Energy storage power station operation and maintenance method, device and equipment based on data driving

Abstract

The invention relates to the technical field of energy storage power generation, and discloses an energy storage power station operation and maintenance method, device and equipment based on data driving, and the method comprises the steps: carrying out the preprocessing of historical feature data, and carrying out the dimension reduction processing, and obtaining sensitive dimension reduction feature data; normal feature data are screened out to train an automatic encoder, whether the data are fault data is detected, and type analysis is performed on the fault data; constructing an equivalent circuit model of the lithium battery, and carrying out a pulse charging and discharging experiment to obtain parameters under different residual electric quantities; and simulating charge and discharge characteristic curves under different fault types based on an equivalent circuit model, extracting fault parameters, performing inversion to obtain real fault parameters, and taking corresponding operation and maintenance measures. The closed-loop intelligent operation and maintenance system is constructed based on a large amount of historical feature data, the fault detection result is fed back timely and accurately, various faults of the energy storage system are effectively monitored, the diagnosis capability and the intelligent level of series faults are improved, and targeted operation and maintenance measures are taken timely.

Description

Technical Field

[0001] The present invention relates to the field of energy storage and power generation technology, and in particular to a data-driven energy storage power station operation and maintenance method, device and equipment. Background Art

[0002] Lithium-ion battery energy storage power stations are energy facilities that store electrical energy in the form of chemical energy in lithium-ion battery packs and release it when needed. They play an increasingly important role in the modern energy system, particularly in areas such as renewable energy integration, grid stabilization, and peak-valley regulation. As global demand for renewable energy increases, the installed capacity of lithium-ion battery energy storage power stations, which can smooth out fluctuations in renewable energy, is also expanding. Typically, a lithium-ion battery energy storage power station comprises an energy storage unit, power conversion system, battery management system, energy management system, thermal management system, auxiliary equipment, and other components, forming a highly integrated and complex electromechanical system. The long-term, efficient, and stable operation of a power station requires timely and efficient system monitoring and maintenance. Accurate monitoring and diagnosis of lithium-ion batteries is a key component of system maintenance.

[0003] In actual operation and maintenance, common lithium battery failures include internal short circuits / micro-short circuits, inconsistencies, connection failures, overcharging, over-discharging, uneven aging, and open circuits, among other soft and hard faults. These failures can reduce the efficiency and stability of energy storage power plants, shorten the lifespan of the power plant, and ultimately lead to failure to meet expected power plant revenue targets. Therefore, the power plant's battery management system and operation and maintenance system prioritize fault diagnosis. Due to the wide variety of faults in energy storage power plants and the large number of individual cells in their energy storage units, the batteries continuously generate massive amounts of operational data during power plant operation. Detecting early-stage faults and micro-faults from this data is a significant challenge. Furthermore, energy storage power plants, often built alongside photovoltaic or wind farms, typically cover large areas and operate in harsh environments. Traditional manual inspection methods struggle to detect problems promptly and comprehensively, and are costly to detect.

[0004] At present, the operation and maintenance technology of lithium battery energy storage power stations mainly focuses on: (1) relying on battery management systems and information collection systems for monitoring. This operation and maintenance technology can provide overall monitoring of the overall operation data of the power station and provide a relatively simple fault alarm function. By setting safety thresholds for the collected values ​​and detection values, an alarm will be issued if the threshold is exceeded or lowered, but it is difficult to perform refined fault diagnosis; (2) analyzing the charge and discharge curves of the lithium battery during the working process. Based on statistical data, the charge and discharge curves during actual operation are compared and statistically analyzed with the charge and discharge curves of normally working batteries to obtain information about the faulty battery and then diagnose the fault. However, this method can only support energy storage systems with a small number of lithium batteries and is often used for offline diagnosis. For large-scale energy storage systems, the huge amount of data will cause the difficulty of diagnosis to increase exponentially. Traditional fault diagnosis models attempt to identify faults through modeling and simulation, but such methods rely on theoretical assumptions, have poor flexibility and adaptability, and have limited ability to analyze complex faults.

[0005] Therefore, existing energy storage power station operation and maintenance methods have many shortcomings: First, the real-time and time consistency of existing data acquisition and monitoring systems often have problems, which may lag behind the actual occurrence of faults, making it difficult to detect problems in a timely manner; second, fault detection and classification mainly rely on expert experience and threshold judgment, and the ability to respond to complex faults is weak. Reasonable threshold setting lacks scientific basis, resulting in a high false alarm rate of faults in the power station management system; and the massive historical data and real-time operation data generated during the operation of energy storage power stations are not fully utilized. Summary of the Invention

[0006] In view of this, the present invention provides a data-driven energy storage power station operation and maintenance method, device and equipment to solve the problems of fault detection lag, high false alarm rate and inability to fully utilize large amounts of data.

[0007] In a first aspect, the present invention provides a data-driven operation and maintenance method for an energy storage power station, wherein the energy storage power station includes a plurality of lithium batteries. The method includes:

[0008] Obtain historical characteristic data of the energy storage power station, pre-process the historical characteristic data, and use a preset algorithm to filter sensitive characteristic data from the historical characteristic data, perform dimensionality reduction processing on the sensitive characteristic data, and obtain sensitive dimensionality reduction characteristic data;

[0009] Normal feature data is filtered out from the sensitive dimensionality reduction feature data, and an autoencoder is trained using the normal feature data. The trained autoencoder is then used to detect whether the actual operating data of the energy storage power station is fault data. The fault data is then type-analyzed to determine the corresponding fault type.

