Energy storage system self-healing and early warning method, equipment and medium

Abstract

The invention discloses an energy storage system self-healing and early warning method and device and a medium, and relates to the technical field. The method comprises the steps of collecting multi-dimensional operation data through a sensor network deployed in a layered manner; carrying out normalization and adaptive discretization on the multi-dimensional operation data by adopting a Bayesian network, and carrying out feature extraction by utilizing Kalman filtering to obtain enhanced feature data adaptive to a time sequence; the enhanced feature data is calculated through a collaborative model of a convolutional neural network and a Transform, and a fault is confirmed through a deep learning model; for the fault, constructing a state space, an action space and a reward function, and selecting an optimal strategy according to a reward value; and after the optimal strategy is executed, predicting a predicted value of a future preset time step through an autoregressive integral moving average model, and calculating a threshold value by combining a historical data mean value and a standard deviation in a time window. According to the method, real-time monitoring, fault prediction and performance optimization of the energy storage system are realized.

Description

Technical Field

[0001] The present application relates to the field of energy storage operation and maintenance technology, and in particular to a method, device, and medium for self-healing and early warning of an energy storage system. Background Art

[0002] With the global energy transition and the rapid development of new energy technologies, energy storage systems are being widely used in areas such as renewable energy and grid peak regulation. The operational stability and safety of these systems have become particularly important. Traditional energy storage system fault handling relies primarily on manual inspections and preset threshold alarms. This approach suffers from delayed response, high misjudgment rates, and insufficient self-healing capabilities. Furthermore, existing technologies lack the ability to deeply mine data from the entire lifecycle of energy storage systems, making it difficult to achieve early warning and automated repair of faults. This results in prolonged system downtime and high maintenance costs.

[0003] Furthermore, as the scale of power grids and energy storage systems expands, traditional fault diagnosis and troubleshooting methods are no longer able to meet the demands of efficient and reliable operation. Existing energy storage system monitoring and troubleshooting methods often neglect real-time monitoring of the energy storage device's corresponding converters and regional environmental factors. This can affect the energy storage and release rates of the energy storage system, thereby impacting the system's response speed and performance.

[0004] Through the above analysis, the problems and defects of the existing technology are as follows: Existing technologies suffer from insufficient data utilization and insufficient intelligent diagnosis, resulting in delayed fault response, high misjudgment rate, and lack of adaptive early warning and self-healing capabilities. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, and medium for self-healing and early warning of an energy storage system, which can solve the problems in the prior art caused by insufficient data utilization and insufficient diagnostic intelligence, such as delayed fault response, high misjudgment rate, and lack of adaptive early warning and self-healing capabilities.

[0006] In the first aspect, an embodiment of the present application provides a method for self-healing and early warning of an energy storage system, the method comprising: collecting multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system through a hierarchically deployed sensor network; normalizing and adaptively discretizing the multi-dimensional operating data using a Bayesian network, and extracting features using a Kalman filter to obtain enhanced feature data adapted to the time series; calculating the enhanced feature data through a collaborative model of a convolutional neural network and a Transformer, and confirming the fault through a deep learning model; constructing a state space, an action space, and a reward function for the fault, and selecting the optimal strategy based on the reward value; after executing the optimal strategy, predicting the predicted value of a preset time step in the future through an autoregressive integral sliding average model, and calculating the threshold by combining the mean and standard deviation of historical data in the time window; triggering a graded warning when the predicted value exceeds the threshold, and adjusting the preset operating strategy.

[0007] In one implementation of the present application, multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system are collected through a layered sensor network, specifically including: deploying the first-layer sensor network to the internal core components of the energy storage system to collect voltage, current, temperature, SOC, and SOH. The internal core components include single cells, battery modules, and battery management systems; deploying the second-layer sensor network to the external environment to collect ambient temperature and humidity. The external environment includes the cabinet surface; deploying the third-layer sensor network to key auxiliary equipment to collect internal temperature, current, and conversion efficiency. Key auxiliary equipment includes a power conversion system; deploying the fourth-layer sensor network to the surrounding environment at a preset distance to collect smoke concentration and combustible gas concentration.

