An intelligent scheduling and fault prediction method, system, device and medium for a light storage system

Abstract

The application provides a kind of optical storage system intelligent scheduling and fault prediction method, device, equipment and medium, belong to intelligent optical storage operation and maintenance technical field, the method: real-time acquisition optical storage system data and pre-processing;The pre-processed data is input into the multimodal fusion model, unified feature representation is carried out through cross attention mechanism, and the fusion feature is obtained;The fusion feature is input into the prediction model to carry out fault prediction and remaining useful life prediction;The prediction result is generated by reinforcement learning algorithm to form the state space with fault type, fault probability and real-time electricity price, and the scheduling strategy with the action space of energy storage battery pack charging and discharging power is set, and the reward function is set;According to the actual operation data of optical storage system, the experience playback pool is constructed, and the strategy network of reinforcement learning algorithm is regularly updated.The application collects data through multimodal sensor, combines deep learning and reinforcement learning, realizes intelligent fault prediction and dynamic scheduling of optical storage system, and improves operation efficiency and reliability.

Description

Technical Field

[0001] This application belongs to the field of intelligent photovoltaic energy storage operation and maintenance technology, specifically involving a method, system, equipment and medium for intelligent scheduling and fault prediction of photovoltaic energy storage systems. Background Technology

[0002] With the development of new energy sources, photovoltaic (PV) and energy storage systems, as efficient renewable energy solutions, can effectively address the intermittency and instability of photovoltaic power generation, playing a crucial role in improving energy efficiency and power supply reliability. However, PV and energy storage systems face numerous challenges during operation.

[0003] First, traditional scheduling methods are mainly based on fixed rules or simplified models, such as simple strategies like low-cost charging and high-cost discharging, or algorithms like particle swarm optimization. However, these methods are difficult to adapt to the randomness of renewable energy generation and the rapid fluctuations in load. Furthermore, due to the singular optimization objective, they often get stuck in local optima and lack consideration for the dynamic changes in equipment health status, leading to a disconnect between scheduling strategies and actual operating conditions. Second, current photovoltaic-storage systems mostly rely on manual inspections, periodic maintenance, or threshold monitoring from a single data source, which is inefficient and makes it difficult to accurately identify early faults or complex fault models. For example, traditional fault monitoring is usually based on electrical parameters (current, voltage) or simple threshold judgments, making it difficult to predict potential faults in advance and assess their impact on scheduling strategies. In addition, existing fault prediction methods often lack the comprehensive utilization of multi-source information, limiting the accuracy and reliability of fault prediction. Third, current photovoltaic-storage systems typically treat scheduling and fault diagnosis as independent modules. Scheduling strategies do not fully consider the potential fault risks of equipment, and fault diagnosis information is not fed back to the scheduling system in real time. Summary of the Invention

[0004] In a first aspect, embodiments of this application provide a method for intelligent scheduling and fault prediction of a photovoltaic-storage system. The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. The method includes the following steps: S1. Through a multi-modal sensor array deployed in photovoltaic arrays, energy storage battery packs, and energy storage converters, electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records are collected in real time, and the collected data is preprocessed. S2. Input the preprocessed data into the Transformer-based multimodal fusion model, and use the cross-attention mechanism to perform unified feature representation to obtain fused features; S3. Input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs, and energy storage converters, and to perform remaining service life prediction for each. S4. The prediction results are used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm. A scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space is set, and a reward function is set based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. S5. Construct an experience replay pool based on the actual operating data of the optical storage system, and periodically 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 step, and the instantaneous reward value of the reward function in the previous time step.

[0005] Furthermore, the specific steps of step S1 are as follows: S11. Collect the string voltage and string current of the photovoltaic array through a smart meter, collect the thermal imaging data of the photovoltaic array through an infrared camera, and collect the acoustic signals of the photovoltaic array through a microphone array; It should be noted that thermal imaging data can be used to locate hot spot faults in photovoltaic arrays, while acoustic signals can be used to detect loose or cracked components in photovoltaic arrays. S12. Collect the individual cell voltage and temperature of the energy storage battery pack through the battery management controller (BMS); S13. Collect the temperature of the heat sink of the energy storage converter using a temperature sensor; S14. Obtain environmental data through a 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. The string voltage, string current, and individual cell voltage of the photovoltaic array are processed using sliding serial port normalization based on environmental data, and the mean and standard deviation characteristics within the window are calculated. S17. Use convolutional enhancement algorithm to enhance the temperature of the photovoltaic array thermal imaging data, the cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter. S18. The acoustic signal of the photovoltaic array is converted into the acoustic characteristics in the frequency domain by short-time Fourier transform; S19. Use the BERT model to generate text embedding features for historical operation and maintenance records.