[0010] Construct an equivalent circuit model of the lithium battery and conduct pulse charge and discharge experiments at different residual capacities to obtain the parameters of the equivalent circuit model at different residual capacities.

[0011] Based on the equivalent circuit model, the charge and discharge characteristic curves of lithium batteries under different fault types are simulated. The fault parameters are extracted according to the charge and discharge characteristic curves, and the gradient descent algorithm is used for inversion to obtain the true fault parameters. Corresponding operation and maintenance measures are taken based on the true fault parameters.

[0012] The data-driven energy storage power station operation and maintenance method provided by the present invention builds a closed-loop intelligent operation and maintenance system based on a large amount of historical feature data, provides timely and accurate feedback on fault detection results, adapts to the needs of actual application scenarios, effectively monitors various faults in the energy storage system, improves the diagnostic capability and intelligence level of a series of faults, takes targeted operation and maintenance measures in a timely manner, optimizes operation and maintenance strategies, reduces operation and maintenance costs, and is conducive to extending the service life of energy storage power station equipment.

[0013] In an optional embodiment, a preset algorithm is used to filter sensitive feature data from historical feature data, and dimensionality reduction processing is performed on the sensitive feature data to obtain sensitive reduced-dimensionality feature data, including:

[0014] Normalize the original values ​​of each historical feature data to obtain the standard value of each historical feature data;

[0015] The information gain algorithm is used to calculate the information gain value of each historical feature data. The information gain value is used to represent the contribution of the historical feature data to fault judgment.

[0016] Based on the sliding window algorithm, the classification performance of the sliding window is calculated after the classifier is used for classification. The optimal classification threshold of the information gain value is selected according to the classification performance. The sensitive feature data is filtered out from the historical feature data based on the optimal classification threshold.

[0017] By analyzing the similarity of sensitive feature data, the sensitive feature data is subjected to dimensionality reduction mapping to obtain sensitive dimensionality reduction feature data.

[0018] In an optional embodiment, after performing classification using a classifier based on a sliding window algorithm, the classification performance corresponding to the sliding window is calculated, an optimal classification threshold of the information gain value is selected according to the classification performance, and sensitive feature data is screened out from the historical feature data based on the optimal classification threshold, including:

[0019] Sort the information gain values ​​of each historical feature data in ascending order, and determine at least one candidate threshold value for each historical feature data;

[0020] Using a sliding window to traverse all candidate thresholds, and using a classifier to perform classification, the classification performance of the classification results is calculated, wherein the classification performance includes accuracy and learning calculation time;

[0021] Select the candidate threshold with the highest accuracy and the shortest learning calculation time as the optimal classification threshold;

[0022] Based on the optimal classification threshold and the information gain value of each historical feature data, a classifier is used to classify each historical feature data to determine the sensitive feature data.

[0023] The data-driven energy storage power station operation and maintenance method provided by the present invention realizes feature extraction and data dimensionality reduction of energy storage charging and discharging data through multi-level data processing, combined with the information gain algorithm and dimensionality reduction mapping method. While ensuring the quality of the data set, it greatly reduces the size of the data set, providing the premise and data preparation for theoretical modeling, machine learning and data visualization.

[0024] In an optional embodiment, the normal feature data is used to train an autoencoder, and the trained autoencoder is used to detect whether the actual operation data of the energy storage power station is fault data, including:

[0025] Input the normal feature data into the automatic encoder and perform encoding and decoding in sequence to obtain the reconstructed data;

[0026] Calculate the error between the reconstructed data and the corresponding normal feature data, and optimize the parameters of the autoencoder according to the error until the error is within the preset error range, thus obtaining a trained autoencoder;

[0027] Input the actual running data into the trained autoencoder to obtain the actual reconstructed data, and calculate the actual error between the actual reconstructed data and the actual running data;

[0028] If the actual error is within the preset error range, the actual operating data is not fault data; if the actual error is not within the preset error range, the actual operating data is fault data.

[0029] In an optional implementation, performing type analysis on the fault data to determine the fault type corresponding to the fault data includes:

[0030] Cluster the fault data to obtain multiple clusters;

[0031] Perform fault simulation using an equivalent circuit model to obtain multiple simulated fault data, and label the fault type of each cluster based on the simulated fault data and the fault data;

[0032] The labeled cluster data are used to train a neural network model to obtain a neural network model for identifying fault types.

[0033] The data-driven energy storage power station operation and maintenance method provided by the present invention utilizes an autoencoder and a K-means algorithm to identify and classify faulty batteries, and optimizes the classification by coupling a fully connected neural network adversarial algorithm with a clustering analysis algorithm. An L2 regularization term is added to the loss function to reduce model complexity, thereby automating the detection and classification of battery string faults, minimizing manual intervention, and improving model efficiency.