[0008] In one implementation of the present application, a Bayesian network is used to normalize and adaptively discretize multi-dimensional operation data, and Kalman filtering is used for feature extraction to obtain enhanced feature data that adapts to the time series, specifically including: establishing a multivariate probability model of dimensional operation data, calculating the multivariate posterior probability through the joint probability distribution of known data, identifying and calibrating abnormal noise data; extracting operation feature data through state prediction, covariance prediction, Kalman gain calculation, and the iterative process of state update and covariance update.

[0009] In one implementation of the present application, the method further includes: calculating the fluctuation variance of the operating characteristic data; when the fluctuation variance is greater than the variance threshold, reducing the discretization interval width; when the variance is less than the variance threshold, expanding the interval width.

[0010] In one implementation of the present application, enhanced feature data is calculated through a collaborative model of a convolutional neural network and a Transformer, and faults are confirmed through a deep learning model, specifically including: inputting enhanced feature data, performing convolution operations through convolution kernels, and extracting local features; capturing long-term features of the data through a multi-head attention mechanism; inputting local features and long-term features into a fully connected layer, training through a cross-entropy loss function, and outputting fault confidence; and confirming the fault when the fault confidence exceeds the confidence value.

[0011] In one implementation of the present application, a state space, an action space, and a reward function are constructed for a fault to generate an optimal self-healing strategy, specifically including: when a fault is confirmed, selecting a preset number of candidate actions, the candidate actions including starting the cooling system and switching redundant battery modules; collecting multi-dimensional operating data in real time after execution, and calculating the reward value of the candidate action; and adjusting the priority of the action in the strategy library according to the reward value.

[0012] In one implementation of the present application, an autoregressive integral sliding average model is used to predict the predicted value of a preset time step in the future, and the threshold is calculated in combination with the mean and standard deviation of historical data in the time window. Specifically, the following steps are performed: order difference processing is performed on the time series data to predict the parameter value of the preset time step in the future, and the parameter value includes the battery cell temperature; the standard deviation threshold is calculated by calculating the mean and standard deviation of the data in the time window based on the current operating conditions, and the operating conditions include fast charging conditions, static conditions, and discharge conditions.

[0013] In one implementation of the present application, the method also includes: based on the mean square error, using the stochastic gradient descent algorithm with adaptive learning rate to update the autoregressive integral moving average model and the deep learning model parameters; according to the feedback on the strategy execution effect, adjusting the noise recognition threshold of the Bayesian network and the process noise covariance of the Kalman filter.

[0014] In a second aspect, an embodiment of the present application further provides a device for self-healing and early warning of an energy storage system, the device comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to: collect multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system through a hierarchically deployed sensor network; normalize and adaptively discretize the multi-dimensional operating data using a Bayesian network, and extract features using a Kalman filter to obtain enhanced feature data adapted to the time series; calculate the enhanced feature data through a collaborative model of a convolutional neural network and a Transformer, and confirm the fault through a deep learning model; for the fault, construct a state space, an action space, and a reward function, and select the optimal strategy based on the reward value; after executing the optimal strategy, predict the predicted value of a preset time step in the future through an autoregressive integral sliding average model, and calculate the threshold value by combining the mean and standard deviation of the historical data in the time window; trigger a graded warning when the predicted value exceeds the threshold, and adjust the preset operating strategy.