[0006] Furthermore, the specific steps of step S2 are as follows: S21. The string voltage, string current, and individual cell voltage of the photovoltaic array are encoded using the mean and standard deviation of environmental data using an LSTM model to calculate the hidden state. S22. Enhanced features of 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 encoded using a CNN algorithm; S23. Use a cross-modal attention fusion algorithm to fuse temporal feature encoding and image feature encoding to obtain temporal image features; S24. After modally aligning the acoustic features of the photovoltaic array with the text embedding features of historical operation and maintenance records, the features are combined 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 photovoltaic arrays, energy storage battery packs, and energy storage converters using deep neural networks (DNNs), and train them using historical operating data, whether a fault exists, and the actual fault type. S32. Pre-build lifetime prediction models for photovoltaic arrays, energy storage battery packs, and energy storage converters using the LSTM model framework, and train them using historical runtime sequence data and actual lifetimes. S33. Merge features Input the fault prediction model and output the fault type and corresponding fault probability;

[0008] in, For the Sigmoid function, Fault type This represents the probability of failure. These are the weight coefficients of the deep convolutional network. For bias terms; S34. Input the time series data of the fused features into the lifetime prediction model to predict the remaining lifetime of photovoltaic arrays, energy storage battery packs, and energy storage converters;

[0009] in, For time series data with fused features, This is a vector of predicted remaining lifetime values ​​for photovoltaic arrays, energy storage battery packs, and energy storage converters.

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

[0011] in, For real-time electricity prices, The failure probability of the photovoltaic array. For the remaining lifetime of the photovoltaic array, The failure probability of the energy storage battery pack. For the remaining lifespan of the energy storage battery pack, The failure probability of the energy storage converter. The remaining lifespan of the energy storage battery; S42. Define the action space as the charging and discharging power of the energy storage battery pack;

[0012] in, This represents the maximum charging power of the energy storage battery pack. This represents the maximum discharge power of the Chunni battery pack. For the charging and discharging power control values ​​of the energy storage battery pack; S43. Construct a multi-objective reward function:

[0013] in, It is the interaction power between the energy storage system and the power grid. For real-time electricity prices, This is the photovoltaic fault penalty coefficient. This is the battery failure penalty coefficient. This is the converter fault penalty coefficient. The SOC health weighting coefficient for energy storage batteries. For optimal SOC of energy storage batteries, For real-time SOC of energy storage batteries, For auxiliary service reward coefficient, For power grid ancillary services, This is the penalty coefficient for discarded light. Let be the photovoltaic power generation at time t. This refers to the actual photovoltaic power that is stored or absorbed by the grid.

[0014] Furthermore, the specific steps of step S5 are as follows: S51. During the actual operation of the photovoltaic storage system, the state at each time step t will be recorded. ,action Instant rewards And the next state Experience data is stored as a single data point in the experience replay pool D, with reward values ​​placed there first. Experienced samples with high or large state changes;

[0015]

[0016] Where P represents the priority of experience placement; S52. The Deep Deterministic Policy Gradient Algorithm (DDPG) is used for policy optimization, and a loss function is constructed for the value assessment network Critic:

[0017] in, The current parameters of the Critic value assessment network. The target parameters for the Critic value assessment network; These are the current parameters of the policy network Actor. For the target parameters of the policy network Actor, Here, N is the discount factor, and N is the number of empirical samples sampled in the empirical replay pool. It is based on a multi-objective function Calculated; S53. Gradient direction update strategy for the loss function of the value evaluation network Critic: Network Actor:

[0018] in, This represents calculating the gradient with respect to the policy network parameters θ. It is the objective function of the policy network Actor, where E is the mathematical expectation, i.e., the batch average operation. It is an action Find the gradient. This represents the state-action value function output by the Critic value assessment network, i.e., the state at which the network is in action. Take action at the time The long-term earnings forecast, The policy network Actor network represents the state. The action output at that time; S54. Obtain the actual operating data of the photovoltaic storage system, the actual operating data including the state vector. and actual fault labels and the state vector The predicted fault labels are obtained by inputting them into the trained fault prediction model. A cross-entropy loss function is then constructed to measure the difference between the predicted fault labels and the actual fault labels.

[0019] in, The total number of fault types. The encoding of the actual fault label. The probability of the predicted type c fault; S55. The model parameters of the fault prediction model are updated based on the cross-entropy loss function and the backpropagation algorithm.

[0020] Furthermore, the specific steps of step S55 are as follows: S551. Take the derivative of 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.

[0021] Secondly, embodiments of this application also provide a method and apparatus for intelligent scheduling and fault prediction of a photovoltaic-storage system. The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. The apparatus 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 multi-modal sensor array deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters, and to preprocess the collected data. The data fusion module is used to input preprocessed data into a Transformer-based multimodal fusion model, and obtain fused features by performing unified feature representation through a cross-attention mechanism. The fault and lifetime prediction module is used to input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs and energy storage converters, and to predict the remaining lifetime for each. The intelligent scheduling module is used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm, and a scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space. It also sets a reward function based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. The learning optimization module is used to construct an experience replay pool based on the actual operating data of the optical storage system, and to periodically 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 step, and the instantaneous reward value of the reward function in the previous time step.