[0034] In an optional embodiment, the equivalent circuit model of the lithium battery includes: an internal resistance, a first resistor, a first capacitor, a second resistor, a second capacitor, and a variable resistor, wherein:

[0035] an internal resistor, one end of which is connected to the positive electrode of the lithium battery and the other end of which is connected to the first end of the first resistor;

[0036] a first resistor, connected in parallel with the first capacitor, with a second end of the first resistor connected to a first end of the second resistor;

[0037] a second resistor, connected in parallel with the second capacitor, with a second end connected to the first end of the variable resistor;

[0038] A variable resistor, wherein an equivalent output voltage of the lithium battery is formed between the second end of the variable resistor and the negative electrode of the lithium battery;

[0039] Pulse charge and discharge experiments were performed at different residual capacities to obtain the parameters of the equivalent circuit model at different residual capacities, including:

[0040] Conduct pulse charge and discharge experiments on lithium batteries at different remaining capacities, and collect voltage and time data during the pulse charge and discharge process.

[0041] According to the collected voltage data and time data, the parameters of the equivalent circuit model are identified to obtain the parameter values ​​in the equivalent circuit model under different residual capacities.

[0042] The data-driven energy storage power station operation and maintenance method provided by the present invention uses an improved equivalent circuit model to perform parameter identification on the model parameters through non-invasive charge and discharge experiments. This can accurately predict the operating characteristics of lithium batteries under charge and discharge conditions, and can also ensure the prediction accuracy of lithium batteries under fault conditions.

[0043] In an optional embodiment, the gradient descent algorithm is used for inversion to obtain the true fault parameters, including:

[0044] Based on the equivalent circuit model, simulate the charge and discharge characteristic curves of lithium batteries under different fault types;

[0045] Extract fault parameters based on the charge and discharge characteristic curve and construct the response equation of the lithium battery:

[0046] Based on the response equation, different fault parameters are simulated to obtain the predicted values ​​of the fault parameters;

[0047] The mean square error between the predicted value and the measured value is defined as the objective function. The gradient of the objective function to the optimized parameter is calculated using the gradient descent algorithm, and the optimization is performed in the opposite direction of the gradient until the objective function converges to obtain the true fault parameter.

[0048] The data-driven energy storage power station operation and maintenance method provided by the present invention simulates various faults based on an improved equivalent circuit model, establishes a response equation, takes the mean square error of measured data and model predicted data as the objective function, adopts the gradient descent method to minimize the objective function, approximates the optimal solution optimization of the fault parameters, obtains the actual fault parameters corresponding to the fault when it occurs, effectively improves the fault parameter inversion accuracy, is suitable for online monitoring and early warning, provides a reliable basis for lithium battery health assessment, fault diagnosis and life prediction, and enhances system safety and management efficiency.

[0049] In a second aspect, the present invention provides a data-driven energy storage power station operation and maintenance device, wherein the energy storage power station includes multiple lithium batteries, and the device includes:

[0050] The data processing module is used to obtain historical characteristic data of the energy storage power station, pre-process the historical characteristic data, filter sensitive characteristic data from the historical characteristic data using a preset algorithm, and perform dimensionality reduction processing on the sensitive characteristic data to obtain sensitive reduced dimensionality characteristic data;

[0051] The fault classification module is used to filter out normal feature data from the sensitive dimensionality reduction feature data, use the normal feature data to train an autoencoder, and use the trained autoencoder to detect whether the actual operating data of the energy storage power station is fault data. It then performs type analysis on the fault data to determine the corresponding fault type.

[0052] An equivalent circuit model parameter determination module is used to construct an equivalent circuit model of a lithium battery and perform pulse charge and discharge experiments at different remaining capacities to obtain the parameters of the equivalent circuit model at different remaining capacities;

[0053] The fault inversion module is used to simulate the charge and discharge characteristic curves of lithium batteries under different fault types based on the equivalent circuit model, extract fault parameters based on the charge and discharge characteristic curves, and use the gradient descent algorithm to perform inversion to obtain the true fault parameters. Corresponding operation and maintenance measures are taken based on the true fault parameters.

[0054] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0055] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0057] Figure 1 is a flow chart of a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0058] Figure 2 is a flow chart of another data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0059] Figure 3 1. It is a flowchart of data dimensionality reduction based on information gain and t-SNE algorithm in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0060] Figure 4 2 is a schematic diagram of the change in classification accuracy of the classifier within different IG score threshold intervals in the data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0061] Figure 5 1. A schematic diagram of a fault diagnosis process based on an autoencoder and an adversarial neural network in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0062] Figure 6 is an error comparison diagram of different working conditions using an automatic encoder for fault detection in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0063] Figure 7 is an improved second-order RC equivalent circuit diagram including nonlinear internal resistance in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0064] Figure 8 1. A schematic diagram of a voltage recovery curve after the pulse discharge of a lithium battery is terminated, including a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0065] Figure 91 is a flow chart of a parameter estimation method based on an improved equivalent circuit in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0066] Figure 10 1. It is a flow chart of a method for performing fault inversion based on an equivalent circuit in a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention;

[0067] Figure 11 is a structural block diagram of a data-driven energy storage power station operation and maintenance device according to an embodiment of the present invention;

[0068] Figure 12 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0070] The embodiment of the present invention provides a data-driven energy storage power station operation and maintenance method, which builds a closed-loop intelligent operation and maintenance system based on a large amount of historical feature data to achieve timely and accurate feedback of fault detection results and optimize the operation and maintenance strategy.

