update the parameters of the failure prediction model based on the gradient of the model parameters and according to a set learning rate.

[0021] In a second aspect, the embodiments of the present application also provide a method and device for intelligent scheduling and failure prediction of a light storage system. The light storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid-connected interface. The device includes: A data acquisition and processing module is configured to acquire electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time through a multi-modal sensor array deployed in the photovoltaic array, the energy storage battery pack, and the energy storage converter, and to preprocess the acquired data. A data fusion module is configured to input the preprocessed data into a multi-modal fusion model based on a Transformer, to perform unified feature representation through a cross-attention mechanism, and to obtain fusion features. A failure and life prediction module is configured to input the fusion features into a prediction model, to perform failure prediction including failure types and failure probabilities for the photovoltaic array, the energy storage battery pack, and the energy storage converter, respectively, and to perform residual service life prediction for the photovoltaic array, the energy storage battery pack, and the energy storage converter, respectively. An intelligent scheduling module is configured to generate a scheduling strategy with a state space composed of failure types, failure probabilities, and real-time electricity prices, and an action space composed of charging and discharging power of the energy storage battery pack, by an reinforcement learning algorithm, and to set a reward function according to device failure penalties of the light storage system, SOC health of the energy storage battery, and photovoltaic power generation light abandonment penalties. A learning optimization module is configured to construct an experience replay pool according to actual operation data of the light storage system, and to regularly update a policy network of the reinforcement learning algorithm based on the experience replay pool. The actual operation data includes action data of the action space at a previous time step, state data of the state space at a current step, and immediate reward values of the reward function at the previous time step.

[0022] In a third aspect, the embodiments of the present application also provide an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for intelligent scheduling and failure prediction of a light storage system according to the first aspect are implemented.

[0023] In a fourth aspect, the embodiments of the present application further provide a storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the intelligent scheduling and fault prediction method of the optical storage system according to the first aspect.

[0024] From the above technical solutions, the present application has the following advantages: The intelligent scheduling and fault prediction method, system, device and medium of the optical storage system provided by the present application realize intelligent operation and health management of the optical storage system by integrating multi-modal data acquisition, deep learning fusion, reinforcement learning scheduling and online adaptive updating, which can improve the operation efficiency and reliability of the optical storage system, reduce operation and maintenance costs, prolong the service life of the equipment, and improve the overall performance and adaptability of the optical storage system. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0026] Figure 1 The flowchart of the intelligent scheduling and fault prediction method of the optical storage system of the present application.

[0027] Figure 2 The schematic diagram of the intelligent scheduling and fault prediction device of the optical storage system of the present application. DETAILED DESCRIPTION

[0028] In the following detailed description of the specific steps of the intelligent scheduling and fault prediction method of the optical storage system, various embodiments of the present disclosure will be described more fully. The present disclosure can have various embodiments, and adjustments and changes can be made therein. However, it should be understood that there is no intention to limit various embodiments of the present disclosure to the specific embodiments disclosed herein, but the present disclosure should be understood to cover all adjustments, equivalents and / or alternatives falling within the spirit and scope of various embodiments of the present disclosure.

[0029] Illustratively, with the rapid development of the new energy industry, the optical storage system as a kind of efficient renewable energy solution has important significance for improving energy utilization efficiency and enhancing power supply reliability. However, in actual operation, the optical storage system faces a series of problems to be solved.

[0030] On the one hand, traditional photovoltaic energy storage system scheduling strategies usually rely on fixed rules or simplified models. For example, the common low-price charging and high-price discharging strategy, as well as particle swarm optimization algorithm, although can achieve basic scheduling function to some extent, but these methods are difficult to adapt to the randomness and volatility of new energy power generation and the rapid changes of load demand. In addition, due to the relatively single optimization goal of these traditional methods, it is easy to fall into local optimal solution, and cannot fully consider the dynamic changes of equipment health status, resulting in the disconnection between scheduling strategy and actual running state.

[0031] On the other hand, the current fault diagnosis of photovoltaic energy storage system mainly relies on artificial inspection, regular maintenance or threshold monitoring based on single data source. These methods not only have low efficiency, but also are difficult to accurately identify early faults or complex fault patterns. For example, traditional fault monitoring usually only based on electrical parameters (such as current, voltage) or simple threshold judgment, it is difficult to predict potential faults in advance and evaluate their impact on scheduling strategy. In addition, existing fault prediction methods often fail to fully utilize multi-source information, which greatly limits the accuracy and reliability of fault prediction.