[0015] On the third aspect, the embodiment of the present application also provides a non-volatile computer storage medium for self-healing and early warning of an energy storage system, which stores computer executable instructions, and the computer executable instructions are set to: collect multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system through a hierarchically deployed sensor network; use a Bayesian network to normalize and adaptively discretize the multi-dimensional operating data, and use Kalman filtering to extract features to obtain enhanced feature data that adapts to the time series; calculate the enhanced feature data through a collaborative model of convolutional neural network and Transformer, and confirm the fault through a deep learning model; for the fault, construct a state space, action space and reward function, and select the optimal strategy according to the reward value; after executing the optimal strategy, predict the predicted value of the future preset time step through the autoregressive integral sliding average model, and calculate the threshold by combining the mean and standard deviation of historical data in the time window; trigger a graded warning when the predicted value exceeds the threshold, and adjust the preset operating strategy.

[0016] The embodiments of the present application provide a method, device, and medium for self-healing and early warning of an energy storage system. The method realizes collaborative collection of multi-dimensional data through a layered sensor network, improves data quality by combining Bayesian networks and Kalman filtering, and accurately identifies faults using a CNN-Transformer collaborative model. The method dynamically generates self-healing strategies based on reinforcement learning, and realizes graded early warning through an ARIMA model and dynamic thresholds. The method integrates multi-dimensional data with time series predictions to identify hidden dangers such as battery thermal runaway and SOC deviation in advance. The method drives the generation of self-healing strategies through a reward function, and automatically performs redundant switching, power adjustment, and other operations. The threshold is adaptively adjusted based on the operating conditions to reduce the false alarm rate and trigger a graded response. The model parameters and strategy library are optimized based on feedback to improve the robustness of the system in different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flow chart of a method for self-healing and early warning of an energy storage system provided in an embodiment of the present application; Figure 2 A schematic diagram of the internal structure of a device for self-healing and early warning of an energy storage system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] The embodiments of the present application provide a method, device, and medium for self-healing and early warning of an energy storage system, which solve the problems in the prior art such as delayed fault response, high misjudgment rate, and lack of adaptive early warning and self-healing capabilities caused by insufficient data utilization and insufficient diagnostic intelligence.

[0020] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0021] Figure 1 This is a flow chart of a method for self-healing and early warning of an energy storage system provided in an embodiment of the present application. Figure 1 As shown, the embodiment of the present application provides a method for energy storage system self-healing and early warning, which specifically includes the following steps: Step 10: Collect multi-dimensional operational data from the energy storage system’s core components, external environment, and key auxiliary equipment through a layered sensor network. In this step, a sensor network covering all key parts of the energy storage system and its surrounding environment is built to enable real-time acquisition of internal and external environmental data. Internal operating data includes voltage, current, temperature, SOC (State of Charge), SOH (State of Health), and other data that directly reflect the operating status of the energy storage system's core components. External environmental data includes ambient temperature and humidity, which significantly impact the performance and stability of the energy storage system. Comprehensive collection of this data is essential for accurately assessing the system's operating status.

[0022] As an optional embodiment, a layered sensor network is deployed to collect multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system. Specifically, the following may be included: Step 101: Deploy the first-layer sensor network to the internal core components of the energy storage system to collect voltage, current, temperature, SOC, and SOH. The internal core components include single cells, battery modules, and battery management systems; Step 102: Deploy the second-layer sensor network to the external environment to collect ambient temperature and humidity. The external environment includes the cabinet surface; Step 103: Deploy the third-layer sensor network to key auxiliary equipment to collect internal temperature, current, and conversion efficiency. Key auxiliary equipment includes the power conversion system; Step 104: Deploy the fourth-layer sensor network to the surrounding environment at a preset distance to collect smoke concentration and combustible gas concentration.

[0023] In this step, the data types collected cover all aspects that affect the safe and stable operation of the energy storage system. Core operating data within the energy storage system includes the voltage, current, and temperature of individual cells / modules, as well as key parameters such as the state of charge and health reported by the battery management system (BMS). This data directly reflects the real-time operating condition and aging of the energy storage units and serves as a key basis for fault diagnosis and lifespan prediction. External environmental data, such as the ambient temperature and humidity at the energy storage device's location, must be monitored. Extreme environmental conditions can significantly affect battery performance and lifespan, and even pose safety risks, and therefore must be included in the monitoring scope. Key auxiliary component data focuses on monitoring the operating status of the power conversion system (PCS), which is closely related to energy conversion in the energy storage device, such as its internal temperature, current, voltage, and conversion efficiency. A PCS failure can also cause the entire energy storage system to fail. Regional environmental monitoring data, for large energy storage power plants, may also include safety-related environmental monitoring data such as smoke and combustible gas concentrations within the area.