[0071] According to an embodiment of the present invention, an embodiment of a data-driven energy storage power station operation and maintenance method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0072] In this embodiment, a data-driven energy storage power station operation and maintenance method is provided, which can be used in the above-mentioned computer system. Figure 1 is a flow chart of a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0073] Step S101 , obtaining historical feature data of the energy storage power station, preprocessing the historical feature data, screening sensitive feature data from the historical feature data using a preset algorithm, performing dimensionality reduction processing on the sensitive feature data, and obtaining sensitive reduced dimensionality feature data.

[0074] Specifically, the energy storage station's historical characteristic data includes, but is not limited to, the capacity, voltage, current, temperature, remaining charge, and health status of individual cells and / or lithium battery strings. This data is collected for both normal and various fault conditions. In this embodiment, a total of 212,084 sample data points were obtained, and the number of labeled samples for each operating state is shown in Table 1.

[0075] Table 1

[0076]

[0077] Historical feature data is standardized so that each historical feature data point has a mean of 0 and a variance of 1. A preset algorithm is used to filter out sensitive feature data that have a significant impact on the operating status from all standardized historical feature data. Each sensitive feature data point is large in size, requiring significant computing resources for subsequent applications. To reduce the data volume while maintaining the structure between data points, the t-distributed Stochastic Neighbor Embedding (t-SNE) analysis method can be used to reduce the dimensionality of the sensitive feature data, mapping the high-dimensional data into a low-dimensional space.

[0078] In step S102, normal feature data is filtered out from the sensitive dimensionality reduction feature data, an autoencoder is trained using the normal feature data, and the trained autoencoder is used to detect whether the actual operating data of the energy storage power station is fault data. The fault data is type analyzed to determine the fault type corresponding to the fault data.

[0079] Specifically, the autoencoder is trained using the selected normal feature data, with the goal of minimizing the reconstruction error between the input and output data. The mean squared error (MSE) can be used as the loss function. During training, the weights and biases of the neural network are continuously adjusted through a backpropagation algorithm to optimize the autoencoder parameters. When the loss function gradually converges and remains at a low level, the autoencoder training is considered complete, indicating that the autoencoder has learned the characteristic representation of the normal operating data of the energy storage power station.

[0080] During actual monitoring, the actual operating data is input into the automatic encoder. If the error between the output reconstructed data and the actual operating data is within the preset error range, the actual operating data is considered to be normal data. If the error between the output reconstructed data and the actual operating data exceeds the preset error range, the actual operating data is considered to be fault data.

[0081] The autoencoder only determines whether a fault exists, but cannot determine the specific type of the fault. Therefore, a trained classification model can be used to perform type analysis on the fault data to determine the fault type, which is only an example and not limited to this.

[0082] Step S103 , constructing an equivalent circuit model of the lithium battery, and performing pulse charge and discharge experiments at different remaining capacities to obtain parameters of the equivalent circuit model at different remaining capacities.

[0083] Specifically, by constructing an equivalent circuit model of the lithium battery and simulating various possible faults, the authors facilitated the measurement of parameter values ​​under various fault conditions. Pulse charge and discharge experiments were conducted at varying levels of remaining charge, enabling the collection of fault-sensitive characteristic data across all states. Pulse operating conditions (such as HPPC testing) closely mirrored the start and shutdown of energy storage power station systems, stimulating polarization effects within the lithium battery and making parameter changes more significant. Applying pulses at a fixed SOC eliminated the interference of SOC fluctuations on the parameters, allowing for the precise extraction of fault-related features.

[0084] In step S104, based on the equivalent circuit model, the charge and discharge characteristic curves of the lithium battery under different fault types are simulated, the fault parameters are extracted according to the charge and discharge characteristic curves, and the gradient descent algorithm is used for inversion to obtain the real fault parameters, and corresponding operation and maintenance measures are taken based on the real fault parameters.

[0085] Specifically, based on the analysis in steps S101-S103, the fault type can be determined based on the actual operating data of the energy storage power station. To implement more effective operation and maintenance measures, an equivalent circuit model can be used to model different fault types, thereby obtaining the charge-discharge characteristic curves of the lithium battery under different fault types. Fault parameters are extracted based on the charge-discharge characteristic curves. The fault parameters extracted here are generally sensitive fault parameters that have a significant impact on the fault type, such as lithium battery string voltage, lithium battery string current, lithium battery cell voltage, lithium battery cell current, lithium battery temperature, etc., which are examples only and are not limited to these.

[0086] Fault parameters are extracted based on the charge-discharge characteristic curve, and the measured values ​​and true values ​​are analyzed to determine the relationship between the measured values ​​and the true values. When performing fault diagnosis, the true values ​​are inferred based on the measured values ​​of the parameters, and then the fault type and the operation and maintenance measures that should be taken based on the true values ​​of the fault parameters are determined based on the true values.

[0087] The data-driven energy storage power station operation and maintenance method provided in this embodiment builds a closed-loop intelligent operation and maintenance system based on a large amount of historical feature data. It provides timely and accurate feedback on fault detection results, adapts to the needs of actual application scenarios, effectively monitors various faults in the energy storage system, improves the diagnostic capabilities and intelligence level of a series of faults, and takes targeted operation and maintenance measures in a timely manner. This optimizes the operation and maintenance strategy, reduces operation and maintenance costs, and helps extend the service life of energy storage power station equipment.