[0032] Finally, the current photovoltaic energy storage system usually regards scheduling and fault diagnosis as two independent modules, lacking effective coordination mechanism. The scheduling strategy fails to fully consider the potential fault risk of equipment, while the fault diagnosis information also fails to be fed back to the scheduling system in real time, thus affecting the overall operation efficiency and reliability of the system.

[0033] In view of the above problems, the embodiment provides a photovoltaic energy storage system intelligent scheduling and fault prediction method, which comprehensively collects photovoltaic energy storage system operation data through a multi-modal sensor array, combines deep learning and reinforcement learning algorithm, realizes intelligent collaborative optimization of fault prediction and dynamic scheduling, ensures the system to quickly respond to dynamic changes, and improves the operation efficiency and reliability.

[0034] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0035] Please refer to Figure 1 Fig. 1 is a flowchart of a photovoltaic energy storage system intelligent scheduling and fault prediction method in a specific embodiment, the photovoltaic energy storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter and a grid-connected interface, and the method includes the following steps: S1. Real-time collection of electrical parameters, thermal imaging data, acoustic signals, environmental data and historical operation and maintenance records through a multi-modal sensor array deployed in the photovoltaic array, energy storage battery pack and energy storage converter, and preprocessing of the collected data; It should be noted that by collecting data of multiple modalities, multi-source input is provided for the deep learning model, early faults are detected through thermal imaging and acoustic signals, the accuracy of fault diagnosis is improved, the data quality is improved, and a basis is provided for subsequent feature extraction and model training; S2. Input the preprocessed data into a multi-modal fusion model based on Transformer, and obtain the fusion features through cross-attention mechanism for unified feature representation; It should be noted that through Transformer and cross-attention mechanism, the features of different modalities are deeply fused, the diversity and accuracy of feature representation are improved, and a unified feature representation is provided for subsequent fault prediction and life prediction; S3. Input the fusion features into the prediction model to perform fault prediction including fault type and probability for the photovoltaic array, energy storage battery pack and energy storage converter, respectively, and perform residual service life prediction, respectively; It should be noted that through DNN and LSTM, the fault type and probability are predicted, the accuracy and reliability of fault prediction are improved, the residual service life of the equipment is predicted through the LSTM model, and a basis is provided for preventive maintenance, and through the combination of fault prediction and life prediction model, a comprehensive evaluation of the health status of the equipment is realized; S4. The prediction results are generated into a state space composed of fault type, fault probability and real-time electricity price through a reinforcement learning algorithm, a scheduling strategy is generated with the charging and discharging power of the energy storage battery pack as the action space, and a reward function is set according to the device fault penalty of the energy storage system, the SOC health of the energy storage battery and the penalty for photovoltaic power generation light abandonment; It should be noted that the dynamic scheduling strategy is generated through the reinforcement learning algorithm, which can optimize the scheduling according to the real-time electricity price and the health status of the equipment, and through the comprehensive consideration of the device fault penalty, the battery SOC health and the light abandonment penalty, the multi-objective optimization is realized, and the policy network is updated regularly based on the experience replay pool to ensure that the scheduling strategy can adapt to the dynamic changes of the system; S5. Construct an experience replay pool according to the actual operation data of the energy storage system, and update the policy network of the reinforcement learning algorithm regularly based on the experience replay pool; the actual operation data includes action data of the action space at the previous time step, state data of the state space at the current step and immediate reward value of the reward function at the previous time step; It should be noted that the DDPG algorithm is used to optimize the scheduling strategy, improve the stability and adaptability of the strategy, and improve the model's ability to recognize new failure modes by updating the parameters of the fault prediction model online. The model parameters are updated based on real-time operation data to ensure that the system can quickly adapt to dynamic changes.

[0036] This embodiment collects a variety of data through a multimodal sensor array, which can fully perceive the operating status of the photovoltaic storage system. Through deep learning and reinforcement learning algorithms, it realizes intelligent fault prediction and dynamic scheduling. By combining fault prediction with scheduling strategies, it achieves collaborative optimization of system operation. Based on the real-time data update strategy network, it ensures that the system can quickly respond to dynamic changes.