[0024] Step 20: Use the Bayesian network to normalize and adaptively discretize the multi-dimensional operation data, and use the Kalman filter to extract features to obtain enhanced feature data that adapts to the time series; As an optional embodiment, a Bayesian network is used to normalize and adaptively discretize multi-dimensional operation data, and Kalman filtering is used to extract features to obtain enhanced feature data that adapts to the time series. Specifically, it may include: Step 201: Establish a multivariate probability model for dimensional operation data, calculate the multivariate posterior probability through the joint probability distribution of known data, and identify and calibrate abnormal noise data; Step 202: Extract operation feature data through state prediction, covariance prediction, Kalman gain calculation, and the iterative process of state update and covariance update.

[0025] In this step, the original collected data often contains noise, outliers or certain measurement errors. Direct use will affect the accuracy of the subsequent analysis model. The Bayesian network is used to reduce noise and calibrate the collected data. In the Bayesian network, based on the Bayesian theorem, for multiple random variables , its joint probability distribution can be expressed as:

[0026] By calculating the probabilistic relationship between known data, the posterior probability of each variable is used to identify and remove noise interference in the data. At the same time, the data is calibrated to ensure its accuracy and consistency.

[0027] Kalman filtering is used to extract features from data. The core formula of Kalman filtering is as follows: Status prediction:

[0028] Covariance prediction:

[0029] Update steps: Kalman gain:

[0030] Status Update:

[0031] Covariance update:

[0032] in, Denotes the state estimate, P denotes the covariance, F is the state transition matrix, B is the control input matrix, u is the control input, H is the observation matrix, z is the observation value, Q is the process noise covariance, R is the observation noise covariance, and K is the Kalman gain. By continuously iterating the above process, the data information that best reflects the operating characteristics of the energy storage system in the presence of noise is extracted, providing high-quality data support for subsequent analysis.

[0033] Normalize the collected data to the range [0,1]. Use the minimum-maximum normalization method, and the formula is:

[0034] Among them, x is the original data, and are the minimum and maximum values ​​in the data, respectively. The data is normalized. This method can eliminate the dimensional differences between different data dimensions, making the data comparable. It also helps improve the training efficiency and accuracy of subsequent deep learning models and avoid model training bias caused by different data scales.

[0035] Discretize the data and discretize the continuous data into time series data. The common equal-width discretization method divides the data range into several equal-width intervals. Assuming that the data range is [a, b] and is divided into n intervals, the width of each interval is ,By judging the interval where the data is located, the continuous data is mapped to the corresponding ,discrete interval, which facilitates the subsequent use of time series analysis ,methods to monitor and predict the operating status of the energy storage system.

[0036] As an optional embodiment, the method may further include: step 203: calculating the fluctuation variance of the operating characteristic data; step 204: reducing the discretization interval width when the fluctuation variance is greater than the variance threshold; step 205: expanding the interval width when the variance is less than the variance threshold.

[0037] Step 30: Calculate the enhanced feature data using a collaborative model of a convolutional neural network and a Transformer, and confirm the fault using a deep learning model. As an optional embodiment, the enhanced feature data is calculated through a collaborative model of a convolutional neural network and a Transformer, and the fault is confirmed through a deep learning model. Specifically, the following may be included: Step 301: inputting the enhanced feature data, performing a convolution operation through a convolution kernel, and extracting local features; Step 302: capturing the long-term features of the data through a multi-head attention mechanism; Step 303: inputting the local features and the long-term features into a fully connected layer, training through a cross-entropy loss function, and outputting the fault confidence; Step 304: confirming the fault when the fault confidence exceeds the confidence value.