[0088] In this embodiment, a data-driven energy storage power station operation and maintenance method is provided, which can be used in the above-mentioned computer system. Figure 2 is a flow chart of a data-driven energy storage power station operation and maintenance method according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps:

[0089] Step S201 , obtaining historical feature data of the energy storage power station, preprocessing the historical feature data, screening sensitive feature data from the historical feature data using a preset algorithm, performing dimensionality reduction processing on the sensitive feature data, and obtaining sensitive reduced dimensionality feature data.

[0090] Specifically, the above step S201 includes:

[0091] Step S2011 , normalizing the original values ​​of each historical feature data to obtain a standard value of each historical feature data.

[0092] Specifically, if Figure 3 The figure shows a flow chart of data dimensionality reduction based on information gain and t-SNE algorithm. For the collected battery charge and discharge data, the Z-score normalization method is used to normalize the original values ​​of each historical feature data. For multiple original values ​​of the same type of historical feature data, the processing formula is:

[0093]

[0094] Among them, x i Represents the i-th original value, μ represents the mean of the original values ​​under this feature, σ represents the standard deviation of the original values ​​under this feature, and Z represents the standard value obtained after normalization of the original values, ranging from [-1, 1].

[0095] Step S2012: Calculate the information gain value of each historical feature data using an information gain algorithm. The information gain value is used to represent the contribution of the historical feature data to fault determination.

[0096] Specifically, the information gain (IG) score of each feature data is calculated through the information gain (IG) algorithm, and the feature set with the highest information gain is selected. The information gain is used to evaluate the contribution of each feature to fault classification. The calculation process includes:

[0097] (1) Based on all feature data, calculate the original entropy:

[0098]

[0099] Among them, E(S) represents the original entropy of all feature data S, C represents the total number of fault types, and p i represents the probability of type i failure.

[0100] (2) Using multiple standard values ​​corresponding to the feature data, calculate the information gain value of each feature:

[0101]

[0102] Among them, Values(A) is all possible values ​​of feature A, S v is a subset of feature A with value v, |S v | and |S| are the sizes of the subset and the total set respectively.

[0103] Step S2013: Based on the sliding window algorithm, the classification is performed using a classifier and the classification performance corresponding to the sliding window is calculated. The optimal classification threshold of the information gain value is selected according to the classification performance, and sensitive feature data is filtered out from the historical feature data based on the optimal classification threshold.

[0104] Specifically, the IG score in this embodiment's information gain algorithm uses a sliding window test algorithm to determine the optimal threshold, ensuring that the selected feature subset achieves optimal classification performance. The sliding window algorithm traverses all possible thresholds for a feature and calculates the information gain corresponding to each threshold, thereby finding the optimal split point and discretizing continuous feature data into binary splits.

[0105] In some optional implementations, the above step S2013 includes:

[0106] Step a1: sort the information gain values ​​of each historical feature data in ascending order, and determine at least one candidate threshold value for each historical feature data.

[0107] Specifically, for each feature, the IG scores of the historical feature data are sorted: {IG1, IG2, ..., IG n}, define a sliding window W(t) of length w, which contains w consecutive IG scores starting from the t-th IG score: W(t) = {IG t ,IGt+1 ,…,IG t+w-1}, the minimum IG score in each sliding window is a candidate threshold.

[0108] In step a2, all candidate thresholds are traversed using a sliding window, and classification is performed using a classifier to calculate the classification performance of the classification results. The classification performance includes accuracy and learning calculation time.

[0109] Specifically, all possible sliding windows are traversed, and the information gain value under each window is calculated to find the window with the best classification performance W(t * ).

[0110] In step a3, the candidate threshold corresponding to the highest accuracy and the shortest learning calculation time is selected as the optimal classification threshold.

[0111] Specifically, the minimum IG score within the optimal window is defined as the optimal threshold: This is just an example, but not limiting.

[0112] Step a4: Based on the optimal classification threshold and the information gain value of each historical feature data, a classifier is used to classify each historical feature data to determine sensitive feature data.

[0113] Specifically, the classifier is used to determine the degree of influence of each feature data on whether a fault occurs, and the feature data with the highest degree of influence on the occurrence of the fault is used as the sensitive feature data. For example, the feature data with the highest degree of influence is used as the sensitive feature data. The sensitive feature data in this embodiment include: lithium battery string current I s , lithium battery string voltage V s , Single lithium battery current I i , Single lithium battery voltage V i , Single lithium battery temperature T i , is only used as an example, but is not limited to this.

[0114] Step S2014: By performing similarity analysis on the sensitive feature data, the sensitive feature data is subjected to dimensionality reduction mapping to obtain sensitive dimensionality reduction feature data.

[0115] Specifically, the process of using t-SNE analysis to reduce the dimension of sensitive feature data includes:

[0116] (1) For each pair of data points x in the high-dimensional dataset (sensitive feature data) i and x j , using Gaussian distribution to define its similarity:

[0117]

[0118] Among them, σ iis the data point x i The local area width.