[0037] Furthermore, as a refinement and expansion of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another method for intelligent scheduling and fault prediction of a photovoltaic storage system is provided. The photovoltaic storage system includes a photovoltaic array, an energy storage battery group, an energy storage converter, and a grid-connected interface. The method includes the following steps: S1. A multimodal sensor array deployed on the photovoltaic array, energy storage battery pack, and energy storage converter collects electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time, and preprocesses the collected data. The specific steps of step S1 are as follows: S11. Collecting the string voltage and current of the photovoltaic array through the smart meter, collecting thermal imaging data of the photovoltaic array through the infrared camera, and collecting acoustic signals of the photovoltaic array through the microphone array; It should be noted that hot spot faults in photovoltaic arrays can be located through thermal imaging data, and loose or cracked components in photovoltaic arrays can be detected through acoustic signals; S12. Collecting the battery cell voltage and temperature of the energy storage battery pack through the battery management controller BMS; S13. Collect the heat sink temperature of the energy storage converter through the temperature sensor; S14. Obtain environmental data through the weather station system, including light intensity, temperature, humidity and wind speed; S15. Obtain structured historical operation and maintenance records through the operation and maintenance system interface; S16. Process the photovoltaic array string voltage, string current, and battery cell voltage of the energy battery group using sliding serial port normalization based on the environmental data, and calculate the mean and standard deviation characteristics within the window;

[0038] in, 、 is the mean and standard deviation within the window; S17. The thermal imaging data of the photovoltaic array, the battery cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter are subjected to temperature enhancement using a convolution enhancement algorithm;

[0039] wherein, is a convolution weight coefficient, is a bias term, represents a temperature value, represents a value of convolution enhancement; S18. The acoustic signal of the photovoltaic array is converted into acoustic features in the frequency domain through short-time Fourier transform;

[0040] S19. The historical operation and maintenance records are subjected to a BERT model to generate text embedding features;

[0041] It should be noted that collecting multiple modal data, including electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records, provides diverse inputs for subsequent deep learning models. Through thermal imaging and acoustic signal detection, early faults are detected, and the accuracy of fault diagnosis is improved. Through the preprocessing steps of normalization, convolution enhancement, and frequency domain conversion, the data quality is improved, providing a foundation for subsequent feature extraction and model training. Exemplarily, at a certain time, the string voltage of the photovoltaic array is 24V, the string current is 8A, the environmental temperature is 25°C, and the light intensity is 800W / m². After sliding window normalization processing, the mean value of the voltage is 24V, and the standard deviation is 0.5V; the mean value of the current is 8A, and the standard deviation is 0.2A; S2. The preprocessed data is input into a multi-modal fusion model based on Transformer, and unified feature representation is performed through cross-attention mechanism to obtain fusion features; the specific steps of step S2 are as follows: S21. The mean and standard deviation features of the string voltage, string current, battery cell voltage of the energy storage battery, and environmental data of the photovoltaic array are subjected to time series feature coding using an LSTM model to calculate the hidden state;

[0042] wherein, represents the mean and standard deviation features of the voltage, represents the mean and standard deviation features of the current, represents the mean and standard deviation features of the environmental data, represents the hidden state; S22. Use the CNN algorithm to perform image feature encoding on the enhanced features of the thermal imaging data of the photovoltaic array, the battery cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter;

[0043] S23. Use the cross-modal attention fusion algorithm to fuse the temporal feature encoding with the image feature encoding to obtain the temporal image feature ;

[0044] in, Q , K , V is the query, key, and value matrix, is the dimension scaling factor; S24. Modally align the acoustic features of the photovoltaic array with the text embedding features of the historical operation and maintenance records, and then combine them with the time series image features to form a global fusion feature ; It should be noted that the Transformer and cross-attention mechanism deeply fuses features from different modalities, improving the diversity and accuracy of feature representation. By combining time series features and image features, it can more comprehensively reflect the operating status of the system. By generating global fusion features, it provides a unified feature representation for subsequent fault prediction and life prediction. For example, assume that after encoding by the LSTM model, the hidden state of the electrical parameter is [0.2, 0.3, 0.4]; after encoding by the CNN algorithm, the image features of the thermal imaging data are [0.5, 0.6, 0.7]; through the cross-modal attention fusion algorithm, the fused features are [0.35, 0.45, 0.55]; then the acoustic features [0.1, 0.2] and the text embedding features [0.8, 0.9] are concatenated to obtain the global fused features [0.35, 0.45, 0.55, 0.1, 0.2, 0.8, 0.9]; S3. Input the fused features into the prediction model to perform fault prediction including fault type and fault probability for the photovoltaic array, energy storage battery pack, and energy storage converter, as well as remaining service life prediction for each. The specific steps of step S3 are as follows: S31. Pre-build fault prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using a deep neural network (DNN). Train the models using historical operating data, fault status, and actual fault types. S32. Pre-build life prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using the LSTM model framework, and train them using historical runtime data and actual lifespan. S33. Fusion features The input fault prediction model outputs a fault type and a corresponding fault probability;