[0022] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent scheduling and fault prediction method for the photovoltaic storage system as described in the first aspect.

[0023] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the intelligent scheduling and fault prediction method for the optical storage system as described in the first aspect.

[0024] As can be seen from the above technical solutions, this application has the following advantages: The intelligent scheduling and fault prediction method, system, equipment and medium for photovoltaic and energy storage systems provided in this application integrate multimodal data acquisition, deep learning fusion, reinforcement learning scheduling and online adaptive updates to realize intelligent operation and health management of photovoltaic and energy storage systems. This can improve the operating efficiency and reliability of photovoltaic and energy storage systems, reduce operation and maintenance costs, extend equipment life, and enhance the overall performance and adaptability of photovoltaic and energy storage systems. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating the intelligent scheduling and fault prediction method for the photovoltaic-storage system of the present invention.

[0027] Figure 2 This is a schematic diagram of the intelligent scheduling and fault prediction device for the photovoltaic energy storage system of the present invention. Detailed Implementation

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

[0029] For example, with the rapid development of the new energy industry, photovoltaic (PV) and energy storage systems, as a highly efficient renewable energy solution, are of great significance for improving energy utilization efficiency and enhancing power supply reliability. However, in actual operation, PV and energy storage systems face a series of problems that urgently need to be solved.

[0030] On the one hand, traditional solar-storage system scheduling strategies typically rely on fixed rules or simplified models. For example, common low-cost charging and high-cost discharging strategies, as well as particle swarm optimization algorithms, can achieve basic scheduling functions to a certain extent, but these methods are difficult to adapt to the randomness, volatility, and rapid changes in load demand of renewable energy generation. Furthermore, because these traditional methods have relatively singular optimization objectives, they are prone to getting trapped in local optima and cannot fully consider the dynamic changes in equipment health, leading to a disconnect between the scheduling strategy and the actual operating state.

[0031] On the other hand, current fault diagnosis in photovoltaic and energy storage systems mainly relies on manual inspections, periodic maintenance, or threshold monitoring based on a single data source. These methods are not only inefficient but also struggle to accurately identify early-stage or complex fault modes. For example, traditional fault monitoring typically relies solely on electrical parameters (such as current and voltage) or simple threshold judgments, making it difficult to predict potential faults in advance and assess their impact on scheduling strategies. Furthermore, existing fault prediction methods often fail to fully utilize multi-source information, which significantly limits the accuracy and reliability of fault prediction.

[0032] Finally, current photovoltaic-storage systems typically treat scheduling and fault diagnosis as two independent modules, lacking an effective coordination mechanism. Scheduling strategies fail to adequately consider potential equipment failure risks, and fault diagnosis information is not fed back to the scheduling system in real time, thus impacting the overall system efficiency and reliability.

[0033] To address the aforementioned issues, this embodiment provides an intelligent scheduling and fault prediction method for a photovoltaic-storage system. By comprehensively collecting operational data of the photovoltaic-storage system through a multi-modal sensor array and combining deep learning and reinforcement learning algorithms, intelligent collaborative optimization of fault prediction and dynamic scheduling is achieved, ensuring that the system can quickly respond to dynamic changes and improving operational efficiency and reliability.

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 The diagram shows a flowchart of an intelligent scheduling and fault prediction method for a photovoltaic-storage system in a specific embodiment. The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. The method includes the following steps: S1. Through a multi-modal sensor array deployed in photovoltaic arrays, energy storage battery packs, and energy storage converters, electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records are collected in real time, and the collected data is preprocessed. It should be noted that by collecting data from multiple modalities, multi-source inputs are provided for deep learning models. Early faults are detected through thermal imaging and acoustic signals, improving the accuracy of fault diagnosis, enhancing data quality, and providing a foundation for subsequent feature extraction and model training. S2. Input the preprocessed data into the Transformer-based multimodal fusion model, and use the cross-attention mechanism to perform unified feature representation to obtain fused features; It should be noted that by using Transformer and cross-attention mechanism, features from different modalities are deeply fused to improve the diversity and accuracy of feature representation. By generating global fused features, a unified feature representation is provided for subsequent fault prediction and lifetime prediction. S3. Input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs, and energy storage converters, and to perform remaining service life prediction for each. It should be noted that by using DNN and LSTM to predict fault types and probabilities, the accuracy and reliability of fault prediction can be improved. The remaining service life of the equipment can be predicted by the LSTM model, providing a basis for preventive maintenance. By combining fault prediction and life prediction models, a comprehensive assessment of the health status of the equipment can be achieved. S4. The prediction results are used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm. A scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space is set, and a reward function is set based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. It should be noted that the dynamic scheduling strategy generated by the reinforcement learning algorithm can optimize scheduling based on real-time electricity price and equipment health status. By comprehensively considering equipment failure penalties, battery SOC health and curtailment penalties, multi-objective optimization is achieved. The strategy 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 based on the actual operating data of the optical storage system, and periodically 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 step, and the instantaneous reward value of the reward function in 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 enhance the model's ability to identify new fault modes by updating the parameters of the fault prediction model online. The model parameters are updated based on real-time operating data to ensure that the system can quickly adapt to dynamic changes.