[0038] In this step, an energy storage system health status assessment model is constructed based on a deep learning model called a convolutional neural network (CNN). In a CNN, the convolution layer extracts features by convolving the input data with a convolution kernel. Assuming the input data is X and the convolution kernel is K, the convolution operation formula is:

[0039] in, is the value of the convolution result at position (i, j), and M and N are the sizes of the convolution kernel. Through structures such as convolutional and pooling layers, local and abstract features in energy storage system operating data can be automatically extracted, enabling a comprehensive assessment of the system's health. The model is trained using historical data to learn data characteristic patterns under normal operating conditions and various fault conditions.

[0040] Combining real-time data with historical trends, the Transformer time series prediction model is used to predict the changing trends of key parameters. The formula for the multi-head attention mechanism in the Transformer model is:

[0041] Among them, Q (Query), K (Key), and V (Value) are input vectors. is the scaling factor. Through its multi-head attention mechanism, the Transformer effectively processes long sequences of data and captures long-term dependencies within the data, enabling accurate prediction of future changes in key parameters such as voltage, current, temperature, and SOC. By comparing the predicted results with historical data and data features from normal operation, potential faults such as battery thermal runaway, SOC estimation deviation, and circuit aging can be identified.

[0042] The fault is secondary confirmed and diagnosed based on the deep learning model. After the potential fault is initially identified, the deep learning model is used again to conduct in-depth analysis and judgment of the fault to ensure the accuracy of the fault diagnosis results and avoid misjudgment and missed judgment.

[0043] Step 40: Construct the state space, action space, and reward function for the fault, and select the optimal strategy based on the reward value; As an optional embodiment, for faults, a state space, an action space, and a reward function are constructed to generate an optimal self-healing strategy, which may specifically include: Step 401: When a fault is confirmed, a preset number of candidate actions are selected, and the candidate actions include starting the cooling system and switching redundant battery modules; Step 402: After execution, multi-dimensional operating data is collected in real time and the reward value of the candidate action is calculated; Step 403: The priority of the action in the strategy library is adjusted according to the reward value.

[0044] For different fault types, a self-healing strategy is generated using a reinforcement learning algorithm. In reinforcement learning, an agent interacts with the environment. When it performs an action a in state s, it receives a reward r and moves to a new state s′. The core formula is the Bellman equation:

[0045] in, is the state-action value function, and γ is the discount factor. By allowing the agent to continuously learn through trial and error in the environment, the strategy is optimized based on the reward signal fed back by the environment. This allows the agent to generate optimal self-healing strategies for different situations, such as battery module failure or power conversion unit failure, such as automatically switching redundant battery modules or power conversion units, dynamically adjusting charge and discharge power or SOC thresholds, and activating internal balancing circuits or cooling systems.

[0046] Step 50: After executing the optimal strategy, the predicted value for the preset time step in the future is predicted using the autoregressive integrated moving average model, and the threshold is calculated by combining the mean and standard deviation of the historical data in the time window; In this step, based on real-time data and historical trends, the ARIMA (Autoregressive Integrated Moving Average) time series forecasting model is used to predict the changing trends of key parameters. The formula of the ARIMA (p, d, q) model is:

[0047] in, is a d-order difference operator, is the original time series, is the autoregressive coefficient, is the moving average coefficient, By fitting and analyzing historical data, we can predict the changes of key parameters in the future and discover potential failure risks in advance.

[0048] As an optional embodiment, the predicted value of the future preset time step is predicted by an autoregressive integral moving average model, and the threshold is calculated by combining the mean and standard deviation of the historical data in the time window. Specifically, it may include: Step 501: performing order difference processing on the time series data to predict the parameter value of the future preset time step, and the parameter value includes the battery cell temperature; Step 502: calculating the standard deviation threshold by calculating the mean and standard deviation of the data in the time window based on the current operating conditions, and the operating conditions include fast charging conditions, static conditions and discharge conditions.