[0045] wherein, is a Sigmoid function, is a fault type, is a fault probability, is a deep convolutional network weight coefficient, is a bias term; S34. Input the time series data of the fusion features into a life prediction model to predict the remaining life of the photovoltaic array, the energy storage battery pack, and the energy storage converter;

[0046] wherein, is time series data of the fusion features, is a remaining life prediction value vector of the photovoltaic array, the energy storage battery pack, and the energy storage converter; It should be noted that the deep neural network DNN and the long short-term memory network LSTM are used to predict the fault type and probability, improve the accuracy and reliability of fault prediction, predict the remaining service life of the equipment through the LSTM model, provide a basis for preventive maintenance, and comprehensively evaluate the health status of the equipment by combining the fault prediction and life prediction models; For example, the input fusion features are [0.35, 0.45, 0.55, 0.1, 0.2, 0.8, 0.9], the DNN model predicts that the fault type of the photovoltaic array is "hot spot fault", and the fault probability is 0.8; the LSTM model predicts that the remaining service life of the photovoltaic array is 1000 hours; S4. The prediction results are generated by a reinforcement learning algorithm to form a state space composed of a fault type, a fault probability, and a real-time electricity price, a scheduling strategy is generated with the charging and discharging power of the energy storage battery pack as an action space, and a reward function is set according to the device fault penalty of the energy storage system, the SOC health of the energy storage battery, and the light generation abandonment penalty of the photovoltaic power generation; the specific steps of step S4 are as follows: S41. Construct a state space of reinforcement learning:

[0047] wherein, is a real-time electricity price, is a fault probability of the photovoltaic array, is a remaining life of the photovoltaic array, is a fault probability of the energy storage battery pack, is a remaining life of the energy storage battery pack, For the failure probability of the energy storage converter, For the remaining life of the energy storage battery; S42. Define the action space as the charge and discharge power of the energy storage battery pack;

[0048] Wherein, The maximum charging power of the energy storage battery pack, The maximum discharging power of the energy storage battery pack, The charge and discharge power control value of the energy storage battery pack; S43. Construct a multi-objective reward function:

[0049] Wherein, The interaction power of the energy storage system and the grid, The real-time electricity price, The photovoltaic failure penalty coefficient, The battery failure penalty coefficient, The converter failure penalty coefficient, The energy storage battery SOC health weight coefficient, The optimal SOC of the energy storage battery, The real-time SOC of the energy storage battery, The auxiliary service reward coefficient, The grid auxiliary service item, The abandoned light penalty coefficient, The photovoltaic power generation power at time t, The actual photovoltaic power consumed by the energy storage or the grid; It should be noted that the grid auxiliary service item includes the frequency modulation and standby service of the light storage system;

[0050] The power sum of the interaction power of the energy storage system and the grid at time t and the charge and discharge power of the energy storage battery Take the smaller one of the power sum and the charge and discharge power of the energy storage battery; Calculate the photovoltaic power generation power at time t The actual photovoltaic power consumed by the energy storage or the grid The power difference of the actual photovoltaic power consumed by the energy storage or the grid Take the value of the power difference when the power difference is greater than 0, and take 0 when the power difference is less than 0; It should be noted that the dynamic scheduling strategy generated by the reinforcement learning algorithm can be optimized according to the real-time electricity price and the device health state, the multi-objective optimization can be realized by comprehensively considering the device fault penalty, the battery SOC health and the light abandonment penalty, the scheduling strategy can be ensured to adapt to the dynamic changes of the system by regularly updating the policy network based on the experience replay pool; S5. Constructing an experience replay pool according to the actual operation data of the optical storage system, and regularly updating the policy network of the reinforcement learning algorithm based on the experience replay pool; the actual operation data includes action data of the action space at the previous time step, state data of the state space at the current step and immediate reward value of the reward function at the previous time step; the specific steps of step S5 are as follows: S51. In the actual operation process of the optical storage system, the state , action , immediate reward and next state of each time step t are stored as an experience data in the experience replay pool D, and the experience samples with high reward value or large state change are preferentially placed;

[0051]

[0052] Wherein, P represents the experience placement priority; S52. The deep deterministic policy gradient algorithm DDPG is used for policy optimization, and a loss function is constructed for the value evaluation network Critic:

[0053] Wherein, is the current parameter of the value evaluation network Critic, is the target parameter of the value evaluation network Critic; is the current parameter of the policy network Actor, is the target parameter of the policy network Actor, is the discount factor, and N is the number of sampled experience samples in the experience replay pool; is calculated according to the multi-objective function ; Exemplarily, 0.95 is taken; S53. Update the policy network Actor based on the gradient direction of the loss function of the value evaluation network Critic:

[0054] Wherein, represents the gradient of the policy network parameter θ is the objective function of the policy network Actor, E is the mathematical expectation, i.e. the batch average operation, is the gradient of action a, represents the state-action value function output by the value evaluation network Critic, i.e. the long-term revenue prediction value of taking action in state ; represents the action output by the policy network Actor network in state ; It should be noted that, is the update direction of the parameter θ to maximize the expected revenue, is the score gradient of the action of the value evaluation network Critic, guiding the policy network Actor to adjust the action to obtain higher revenue; Replace the action a with the action actually output by the policy network in state , ensuring that the gradient calculation is based on the action generated by the current policy, rather than an arbitrary action; The influence gradient of the policy network parameter θ on the output action guides the parameter update direction; The overall meaning of the formula is that the policy network Actor updates the parameters in the direction of the Q value evaluated by the value evaluation network Critic, so that the policy gradually tends to the direction of maximizing the expected revenue; S54. Obtain actual operation data of the optical storage system, the actual operation data comprising a state vector and an actual fault label , and input the state vector into the trained fault prediction model to obtain a predicted fault label, and construct a cross-entropy loss function to measure the difference between the predicted fault label and the actual fault label:

[0055] Wherein, C is the total number of fault types, is the encoding of the actual fault label, is the probability of the predicted cth fault; S55. Update the model parameters of the fault prediction model based on the cross-entropy loss function and through a back propagation algorithm; the specific steps of step S55 are as follows: S551. Derive the cross-entropy loss function based on the fault prediction model parameters to obtain the gradient of the model parameters; S552. Based on the gradient of the model parameters, the parameters of the fault prediction model are updated according to the set learning rate; It should be noted that the fault prediction model can continuously adjust its parameters based on real-time operating data, thereby better adapting to the dynamic changes and new fault modes of the solar-storage system; The Deep Deterministic Policy Gradient (DDPG) algorithm is used to optimize scheduling strategies, improving their stability and adaptability. The fault prediction model's parameters are updated online to enhance the model's ability to identify new fault modes. Model parameters are updated based on real-time operational data, ensuring the system can quickly adapt to dynamic changes. For example, at a certain time step, the state is [0.5, 0.8, 1000, 0.2, 800], action Taking 50kW as an example, instant reward is -0.3, the next state The data are stored in the experience replay pool, and the policy network parameters are updated through the DDPG algorithm to optimize the scheduling strategy. At the same time, the parameters of the fault prediction model are updated through the cross entropy loss function to improve the model's ability to recognize new fault modes.

[0056] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0057] like Figure 2 As shown, the following is an embodiment of the intelligent scheduling and fault prediction device for the photovoltaic storage system provided by the embodiment of the present disclosure. This system and the intelligent scheduling and fault prediction method for the photovoltaic storage system in the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the intelligent scheduling and fault prediction device for the photovoltaic storage system, please refer to the embodiment of the above-mentioned intelligent scheduling and fault prediction method for the photovoltaic storage system.

[0058] The photovoltaic storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter and a grid-connected interface, and the device includes: The data acquisition and processing module is used to collect electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time through a multimodal sensor array deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters, and to pre-process the collected data. The data fusion module is used to input the preprocessed data into the Transformer-based multimodal fusion model, unify the feature representation through the cross-attention mechanism, and obtain the fused features; a fault and life prediction module, configured to input the fused features into a prediction model to perform fault prediction including fault types and fault probabilities for the photovoltaic array, the energy storage battery pack, and the energy storage converter, respectively, and to perform remaining useful life prediction, respectively; an intelligent scheduling module, configured to generate a scheduling strategy by an reinforcement learning algorithm using a state space composed of fault types, fault probabilities, and real-time electricity prices, and an action space composed of charging and discharging powers of the energy storage battery pack, and to set a reward function according to a device fault penalty of the photovoltaic energy storage system, a health of the energy storage battery SOC, and a photovoltaic power generation light abandonment penalty as a target; a learning optimization module, configured to construct an experience replay pool according to actual operation data of the photovoltaic energy storage system, and to update a policy network of the reinforcement learning algorithm based on the experience replay pool periodically; the actual operation data includes action data of the action space at a previous time step, state data of the state space at a current step, and an immediate reward value of the reward function at the previous time step.

[0059] The embodiment realizes intelligent fault prediction and dynamic scheduling of the photovoltaic energy storage system, and improves operation efficiency and reliability through interaction and cooperation of the data acquisition and processing module, the data fusion module, the fault and life prediction module, the intelligent scheduling module, and the learning optimization module.