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

[0037] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, another intelligent scheduling and fault prediction method for a photovoltaic-storage system is provided. The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. The method includes the following steps: S1. Using a multi-modal sensor array deployed in photovoltaic arrays, energy storage battery packs, and energy storage converters, electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records are collected in real time, and the collected data is preprocessed; the specific steps of step S1 are as follows: S11. Collect the string voltage and string current of the photovoltaic array through a smart meter, collect the thermal imaging data of the photovoltaic array through an infrared camera, and collect the acoustic signals of the photovoltaic array through a microphone array; It should be noted that thermal imaging data can be used to locate hot spot faults in photovoltaic arrays, while acoustic signals can be used to detect loose or cracked components in photovoltaic arrays. S12. Collect the individual cell voltage and temperature of the energy storage battery pack through the battery management controller (BMS); S13. Collect the temperature of the heat sink of the energy storage converter using a temperature sensor; S14. Obtain environmental data through a 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. The string voltage, string current, and individual cell voltage of the photovoltaic array are processed using sliding serial port normalization based on environmental data, and the mean and standard deviation characteristics within the window are calculated.

[0038] in, , The mean and standard deviation within the window; S17. Use convolutional enhancement algorithm to enhance the temperature of the photovoltaic array thermal imaging data, the cell temperature of the energy storage battery, and the heat sink temperature of the energy storage converter.

[0039] in, These are the convolution weight coefficients. For bias terms, Indicates the temperature value. This represents the value of convolution enhancement; S18. The acoustic signal of the photovoltaic array is converted into the acoustic characteristics in the frequency domain by short-time Fourier transform;

[0040] S19. Use the BERT model to generate text embedding features for historical operation and maintenance records;

[0041] It should be noted that data from multiple modalities, including electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records, are collected to provide diverse inputs for subsequent deep learning models. Early faults are detected through thermal imaging and acoustic signals, improving the accuracy of fault diagnosis. Preprocessing steps such as normalization, convolution enhancement, and frequency domain transformation are used to improve data quality, providing a foundation for subsequent feature extraction and model training. For example, taking a certain moment when the string voltage of the photovoltaic array is 24V, the string current is 8A, the ambient temperature is 25°C, and the irradiance is 800W / m², after normalization through a sliding window, the mean voltage is 24V and the standard deviation is 0.5V; the mean current is 8A and the standard deviation is 0.2A. S2. Input the preprocessed data into the Transformer-based multimodal fusion model, and perform unified feature representation through a cross-attention mechanism to obtain fused features; the specific steps of step S2 are as follows: S21. The string voltage, string current, and individual cell voltage of the photovoltaic array are encoded using the mean and standard deviation of environmental data using an LSTM model to calculate the hidden state.

[0042] in, This indicates the characteristics of the mean and standard deviation of voltage. This indicates the characteristics of the mean and standard deviation of the current. This indicates the characteristics of the mean and standard deviation of environmental data. Indicates a hidden state; S22. Enhanced features of 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 encoded using a CNN algorithm;

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

[0044] in, Q , K , V For query, key-value matrix, This is the dimension scaling factor; S24. After modally aligning the acoustic features of the photovoltaic array with the text embedding features of historical operation and maintenance records, concatenate them with time-series image features to form a global fusion feature. ; It should be noted that by using Transformer and cross-attention mechanism, features from different modalities are deeply fused to improve the diversity and accuracy of feature representation. By combining temporal features and image features, the system's operating status can be more comprehensively reflected. By generating global fused features, a unified feature representation can be provided for subsequent fault prediction and lifetime prediction. For example, suppose that after encoding by the LSTM model, the hidden state of the electrical parameters 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, and to predict the remaining service life for each. The specific steps of step S3 are as follows: S31. Pre-build fault prediction models for photovoltaic arrays, energy storage battery packs, and energy storage converters using deep neural networks (DNNs), and train them using historical operating data, whether a fault exists, and the actual fault type. S32. Pre-build lifetime prediction models for photovoltaic arrays, energy storage battery packs, and energy storage converters using the LSTM model framework, and train them using historical runtime sequence data and actual lifetimes. S33. Merge features Input the fault prediction model and output the fault type and corresponding fault probability;

[0045] in, For the Sigmoid function, Fault type This represents the probability of failure. These are the weight coefficients of the deep convolutional network. For bias terms; S34. Input the time series data of the fused features into the lifetime prediction model to predict the remaining lifetime of photovoltaic arrays, energy storage battery packs, and energy storage converters;