[0119] (2) Calculate the conditional probability:

[0120] Among them, N represents the total number of samples in the data set, p ij is the joint probability between two points.

[0121] (3) Calculate the similarity of low-dimensional space: After randomly initializing the low-dimensional space points, calculate each pair of low-dimensional embedding points y i and y j The similarity of , using t distribution to define the similarity of low-dimensional space,

[0122]

[0123] (4) Minimize the similarity difference between high-dimensional space and low-dimensional space, and use KL divergence as the loss function to measure the difference between the similarity of high-dimensional space and low-dimensional space:

[0124]

[0125] (5) Use the gradient descent method to minimize the loss function C. The gradient calculation formula is:

[0126]

[0127] The goal is to continuously adjust the coordinate y in the low-dimensional space i , so that its similarity q ij As close as possible to the high-dimensional space ij The method is similar to using gradient descent to optimize the position of the low-dimensional space point. Finally, the low-dimensional space coordinate y i When the value of changes less than the given tolerance, the y i It can be considered as the optimal value.

[0128] It should be noted that the classification performance can be directly analyzed using high-dimensional data, or it can be analyzed using reduced-dimensional data. In this embodiment, the reduced-dimensional data is used to train the classifier to reduce the amount of computation and evaluate its performance in fault detection. In this embodiment, three classifiers, CART, KNN, and RF, are tested separately to compare their classification accuracy (PACC) and learning computation time (LCT). Figure 4 This is a graph showing the change in classification accuracy of the classifier within different IG score threshold ranges. Figure 4Table 2 and Table 3 show the classifier performance for different sliding window intervals, verifying the effectiveness of the sliding window test algorithm in optimizing feature selection. By selecting an appropriate threshold interval, the classifier accuracy can be significantly improved and the computation time can be reduced, thereby optimizing the overall performance of fault detection.

[0129] Table 2

[0130]

[0131] The data-driven energy storage power station operation and maintenance method provided in this embodiment implements feature extraction and data dimensionality reduction of energy storage charging and discharging data through multi-level data processing, combined with an information gain algorithm and a dimensionality reduction mapping method. This significantly reduces the size of the data set while ensuring its quality, providing the premise and data preparation for theoretical modeling, machine learning, and data visualization.

[0132] In step S202, normal feature data is filtered out from the sensitive dimensionality reduction feature data, an autoencoder is trained using the normal feature data, and the trained autoencoder is used to detect whether the actual operating data of the energy storage power station is fault data. The fault data is type analyzed to determine the fault type corresponding to the fault data.

[0133] Specifically, the above step S202 includes:

[0134] Step S2021: Input the normal feature data into the automatic encoder for encoding and decoding in sequence to obtain reconstructed data.

[0135] Specifically, the high-dimensional feature data of the energy storage power station during normal operation is input into the autoencoder, which is then encoded by the autoencoder to obtain low-dimensional data, which is then decoded and reconstructed to obtain reconstructed data.

[0136] Unsupervised learning is used to identify faulty individuals and automatically encode and decode the operating data of lithium-ion battery cells in energy storage systems. The autoencoder compresses the input high-dimensional feature data into a low-dimensional latent representation, and the autodecoder reconstructs the latent representation into reconstructed data. This essentially maps the high-dimensional input data into a low-dimensional latent space and then reconstructs it into a high-dimensional output.

[0137] The formula for automatic encoding is:

[0138] Z=f(X)=σ e (W e X+b e )

[0139] Among them, W e is the weight matrix of the encoder, b e is the bias, σ e is the activation function.

[0140] The formula for automatic decoding is:

[0141] X′=g(Z)=σ d (W d Z+b d )

[0142] Among them, W d is the weight matrix of the decoder, b d is the bias, σ d is the activation function.

[0143] Step S2022 , calculating the error between the reconstructed data and the corresponding normal feature data, and optimizing the parameters of the autoencoder according to the error until the error is within a preset error range, thereby obtaining a trained autoencoder.

[0144] Specifically, the goal of the autoencoder is to make the reconstructed data X' as close as possible to the original input feature data X, calculate the mean square error between the input feature data and the reconstructed data, and adjust the parameters of the autoencoder based on the relationship between the calculated mean square error and the preset error range until the error is within the preset error range, thereby obtaining a trained autoencoder.

[0145] Step S2023: Input the actual operation data into the trained autoencoder to obtain actual reconstructed data, and calculate the actual error between the actual reconstructed data and the actual operation data.

[0146] Specifically, when using an autoencoder for fault detection, whether a fault exists is determined by calculating the mean square error between actual operating data and actual reconstructed data.

[0147] Step S2024: If the actual error is within the preset error range, the actual operation data is not fault data; if the actual error is not within the preset error range, the actual operation data is fault data.

[0148] Specifically, if the actual error is within the preset error range, the actual operation data is not fault data and the lithium battery operates normally. If the actual error is not within the preset error range, the actual operation data is fault data and the lithium battery fails.

[0149] Since it is unsupervised machine learning, the fault type needs to be classified using other methods later. The goal of the autoencoder is to minimize the mean square error between the input X and the reconstructed output X', as follows:

[0150]

[0151] Where n is the number of samples, X i is the feature data of the original input, X' iis the reconstructed data output after the encoder-decoder reconstruction. When the reconstruction error exceeds the threshold, the actual operation data is considered to contain fault data.