[0060] The photovoltaic energy storage system intelligent scheduling and fault prediction method provided in the embodiment of the application can be applied to an electronic device. Those skilled in the art can understand that the electronic device structure involved in the embodiment of the application does not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than the diagram, or combine certain components, or different component arrangements. In the embodiment of the application, the electronic device includes but is not limited to a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the application described and / or claimed herein.

[0061] The electronic device can include a processor, an external memory interface, an internal memory, a universal serial bus (USB) interface, a charging management module, a power management module, a battery, a wireless communication module, an audio module, a speaker, a microphone, a sensor module, a key, a camera, a display screen, and a SIM card interface, etc.

[0062] It can be understood that the structure shown in the embodiments of the present application does not constitute a specific limitation on the electronic device. In other embodiments of the present application, the electronic device can include more or fewer components than shown, or combine certain components, or split certain components, or different arrangement of components. The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0063] The processor can include one or more processing units, such as: the processor can include a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Among them, different processing units can be independent devices, or can be integrated in one or more processors.

[0064] Among them, the processor can be the nerve center and command center of the electronic device. The controller can generate operation control signals according to instruction operation codes and timing signals, and complete the control of fetching and executing instructions.

[0065] The memory can also be provided in the processor for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. The memory can save instructions or data that the processor has just used or repeatedly uses. If the processor needs to use the instructions or data again, it can directly call from the memory. Avoiding repeated access, reducing the waiting time of the processor, thus improving the efficiency of the system.

[0066] The electronic device realizes the intelligent scheduling and fault prediction method of the optical storage system. Through the multi-modal sensor array deployed in the photovoltaic array, energy storage battery pack and energy storage converter, the electrical parameters, thermal imaging data, acoustic signals, environmental data and historical operation records are collected in real time, and the collected data is preprocessed. The preprocessed data is input into the multi-modal fusion model based on the Transformer, and the unified feature representation is obtained through the cross attention mechanism to obtain the fusion feature. The fusion feature is input into the prediction model to perform fault prediction including fault type and fault probability for the photovoltaic array, energy storage battery pack and energy storage converter, and to perform residual service life prediction respectively. The prediction result is generated by the reinforcement learning algorithm to form a state space composed of fault type, fault probability and real-time electricity price, and a scheduling strategy with energy storage battery pack charging and discharging power as the action space. The reward function is set according to the device fault penalty of the optical storage system, the SOC health of the energy storage battery and the penalty for photovoltaic power generation light abandonment. The experience replay pool is constructed according to the actual operation data of the optical storage system, and the strategy network of the reinforcement learning algorithm is regularly updated based on the experience replay pool. The actual operation data includes action data of the action space at the previous time step, state data of the state space at the current step and immediate reward value of the reward function at the previous time step. The technical scheme achieves the beneficial effects of comprehensively collecting the operation data of the optical storage system through the multi-modal sensor array, combining the deep learning and reinforcement learning algorithm, realizing the intelligent collaborative optimization of fault prediction and dynamic scheduling, ensuring the system to quickly respond to dynamic changes, and improving the operation efficiency and reliability.

[0067] In the storage medium provided in the application, a program product capable of realizing the intelligent scheduling and fault prediction method of the optical storage system is stored.

[0068] The intelligent scheduling and fault prediction method of the optical storage system comprises: through a multi-modal sensor array deployed on a photovoltaic array, an energy storage battery pack, and an energy storage converter, real-time acquisition of electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records is performed, and the acquired data is preprocessed; the preprocessed data is input into a multi-modal fusion model based on a Transformer, unified feature representation is performed through a cross-attention mechanism, and fusion features are obtained; the fusion features are input into a prediction model to perform fault prediction including fault types and fault probabilities for the photovoltaic array, the energy storage battery pack, and the energy storage converter, respectively, and to perform residual service life prediction for the photovoltaic array, the energy storage battery pack, and the energy storage converter, respectively; the prediction results are used to generate a scheduling strategy with a state space composed of fault types, fault probabilities, and real-time electricity prices, and an action space composed of energy storage battery pack charging and discharging power through a reinforcement learning algorithm, and a reward function is set according to device fault penalties of the optical storage system, SOC health of the energy storage battery, and photovoltaic power generation light abandonment penalties; an experience replay pool is constructed according to actual operation data of the optical storage system, and a policy network of the reinforcement learning algorithm is regularly updated based on the experience replay pool; the actual operation data comprises action data of the action space at a previous time step, state data of the state space at a current step, and an immediate reward value of the reward function at the previous time step.