[0046] in, For time series data with fused features, This is a vector of predicted remaining lifetime values ​​for photovoltaic arrays, energy storage battery packs, and energy storage converters. It should be noted that deep neural networks (DNN) and long short-term memory networks (LSTM) are used to predict fault types and probabilities, which improves the accuracy and reliability of fault prediction. The remaining service life of the equipment is predicted by the LSTM model, providing a basis for preventive maintenance. By combining fault prediction and service life prediction models, a comprehensive assessment of the health status of the equipment can be achieved. For example, taking the input fusion features as [0.35, 0.45, 0.55, 0.1, 0.2, 0.8, 0.9], the DNN model predicts the fault type of the photovoltaic array as "hot spot fault" with a fault probability of 0.8; the LSTM model predicts the remaining service life of the photovoltaic array as 1000 hours. S4. The prediction results are used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm. A scheduling strategy is used with the charging and discharging power of the energy storage battery pack as the action space. A reward function is set based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. The specific steps of step S4 are as follows: S41. Constructing the state space for reinforcement learning:

[0047] in, For real-time electricity prices, The failure probability of the photovoltaic array. For the remaining lifetime of the photovoltaic array, The failure probability of the energy storage battery pack. For the remaining lifespan of the energy storage battery pack, The failure probability of the energy storage converter. The remaining lifespan of the energy storage battery; S42. Define the action space as the charging and discharging power of the energy storage battery pack;

[0048] in, This represents the maximum charging power of the energy storage battery pack. This represents the maximum discharge power of the Chunni battery pack. For the charging and discharging power control values ​​of the energy storage battery pack; S43. Construct a multi-objective reward function:

[0049] in, It is the interaction power between the energy storage system and the power grid. For real-time electricity prices, This is the photovoltaic fault penalty coefficient. This is the battery failure penalty coefficient. This is the converter fault penalty coefficient. The SOC health weighting coefficient for energy storage batteries. For optimal SOC of energy storage batteries, For real-time SOC of energy storage batteries, For auxiliary service reward coefficient, For power grid ancillary services, This is the penalty coefficient for discarded light. Let be the photovoltaic power generation at time t. The actual photovoltaic power that is stored or absorbed by the grid; It should be noted that grid ancillary services include frequency regulation and standby services for photovoltaic and energy storage systems;

[0050] The sum of the power generated by the interaction between the energy storage system and the grid at time t and the charging and discharging power of the energy storage battery is calculated; the actual photovoltaic power stored or consumed. Take the smaller value between the power and the charging / discharging power of the energy storage battery; Calculate the photovoltaic power generation at time t Actual photovoltaic power stored or consumed The power difference, the curtailment penalty term is the actual photovoltaic power that is stored or consumed. The power difference is taken when it is greater than 0, and 0 is taken when it is less than 0. It should be noted that the dynamic scheduling strategy generated by the reinforcement learning algorithm can optimize scheduling based on real-time electricity price and equipment health status. By comprehensively considering equipment failure penalties, battery SOC health and curtailment penalties, multi-objective optimization is achieved. By periodically updating the strategy network based on the experience replay pool, it is ensured that the scheduling strategy can adapt to the dynamic changes of the system. S5. Construct an experience replay pool based on the actual operating data of the optical storage system, and periodically 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 step, and the instantaneous reward value of the reward function in the previous time step; the specific steps of step S5 are as follows: S51. During the actual operation of the photovoltaic storage system, the state at each time step t will be recorded. ,action Instant rewards And the next state Experience data is stored as a single data point in the experience replay pool D, with reward values ​​placed there first. Experienced samples with high or large state changes;

[0051]

[0052] Where P represents the experience placement priority; S52. The Deep Deterministic Policy Gradient Algorithm (DDPG) is used for policy optimization to construct the loss function for the value assessment network Critic:

[0053] in, The current parameters of the Critic value assessment network. The target parameters for the Critic value assessment network; These are the current parameters of the policy network Actor. For the target parameters of the policy network Actor, Here, N is the discount factor, and N is the number of empirical samples sampled in the empirical replay pool. It is based on a multi-objective function Calculated; For example, Take 0.95; S53. Gradient direction update strategy for the loss function of the value evaluation network Critic: Network Actor:

[0054] in, This represents calculating the gradient with respect to the policy network parameters θ. It is the objective function of the policy network Actor, where E is the mathematical expectation, i.e., the batch average operation. It is about finding the gradient of action a. This represents the state-action value function output by the Critic value assessment network, i.e., the state at which the network is in action. Take action at the time The long-term earnings forecast, The policy network Actor network represents the state. The action output at that time; It should be noted that, It is the direction of parameter θ update to maximize the expected return. It is a value assessment network Critic for actions The scoring gradient guides the policy network Actor to adjust its actions to obtain higher returns; Replace action 'a' with the policy network in state 'a'. The actual output action This ensures that gradient calculations are based on actions generated by the current policy, rather than arbitrary actions. Policy network parameters θ For output actions The gradient influences the parameter update direction. The overall meaning of the formula is that the policy network Actor updates its parameters along the Q-value gradient direction evaluated by the value evaluation network Critic, so that the policy gradually tends to the direction of maximizing the expected return. S54. Obtain the actual operating data of the photovoltaic storage system, the actual operating data including the state vector. and actual fault labels and the state vector The predicted fault labels are obtained by inputting them into the trained fault prediction model. A cross-entropy loss function is then constructed to measure the difference between the predicted fault labels and the actual fault labels.