[0152] Step S2025: cluster the fault data to obtain multiple clusters.

[0153] Specifically, after detecting fault data, cluster analysis is required to determine the specific fault type. The common K-means method is used for cluster analysis. K-means clustering is the partitioning of the sample set X. Through the K-means operation, the problem of selecting samples into classes can be solved. The strategy is as follows:

[0154] Randomly select K data points c i (i=1~K) as the initial centroid (K is the number of clusters), calculate the Euclidean distance of other data points to the K centroids, divide these data points into the centroids closest to them, and finally divide all the data in the set into K points c i (i=1~K) are the K clusters (classes) with the centroids. The Euclidean distance formula is as follows, where d represents the Euclidean distance, X i Represents the data vector numbered i.

[0155] d ij =d(X i ,X j )=||X i -X j || 2

[0156] For each cluster (class), update the centroid c i , the new centroid is the vector average of all samples in the cluster, where X c Represents the new centroid corresponding vector, and N is the total number of samples in the current cluster.

[0157]

[0158] The steps of calculating the Euclidean distances from other data points to these K centroids are repeated until the centroids do not change, thus completing the K classifications of the sample set X.

[0159] Step S2026 , performing fault simulation using the equivalent circuit model to obtain a plurality of simulated fault data, and marking the fault type of each cluster based on the simulated fault data and the fault data.

[0160] Specifically, cluster analysis classifies data, but it cannot determine the fault type corresponding to each cluster. In order to complete the identification of different fault types of the energy storage system, and at the same time to optimize the selection of K value and eliminate the local optimal problem caused by the random selection of the initial centroid, special processing of the data is required.

[0161] The sample data is processed using a lithium battery equivalent circuit model to simulate various faults, such as internal short circuits, external short circuits, inconsistent capacity, connection failures, overcharge, overdischarge, and uneven aging. The simulated data is then standardized and dimensionally reduced before being merged into the original dataset. The original sample data from the fault dataset is analyzed and labeled, and the labels are incorporated into the fault dataset. Choosing the initial centroid for the K-means classification to be within these labeled categories, as much as possible, can improve the dispersion and stability of the initial centroid.

[0162] Step S2027: Use the labeled cluster data to train a neural network model to obtain a neural network model for identifying fault types.

[0163] Specifically, if Figure 5 Figure 2 shows a fault diagnosis flow chart based on autoencoders and adversarial neural networks. To improve classification accuracy and authenticity, an adversarial training algorithm based on a three-layer fully connected neural network is coupled with a cluster analysis algorithm to optimize the classification. First, a K-means method is used with simulated fault data samples to perform a preliminary classification of fault types and generate labels. Then, using the three-layer fully connected neural network, adversarial samples are generated using the fast gradient signed algorithm, using the initial classification labels and the fault sample data obtained from autoencoder detection. After generating the adversarial samples, the original and adversarial samples are combined, and L2 regularization is applied. This not only enhances model stability but also limits model complexity, prevents overfitting, and improves model generalization.

[0164] The formula for adversarial samples is:

[0165]

[0166] Among them, x i For the original sample, For adversarial samples, y i is the label, ε is the perturbation, sign is the activation function, f θ is the output of the neural network model, which represents the probability distribution of the corresponding category.

[0167] L is the loss function, the loss function of the original sample and the loss function of the adversarial sample, and the multi-classification cross entropy loss function can be used:

[0168]

[0169] Among them, y i,c is the label that sample i belongs to category c, is the probability that the model predicts that sample i belongs to the label of category c, N is the number of samples, and C is the number of categories.

[0170] The L2 regularization loss function can be written as:

[0171]

[0172] Among them, w i is the weight of sample i.

[0173] Then the joint loss function can be written as:

[0174]

[0175] Then, the joint loss function is minimized through algorithms such as gradient descent, and model parameters such as weights and biases are updated, and the algorithm is iterated repeatedly until the model converges.

[0176] like Figure 6 The figure below compares the errors of different operating conditions for fault detection using an autoencoder. The reconstruction error of the operating data under normal operating conditions is small, while the reconstruction error of the operating data under fault conditions is large. Therefore, the autoencoder can be used to determine whether a fault exists.

[0177] The data-driven energy storage power station operation and maintenance method provided in this embodiment utilizes an autoencoder and a K-means algorithm to identify and classify faulty batteries. A fully connected neural network adversarial algorithm coupled with a clustering analysis algorithm is used to optimize the classification. An L2 regularization term is added to the loss function to reduce model complexity, thereby automating the detection and classification of battery string faults, minimizing manual intervention, and improving model efficiency.

[0178] Step S203 , constructing an equivalent circuit model of the lithium battery, and performing pulse charge and discharge experiments at different remaining capacities to obtain parameters of the equivalent circuit model at different remaining capacities.

[0179] Specifically, the constructed equivalent circuit model includes: an internal resistor, one end of which is connected to the positive electrode of the lithium battery and the other end of which is connected to the first end of a first resistor; a first resistor, which is connected in parallel with a first capacitor and whose second end is connected to the first end of a second resistor; a second resistor, which is connected in parallel with the second capacitor and whose second end is connected to the first end of a variable resistor; and an equivalent output voltage of the lithium battery is generated between the second end of the variable resistor and the negative electrode of the lithium battery.