[0049] In this step, the warning level is set based on the dynamic time window dynamic threshold algorithm. Assume that the time window size is w, at time t, the dynamic threshold , can be calculated by the following formula:

[0050] in, is the data within the time window, is the standard deviation of the data within the time window, and k is the adjustment coefficient. This algorithm dynamically adjusts the warning threshold based on recent data fluctuations and trends, ensuring that the warning level more closely matches the actual operating status of the energy storage system. When data fluctuates significantly, the warning sensitivity is appropriately increased; when data is relatively stable, the warning threshold is appropriately relaxed to avoid unnecessary false alarms.

[0051] When parameters approach abnormal thresholds, an alert is issued via SMS or app push notifications, and the higher-level control system is linked to adjust operational strategies. This alert is promptly sent to relevant operations and maintenance personnel, allowing them to immediately understand any abnormalities in the energy storage system. Simultaneously, the system automatically adjusts the energy storage system's operating parameters and operating mode based on the alert, such as reducing charge and discharge power or switching operating modes, to prevent further escalation of the fault.

[0052] Step 60: When the predicted value exceeds the threshold, a graded warning is triggered and the preset operation strategy is adjusted.

[0053] In this step, we use a dynamic threshold algorithm to set warning levels. Based on the ARIMA model's predictions and the threshold ranges determined by the dynamic threshold algorithm, we assign different warning levels, such as general warning and severe warning, so that operations personnel can take appropriate countermeasures based on the warning level.

[0054] When a parameter exceeds the warning threshold, an early warning is issued via SMS, email, etc. Early warning information is delivered to relevant personnel in a timely and multi-channel manner, ensuring that the information can be conveyed accurately and quickly, allowing operation and maintenance personnel to respond in a timely manner.

[0055] As an optional embodiment, the method may also include: updating the autoregressive integral moving average model and deep learning model parameters based on the mean square error and using the stochastic gradient descent algorithm with adaptive learning rate; adjusting the noise recognition threshold of the Bayesian network and the process noise covariance of the Kalman filter according to the feedback of the strategy execution effect.

[0056] In this step, the self-healing strategy library and early warning model are optimized based on feedback data. Based on feedback on strategy execution, the effectiveness of the self-healing strategies and the accuracy of the early warning model are analyzed. Effective self-healing strategies are retained and strengthened, while those with poor performance are optimized and improved. Furthermore, the parameters and structure of the early warning model are adjusted to improve its prediction accuracy and reliability. Optimize the self-healing strategy library and early warning model based on feedback data. In order to measure the relationship between feedback data and strategy effectiveness, a strategy effectiveness evaluation formula is introduced. Let the strategy effectiveness score be SEA, and the fault repair rate weight be , the system recovery weight is ,in ,but

[0057] This formula combines the fault repair rate and system recovery, assigning different weights to comprehensively evaluate the effectiveness of self-healing strategies. Based on the calculated strategy effectiveness score, strategies in the self-healing strategy library can be optimized. Policies with low scores can be improved by adjusting their parameters or structure. For early warning model optimization, let the early warning model prediction value be , the actual value is , the number of samples is N, and the mean square error (MSE) is used as the loss function L to evaluate the accuracy of the model prediction. The formula is:

[0058] Based on this loss function, the stochastic gradient descent method is used to update the early warning model parameters θ. The update formula is:

[0059] in, The parameter value α for the kth iteration is the learning rate. Through continuous iteration, the early warning model parameters are adjusted according to the feedback data, the loss function value is reduced, and the early warning model is optimized to improve its prediction accuracy and reliability.

[0060] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides a device for self-healing and early warning of an energy storage system, the structure of which is as follows: Figure 2 shown.