[0055] Where C represents the total number of fault types. The encoding of the actual fault label. The probability of the predicted type c fault; S55. Update the model parameters of the fault prediction model based on the cross-entropy loss function and through the backpropagation algorithm; the specific steps of step S55 are as follows: S551. Take the derivative of 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; It should be noted that the fault prediction model can continuously adjust its own parameters based on real-time operating data, thereby better adapting to the dynamic changes and new fault modes of the photovoltaic-storage system. The scheduling strategy is optimized by adopting the Deep Deterministic Strategy Gradient (DDPG) algorithm, which improves the stability and adaptability of the strategy. The model's ability to identify new fault modes is enhanced by updating the parameters of the fault prediction model online. The model parameters are updated based on real-time running data to ensure that the system can quickly adapt to dynamic changes. For example, at a certain time step, the state The action is [0.5, 0.8, 1000, 0.2, 800]. Taking 50kW as an example, instant reward The value is -0.3, the next state. The values ​​are [0.5, 0.7, 999, 0.2, 799]. These data are stored in the experience replay pool, and the policy network parameters are updated using the DDPG algorithm to optimize the scheduling policy. At the same time, the parameters of the fault prediction model are updated using the cross-entropy loss function to improve the model's ability to identify new fault modes.

[0056] It should be understood that the sequence number of each step in the above embodiments does not imply 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 are embodiments of the intelligent scheduling and fault prediction device for photovoltaic and energy storage systems provided in this disclosure. This system and the intelligent scheduling and fault prediction methods for photovoltaic and energy storage systems in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the intelligent scheduling and fault prediction device for photovoltaic and energy storage systems, please refer to the embodiments of the intelligent scheduling and fault prediction methods for photovoltaic and energy storage systems described above.

[0058] The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. 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 multi-modal sensor array deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters, and to preprocess the collected data. The data fusion module is used to input preprocessed data into a Transformer-based multimodal fusion model, and obtain fused features by performing unified feature representation through a cross-attention mechanism. The fault and lifetime prediction module is used to input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs and energy storage converters, and to predict the remaining lifetime for each. The intelligent scheduling module is used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm, and a scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space. It also sets a reward function based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. The learning optimization module is used to construct an experience replay pool based on the actual operating data of the optical storage system, and to periodically 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 step, and the instantaneous reward value of the reward function in the previous time step.

[0059] This embodiment achieves intelligent fault prediction and dynamic scheduling of the photovoltaic storage system through the interactive collaboration of the data acquisition and processing module, data fusion module, fault and lifespan prediction module, intelligent scheduling module, and learning optimization module, thereby improving operational efficiency and reliability.

[0060] The intelligent scheduling and fault prediction method for photovoltaic storage systems provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, 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 this application described and / or claimed herein.

[0061] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0062] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0063] A processor may include one or more processing units, such as a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0064] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0065] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0066] The aforementioned electronic equipment implements the intelligent scheduling and fault prediction method for the photovoltaic-storage system proposed in this application. It utilizes a multi-modal sensor array deployed on the photovoltaic array, energy storage battery bank, and energy storage converter to collect electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records in real time. The collected data is preprocessed; the preprocessed data is input into a Transformer-based multi-modal fusion model, and a unified feature representation is obtained through a cross-attention mechanism. The fused features are then input into a prediction model to perform fault predictions (including fault type and probability) for the photovoltaic array, energy storage battery bank, and energy storage converter, as well as remaining service life predictions. The prediction results are then used to generate a state space composed of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm. This paper proposes a scheduling strategy that uses the charging and discharging power of the energy storage battery pack as the action space, and sets a reward function based on the equipment failure penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation. An experience replay pool is constructed based on the actual operating data of the photovoltaic-storage system, and the policy network of the reinforcement learning algorithm is periodically updated 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 step, and the instantaneous reward value of the reward function in the previous time step. This technical solution achieves the beneficial effect of comprehensively collecting the operating data of the photovoltaic-storage system through a multi-modal sensor array, combined with deep learning and reinforcement learning algorithms, to realize intelligent collaborative optimization of fault prediction and dynamic scheduling, ensuring the system's rapid response to dynamic changes and improving operational efficiency and reliability.

[0067] The storage medium provided in this application stores a program product capable of implementing intelligent scheduling and fault prediction methods for optical storage systems.