[0069] In some possible implementation manners, the intelligent scheduling and fault prediction method of the optical storage system of the present disclosure can be implemented in the form of a program product, which comprises program codes for causing terminal equipment to execute the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of the present specification when the program product is run on the terminal equipment.

[0070] The storage medium of the present disclosure can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0071] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent scheduling and fault prediction of a photovoltaic storage system, characterized in that: The photovoltaic storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid-connected interface. The method includes the following steps: S1. Multimodal sensor arrays deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters collect electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time, and preprocess the collected data. S2. Input the preprocessed data into the Transformer-based multimodal fusion model, unify the feature representation through the cross-attention mechanism, and obtain the fused features; S3. Input the fused features into the prediction model to perform fault predictions, including fault type and fault probability, for the PV array, energy storage battery pack, and energy storage converter, as well as remaining useful life predictions for each. S4. The prediction results are applied to a reinforcement learning algorithm to generate a state space consisting of fault type, fault probability, and real-time electricity price. A scheduling strategy is developed using the charge and discharge power of the energy storage battery as the action space. A reward function is then set based on the goals of punishing PV-storage system equipment failures, PV battery SOC health, and PV power curtailment. S5. Build an experience replay pool based on the actual operating data of the solar-powered storage system, and regularly update the policy network of the reinforcement learning algorithm based on the experience replay pool; the actual operating data includes the action data of the action space in the previous time step, the state data of the state space in the current time step, and the immediate reward value of the reward function in the previous time step.

2. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 1, characterized in that: The specific steps of step S1 are as follows: S11. Collecting the string voltage and current of the photovoltaic array through the smart meter, collecting thermal imaging data of the photovoltaic array through the infrared camera, and collecting acoustic signals of the photovoltaic array through the microphone array; It should be noted that hot spot faults in photovoltaic arrays can be located through thermal imaging data, and loose or cracked components in photovoltaic arrays can be detected through acoustic signals; S12. Collecting the battery cell voltage and temperature of the energy storage battery pack through the battery management controller BMS; S13. Collect the heat sink temperature of the energy storage converter through the temperature sensor; S14. Obtain environmental data through the weather station system, including light intensity, temperature, humidity and wind speed; S15. Obtain structured historical operation and maintenance records through the operation and maintenance system interface; S16. Process the photovoltaic array string voltage, string current, and battery cell voltage of the energy battery group using sliding serial port normalization based on the environmental data, and calculate the mean and standard deviation characteristics within the window; S17. Use a convolution enhancement algorithm to perform temperature enhancement on the thermal imaging data of the photovoltaic array, the battery cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter; S18. Convert the acoustic signal of the photovoltaic array into acoustic characteristics in the frequency domain through short-time Fourier transform; S19. Use the BERT model to generate text embedding features for historical operation and maintenance records.

3. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 2, characterized in that: The specific steps of step S2 are as follows: S21. Use the LSTM model to perform temporal feature encoding on the PV array string voltage, string current, and battery cell voltage using the mean and standard deviation features of the environmental data, and calculate the hidden state. S22. Use the CNN algorithm to perform image feature encoding on the enhanced features of the thermal imaging data of the photovoltaic array, the battery cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter; S23. Use a cross-modal attention fusion algorithm to fuse the temporal feature encoding with the image feature encoding to obtain temporal image features; S24. Modally align the acoustic features of the photovoltaic array with the text embedding features of the historical operation and maintenance records, and then combine them with the time series image features to form a global fusion feature.

4. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 3, characterized in that: The specific steps of step S3 are as follows: S31. Pre-build fault prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using a deep neural network (DNN). Train the models using historical operating data, fault status, and actual fault types. S32. Pre-build life prediction models for the photovoltaic array, energy storage battery pack, and energy storage converter using the LSTM model framework, and train them using historical runtime data and actual lifespan. S33. Fusion features Input the fault prediction model to output the fault type and corresponding fault probability; in, is the Sigmoid function, is the fault type, is the failure probability, is the weight coefficient of the deep convolutional network, is the bias term; S34. Input the time series data of the fusion features into the life prediction model to predict the remaining life of the photovoltaic array, energy storage battery pack, and energy storage converter; in, is the time series data of fusion features, is the remaining life prediction value vector of the photovoltaic array, energy storage battery group, and energy storage converter.

5. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 4, characterized in that: The specific steps of step S4 are as follows: S41. Constructing the state space of reinforcement learning: in, is the real-time electricity price, is the failure probability of the PV array, is the remaining life of the PV array, is the failure probability of the energy storage battery pack, is the remaining life of the energy storage battery pack, is the failure probability of the energy storage converter, The remaining life of the energy storage battery; S42 defines the action space as the charge and discharge power of the energy storage battery pack; in, is the maximum charging power of the energy storage battery pack, The maximum discharge power of the Chunnigeng battery pack, is the charge and discharge power control value of the energy storage battery pack; S43. Constructing a multi-objective reward function: in, is the interaction power between the energy storage system and the grid, is the real-time electricity price, is the photovoltaic fault penalty coefficient, is the battery failure penalty coefficient, is the converter fault penalty coefficient, is the SOC health weight coefficient of the energy storage battery, For the optimal SOC of the energy storage battery, Real-time SOC of energy storage battery, is the auxiliary service reward coefficient, For grid auxiliary services, is the light abandonment penalty coefficient, is the photovoltaic power generation power at time t, It is the photovoltaic power actually absorbed by energy storage or grid.

6. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 5, characterized in that: The specific steps of step S5 are as follows: S51. In the actual operation of the solar energy storage system, the state of each time step t is ,action , instant rewards and the next state It is stored as an experience data in the experience replay pool D, and the reward value is placed first. Experience samples with high or large state changes; Among them, P represents the experience placement priority; S52. Use the deep deterministic policy gradient algorithm DDPG for policy optimization and construct a loss function for the value evaluation network Critic: in, is the current parameter of the value evaluation network Critic, is the target parameter of the value assessment network Critic; is the current parameter of the policy network Actor, is the target parameter of the policy network Actor, is the discount factor, N is the number of experience samples sampled in the experience replay pool; Based on the multi-objective function Calculated; S53. Update the strategy network Actor based on the gradient direction of the loss function of the value evaluation network Critic: in, Indicates the gradient of the policy network parameter θ is the objective function of the policy network Actor, E is the mathematical expectation, that is, the batch average operation, is to find the gradient of action a, Represents the state-action value function output by the value evaluation network Critic, that is, in state Take action The long-term return forecast value, Indicates that the policy network Actor network is in state Output action when S54. Acquire actual operation data of the solar storage system, the actual operation data including the state vector and the actual fault label , and the state vector Input the trained fault prediction model to obtain the predicted fault label, and construct a cross-entropy loss function to measure the difference between the predicted fault label and the actual fault label: Where C is the total number of fault types, is the code of the actual fault label, is the predicted probability of type c failure; S55. Based on the cross entropy loss function, the model parameters of the fault prediction model are updated through the back propagation algorithm.

7. The method for intelligent scheduling and fault prediction of a photovoltaic storage system according to claim 6, characterized in that: The specific steps of step S55 are as follows: S551. Derivative the cross entropy loss function based on the fault prediction model parameters to obtain the gradient of the model parameters; S552. Update the parameters of the fault prediction model based on the gradient of the model parameters and according to the set learning rate.

8. A method and device for intelligent scheduling and fault prediction of a photovoltaic storage system, characterized in that: The photovoltaic storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter and a grid-connected interface, and the device includes: The data acquisition and processing module is used to collect electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time through a multimodal sensor array deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters, and to pre-process the collected data. The data fusion module is used to input the preprocessed data into the Transformer-based multimodal fusion model, unify the feature representation through the cross-attention mechanism, and obtain the fused features; The fault and life prediction module is used to input the fusion features into the prediction model to perform fault prediction including fault type and fault probability for the photovoltaic array, energy storage battery pack, and energy storage converter, as well as the remaining service life prediction for each. The intelligent scheduling module uses a reinforcement learning algorithm to generate a state space composed of fault type, fault probability, and real-time electricity price based on the prediction results. It also uses the charge and discharge power of the energy storage battery pack as the action space to formulate a scheduling strategy. The module also sets a reward function based on the equipment failure penalty of the photovoltaic storage system, the state of charge (SOC) health of the energy storage battery, and the penalty for photovoltaic power curtailment. A learning optimization module is used to build an experience replay pool based on the actual operating data of the solar energy storage system and regularly update the policy network of the reinforcement learning algorithm based on the experience replay pool; the actual operating data includes the action data of the action space in the previous time step, the state data of the state space in the current time step, and the immediate reward value of the reward function in the previous time step.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method implements the steps of the method for intelligent scheduling and fault prediction of a photovoltaic storage system as claimed in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent scheduling and fault prediction of a photovoltaic storage system as claimed in any one of claims 1 to 7 are implemented.

Images