[0061] Figure 2This is a schematic diagram of the internal structure of a device for self-healing and early warning of an energy storage system provided in an embodiment of the present application. Figure 2 As shown, the equipment includes: at least one processor 201; and, a memory 202 communicatively coupled to the at least one processor; Among them, the memory 202 stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor 201 so that the at least one processor 201 can: collect multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system through a hierarchically deployed sensor network; use a Bayesian network to normalize and adaptively discretize the multi-dimensional operating data, and use Kalman filtering to extract features to obtain enhanced feature data that adapts to the time series; calculate the enhanced feature data through a collaborative model of convolutional neural network and Transformer, and confirm the fault through a deep learning model; for the fault, construct a state space, action space and reward function, and select the optimal strategy according to the reward value; after executing the optimal strategy, predict the predicted value of the future preset time step through the autoregressive integral sliding average model, and calculate the threshold by combining the mean and standard deviation of historical data in the time window; trigger a graded warning when the predicted value exceeds the threshold, and adjust the preset operating strategy.

[0062] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for self-healing and early warning of an energy storage system stores computer-executable instructions, wherein the computer-executable instructions are configured to: collect multi-dimensional operating data of the internal core components, external environment, and key auxiliary equipment of the energy storage system through a hierarchically deployed sensor network; normalize and adaptively discretize the multi-dimensional operating data using a Bayesian network, and extract features using a Kalman filter to obtain enhanced feature data adapted to the time series; calculate the enhanced feature data through a collaborative model of a convolutional neural network and a Transformer, and confirm the fault through a deep learning model; construct a state space, an action space, and a reward function for the fault, and select the optimal strategy based on the reward value; after executing the optimal strategy, predict the predicted value of a preset time step in the future through an autoregressive integral sliding average model, and calculate the threshold value by combining the mean and standard deviation of historical data in the time window; trigger a graded warning when the predicted value exceeds the threshold, and adjust the preset operating strategy.

[0063] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the IoT device and media embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.

[0064] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0065] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0069] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0070] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0071] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0072] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0073] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for self-healing and early warning of an energy storage system, characterized in that: The method comprises: Through a layered sensor network, multi-dimensional operating data of the energy storage system's internal core components, external environment, and key auxiliary equipment is collected; A Bayesian network is used to normalize and adaptively discretize the multi-dimensional operation data, and a Kalman filter is used to extract features to obtain enhanced feature data that is adapted to the time series; The enhanced feature data is calculated using a collaborative model of a convolutional neural network and a Transformer, and the fault is confirmed using a deep learning model; For the fault, construct the state space, action space and reward function, and select the optimal strategy based on the reward value; After executing the optimal strategy, the predicted value of the future preset time step is predicted through the autoregressive integrated moving average model, and the threshold is calculated by combining the mean and standard deviation of the historical data in the time window; When the predicted value exceeds a threshold, a graded warning is triggered and a preset operation strategy is adjusted.

2. The method for self-healing and early warning of an energy storage system according to claim 1, characterized in that: The hierarchically deployed sensor network collects multi-dimensional operational data on the energy storage system's internal core components, external environment, and key auxiliary equipment, specifically including: Deploy the first-tier sensor network to the core components of the energy storage system to collect voltage, current, temperature, SOC, and SOH. The core components include single cells, battery modules, and the battery management system. Deploy the second layer of sensor network to the external environment to collect ambient temperature and humidity, wherein the external environment includes the cabinet surface; Deploy the third layer of sensor networks to key auxiliary equipment, including power conversion systems, to collect internal temperature, current, and conversion efficiency; The fourth-layer sensor network is deployed to the surrounding environment at a preset distance to collect smoke concentration and combustible gas concentration.

3. The energy storage system self-healing and early warning method according to claim 2, characterized in that: The multi-dimensional operation data is normalized and adaptively discretized using a Bayesian network, and features are extracted using a Kalman filter to obtain enhanced feature data that adapts to the time series, specifically including: Establish a multivariate probability model for the dimensional running data, calculate the multivariate posterior probability through the joint probability distribution of known data, and identify and calibrate abnormal noise data; Operational characteristic data is extracted through the iterative process of state prediction, covariance prediction, Kalman gain calculation, and state update and covariance update.