[0068] The intelligent scheduling and fault prediction method for photovoltaic-storage systems includes: 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 on the photovoltaic array, energy storage battery pack, and energy storage converter; preprocessing the acquired data; inputting the preprocessed data into a Transformer-based multi-modal fusion model, and performing unified feature representation through a cross-attention mechanism to obtain fused features; inputting the fused features into a 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 utilization prediction for each. Lifetime prediction; the prediction results are used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm, and a scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space. A reward function is set based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery, and the curtailment penalty of photovoltaic power generation as objectives. An experience replay pool is constructed based on the actual operating data of the photovoltaic-storage system, and the policy network of the reinforcement learning algorithm is updated periodically 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 step, and the real-time reward value of the reward function in the previous time step.

[0069] In some possible implementations, the intelligent scheduling and fault prediction method for photovoltaic and energy storage systems disclosed herein can be implemented as a program product, which includes program code. When the program product is run on a terminal device, the program code is used to cause the terminal device to perform the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of this disclosure.

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

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, 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 connection interface. The method includes the following steps: S1. Through a multi-modal sensor array deployed in photovoltaic arrays, energy storage battery packs, and energy storage converters, electrical parameters, thermal imaging data, acoustic signals, environmental data, and historical operation and maintenance records are collected in real time, and the collected data is preprocessed. Collect string voltage, string current, thermal imaging data, acoustic features corresponding to acoustic signals of photovoltaic arrays, individual cell voltage of energy storage battery packs, heat sink temperature of energy storage converters, environmental data, and text embedding features corresponding to historical operation and maintenance records. S2. Input the preprocessed data into the Transformer-based multimodal fusion model, and use the cross-attention mechanism to perform unified feature representation to obtain fused features; The specific steps of step S2 are as follows: S21. Use the LSTM model to encode the time-series features of the photovoltaic array string voltage, string current, battery cell voltage of the energy storage battery pack, and environmental data mean and standard deviation, and calculate the hidden state. S22. Enhanced features of the thermal imaging data of the photovoltaic array, the cell temperature of the energy storage battery pack, and the heat sink temperature of the energy storage converter are encoded using a CNN algorithm; S23. Use a cross-modal attention fusion algorithm to fuse temporal feature encoding and image feature encoding to obtain temporal image features; S24. After modally aligning the acoustic features of the photovoltaic array with the text embedding features of historical operation and maintenance records, the features are combined with the time-series image features to form a global fusion feature; S3. Input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs, and energy storage converters, and to perform remaining service life prediction for each. The specific steps of step S3 are as follows: S31. Pre-build fault prediction models for photovoltaic arrays, energy storage battery packs, and energy storage converters using deep neural networks (DNNs), and train them using historical operating data, whether a fault exists, and the actual fault type. S32. Pre-build lifetime prediction models for photovoltaic arrays, energy storage battery packs, and energy storage converters using the LSTM model framework, and train them using historical runtime sequence data and actual lifetimes. S33. Merge features Input the fault prediction model and output the fault type and corresponding fault probability; in, For the Sigmoid function, Fault type This represents the probability of failure. These are the weight coefficients of a deep convolutional network. For bias terms; S34. Input the time series data of the fused features into the lifetime prediction model to predict the remaining lifetime of photovoltaic arrays, energy storage battery packs, and energy storage converters; in, For time series data with fused features, This is a vector of predicted remaining lifetime values ​​for photovoltaic arrays, energy storage battery packs, and energy storage converters. S4. The prediction results are used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm. A scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space is set, and a reward function is set based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery pack, and the curtailment penalty of photovoltaic power generation as objectives. S5. Construct an experience replay pool based on the actual operating data of the optical storage system, and periodically 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 step, and the instantaneous reward value of the reward function in the previous time step.

2. The intelligent scheduling and fault prediction method for a photovoltaic-storage system according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Collect the string voltage and string current of the photovoltaic array through a smart meter, collect the thermal imaging data of the photovoltaic array through an infrared camera, and collect the acoustic signals of the photovoltaic array through a microphone array; It should be noted that thermal imaging data can be used to locate hot spot faults in photovoltaic arrays, while acoustic signals can be used to detect loose or cracked components in photovoltaic arrays. S12. Collect the individual cell voltage and temperature of the energy storage battery pack through the battery management controller (BMS); S13. Collect the temperature of the heat sink of the energy storage converter using a temperature sensor; S14. Obtain environmental data through a 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. The string voltage and string current of the photovoltaic array, the individual cell voltage of the energy storage battery pack, and the environmental data are processed by sliding window normalization, and the mean and standard deviation characteristics within the window are calculated. S17. Use convolutional enhancement algorithm to enhance the temperature of the photovoltaic array's thermal imaging data, the cell temperature of the energy storage battery pack, and the heat sink temperature of the energy storage converter. S18. The acoustic signal of the photovoltaic array is converted into the acoustic characteristics in the frequency domain by short-time Fourier transform; S19. Use the BERT model to generate text embedding features for historical operation and maintenance records.