4. The energy storage system self-healing and early warning method according to claim 1, characterized in that: The method further comprises: Calculating the fluctuation variance of the operation characteristic data; When the fluctuation variance is greater than the variance threshold, narrowing the discretization interval width; When the variance is less than the variance threshold, the interval width is expanded.

5. The energy storage system self-healing and early warning method according to claim 1, characterized in that: The enhanced feature data is calculated using a collaborative model of a convolutional neural network and a Transformer, and the fault is confirmed using a deep learning model, specifically including: Input the enhanced feature data, perform convolution operation through the convolution kernel, and extract local features; Capture long-term data features through multi-head attention mechanism; Input the local features and long-term features into a fully connected layer, train them through a cross entropy loss function, and output fault confidence; The fault is confirmed when the fault confidence exceeds a confidence value.

6. The energy storage system self-healing and early warning method according to claim 1, characterized in that: For the fault, the state space, action space, and reward function are constructed to generate the optimal self-healing strategy, which specifically includes: When a fault is confirmed, a preset number of candidate actions are selected, wherein the candidate actions include starting a cooling system and switching a redundant battery module; After execution, multi-dimensional operation data is collected in real time to calculate the reward value of the candidate action; The priority of actions in the policy library is adjusted according to the reward value.

7. The energy storage system self-healing and early warning method according to claim 1, characterized in that: The autoregressive integrated moving average model is used to predict the predicted value of the preset time step in the future. The threshold is calculated by combining the mean and standard deviation of the historical data in the time window. Specifically, it includes: Performing order difference processing on the time series data to predict parameter values ​​at future preset time steps, the parameter values ​​including battery cell temperature; The standard deviation threshold is calculated by calculating the mean and standard deviation of the data in the time window based on the current operating conditions, wherein the operating conditions include fast charging conditions, static conditions, and discharging conditions.

8. The energy storage system self-healing and early warning method according to claim 1, characterized in that: The method further comprises: Based on the mean square error, the stochastic gradient descent algorithm with adaptive learning rate is used to update the parameters of the autoregressive integrated moving average model and the deep learning model; According to the feedback of strategy execution effect, the noise recognition threshold of Bayesian network and the process noise covariance of Kalman filter are adjusted.

9. A device for self-healing and early warning of energy storage system, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Through a layered sensor network, multi-dimensional operating data of the energy storage system's internal core components, external environment, and key auxiliary equipment is collected; Normalizing and adaptively discretizing the multi-dimensional operating data to obtain enhanced feature data adapted to the time series; The enhanced feature data is calculated using a collaborative model of a convolutional neural network and a Transformer, and the fault is confirmed using a deep learning model; Constructing the state space, action space, and reward function for the fault, and generating the optimal self-healing strategy; The predicted value of the preset time step in the future is predicted through the autoregressive integrated moving average model, and the threshold is calculated by combining the mean and standard deviation of the historical data in the time window; When the predicted value approaches a threshold, a graded warning is triggered and a preset operation strategy is adjusted.

10. A non-volatile computer storage medium for energy storage system self-healing and early warning, storing computer executable instructions, characterized in that: The computer executable instructions are configured to: Through a layered sensor network, multi-dimensional operating data of the energy storage system's internal core components, external environment, and key auxiliary equipment is collected; Normalizing and adaptively discretizing the multi-dimensional operating data to obtain enhanced feature data adapted to the time series; The enhanced feature data is calculated using a collaborative model of a convolutional neural network and a Transformer, and the fault is confirmed using a deep learning model; Constructing the state space, action space, and reward function for the fault, and generating the optimal self-healing strategy; The predicted value of the preset time step in the future is predicted through the autoregressive integrated moving average model, and the threshold is calculated by combining the mean and standard deviation of the historical data in the time window; When the predicted value approaches a threshold, a graded warning is triggered and a preset operation strategy is adjusted.

Images