3. The intelligent scheduling and fault prediction method for a photovoltaic-storage system according to claim 2, characterized in that, The specific steps of step S4 are as follows: S41. Constructing the state space for reinforcement learning: in, For real-time electricity prices, The failure probability of the photovoltaic array. For the remaining lifetime of the photovoltaic array, The failure probability of the energy storage battery pack. For the remaining lifespan of the energy storage battery pack, The failure probability of the energy storage converter. The remaining lifespan of the energy storage converter; S42. Define the action space as the charging and discharging power of the energy storage battery pack; in, This represents the maximum charging power of the energy storage battery pack. This represents the maximum discharge power of the energy storage battery pack. For the charging and discharging power control values ​​of the energy storage battery pack; S43. Construct a multi-objective reward function: in, It is the interaction power between the energy storage system and the power grid. For real-time electricity prices, This represents the photovoltaic fault penalty coefficient. This is the battery failure penalty coefficient. This is the converter fault penalty coefficient. This is the SOC health weighting coefficient for energy storage battery packs. For optimal SOC of energy storage battery packs, For real-time SOC of energy storage battery packs, For auxiliary service reward coefficient, For power grid ancillary services, This is the penalty coefficient for discarded light. Let be the photovoltaic power generation at time t. This refers to the actual photovoltaic power that is stored or absorbed by the grid.

4. The intelligent scheduling and fault prediction method for a photovoltaic-storage system according to claim 3, characterized in that, The specific steps of step S5 are as follows: S51. During the actual operation of the photovoltaic storage system, the state at each time step t will be recorded. ,action Instant rewards And the next state Experience data is stored as a single data point in the experience replay pool D, with reward values ​​placed there first. Experienced samples with high or large state changes; Where 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 assessment network Critic: in, The current parameters of the Critic value assessment network. The target parameters for the Critic value assessment network; These are the current parameters of the policy network Actor. For the target parameters of the policy network Actor, Here, N is the discount factor, and N is the number of empirical samples sampled in the empirical replay pool. It is based on a multi-objective function Calculated; S53. Gradient direction update strategy for the loss function of the value evaluation network Critic: Network Actor: in, This represents calculating the gradient with respect to the policy network parameters θ. It is the objective function of the policy network Actor, where E is the mathematical expectation, i.e., the average operation of the batch. It is about finding the gradient of action a. This represents the state-action value function output by the Critic value assessment network, i.e., the state at which the network is in action. Take action at the time The long-term earnings forecast, The policy network Actor network represents the state. The action output at that time; S54. Obtain the actual operating data of the photovoltaic storage system, the actual operating data including the state vector. and actual fault labels and the state vector The predicted fault labels are obtained by inputting them into the trained fault prediction model. A cross-entropy loss function is then constructed to measure the difference between the predicted fault labels and the actual fault labels. Where C represents the total number of fault types. The encoding of the actual fault label. This represents the predicted probability of type c faults. S55. The model parameters of the fault prediction model are updated based on the cross-entropy loss function and the backpropagation algorithm.

5. The intelligent scheduling and fault prediction method for a photovoltaic-storage system according to claim 4, characterized in that, The specific steps of step S55 are as follows: S551. Take the derivative of 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.

6. A smart scheduling and fault prediction device for a photovoltaic-storage system, comprising executing the method described in any one of claims 1-5, characterized in that, The photovoltaic-storage system includes a photovoltaic array, an energy storage battery pack, an energy storage converter, and a grid connection interface. 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 multi-modal sensor array deployed on photovoltaic arrays, energy storage battery packs, and energy storage converters, and to preprocess the collected data. The data fusion module is used to input preprocessed data into a Transformer-based multimodal fusion model, and obtain fused features by performing unified feature representation through a cross-attention mechanism. The fault and lifetime prediction module is used to input the fused features into the prediction model to perform fault prediction, including fault type and fault probability, for photovoltaic arrays, energy storage battery packs and energy storage converters, and to predict the remaining lifetime for each. The intelligent scheduling module is used to generate a state space consisting of fault type, fault probability, and real-time electricity price through a reinforcement learning algorithm, and a scheduling strategy with the charging and discharging power of the energy storage battery pack as the action space. It also sets a reward function based on the equipment fault penalty of the photovoltaic-storage system, the SOC health of the energy storage battery pack, and the curtailment penalty of photovoltaic power generation as objectives. The learning optimization module is used to construct an experience replay pool based on the actual operating data of the optical storage system, and to periodically 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 step, and the instantaneous reward value of the reward function in the previous time step.

7. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent scheduling and fault prediction method for a photovoltaic storage system as described in any one of claims 1 to 5.

8. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent scheduling and fault prediction method for the photovoltaic storage system as described in any one of claims 1 to 5.

Images