[0180] Specifically, if Figure 7 As shown in the figure, it is an improved second-order RC equivalent circuit diagram including nonlinear internal resistance, where the ohmic internal resistance R0 represents the internal resistance of the internal components of the lithium battery and the contact internal resistance, the RC parallel part represents the electrochemical polarization effect and concentration polarization effect, and the variable resistance R L The internal resistance of the lithium battery increases under the condition of short-term depletion of lithium ions. Compared with the commonly used Thevenin model equivalent circuit, the variable resistor R is added. LThrough experimental observation, it was found that when the discharge current of the lithium battery is small, the commonly used Thevenin model equivalent circuit can accurately reflect the operating characteristics of the lithium battery. However, when the discharge current is large, the commonly used Thevenin model equivalent circuit cannot accurately reflect the operating characteristics of the lithium battery. The battery operating characteristics will deviate. Therefore, the variable resistor is added to reduce the characteristic deviation.

[0181] like Figure 7 In the equivalent circuit shown, U oc Indicates the open circuit voltage of the lithium battery, U t Represents the terminal voltage of the lithium battery when it is working. If U1 and U2 represent the terminal voltages of R1, C2 and R2, C2, that is, the electrochemical polarization voltage and concentration polarization voltage, and i represents the output current of the lithium battery, then according to Kirchhoff's law, we can get:

[0182]

[0183] U t =U oc -(U1+U2+iR0+iR L )

[0184] In order to obtain the parameters R1, C1, R2, C2, R0, R L To determine the specific value of , it is necessary to perform parameter identification on the lithium battery. The parameter identification can adopt an offline parameter identification method. In this embodiment, pulse charge and discharge experiments are performed on the lithium battery under different remaining capacities (State of Charge, SOC), and regression analysis is performed on the changes in the terminal voltage to obtain the corresponding equivalent circuit parameters.

[0185] The above step S203 includes:

[0186] Step S2031 , performing pulse charge and discharge experiments on the lithium battery at different remaining capacities, and collecting voltage data and time data during the pulse charge and discharge process.

[0187] Specifically, place the battery in a constant temperature box, perform pulse discharge at a constant current not exceeding 0.5C for 10 seconds, let it stand for 50 seconds, and then charge it at the same rate for 10 seconds and let it stand for 50 seconds. At this time, since the charge and discharge rate is small, there is no local depletion of lithium ions, and the variable resistance RL can be regarded as zero. When the pulse discharge stops, although the current becomes zero, the polarization effect does not disappear, so the terminal voltage cannot immediately return to the open circuit voltage, but shows a slow rise. At this time, the change of the terminal voltage is as follows: Figure 8 As shown, it is the voltage recovery curve of the lithium battery after the pulse discharge is cut off.

[0188] Step S2032 : performing parameter identification on the equivalent circuit model based on the collected voltage data and time data to obtain parameter values ​​in the equivalent circuit model under different remaining capacities.

[0189] Specifically, Figure 8 In the figure, AB is a linear straight line. It is the moment when the pulse discharge stops. The impedance of C1 and C2 is zero. The linear voltage drop caused by the ohmic internal resistance R0 disappears. The voltage shows an instantaneous linear rise from point B to point A. Therefore, we can get: Due to the existence of the RC circuit, when restored to the AC end, the AC section can be regarded as the zero-state response of the equivalent circuit. Using the voltage response of the AC section, the parameter values ​​of R1, C1 and R2, C2 can be obtained. Ignoring the variable resistor R L , the dynamic response equation of the terminal voltage is:

[0190]

[0191] After sorting, we can get the fitting function: U t =a+be -ct +de -ft

[0192] Among them, a=U oc -i(R0+R1+R2), b=iR1, c=1 / τ1, d=iR2, f=1 / τ2. By fitting the terminal voltage data of the BC segment with the least squares method, the corresponding parameters a, b, c, d, and f can be obtained, which are further converted according to the following formula:

[0193]

[0194] After conversion, the equivalent circuit parameters R1, C1, R2, and C2 are obtained. The battery is charged for 120 seconds at the same rate as the discharge current, and then left to stand for 2 hours until the terminal voltage no longer changes. At this time, a small rate charge and discharge is considered complete.

[0195] like Figure 9 Figure 2 shows a flow chart of the parameter estimation method based on the improved equivalent circuit. At the same temperature, for different SOCs, the above process is repeated to obtain the equivalent circuit parameter values ​​R0, R1, C1, R2, and C2 of the lithium battery at different SOCs. An interpolation table for the five parameters at different SOCs is constructed to ensure that the corresponding values ​​can be obtained from the interpolation table at any SOC.

[0196] Perform a high current charge and discharge experiment on the battery. At this time, the variable resistor R L It shows a nonlinear increasing trend, and RL and terminal voltage Ut satisfy the following equation:

[0197]

[0198] Under zero boundary conditions, the equation has the following solutions:

[0199]

[0200] Among them, a and b are parameters that need to be determined through high-current discharge voltage identification.

[0201] Under high current pulse discharge conditions, the voltage U across the variable resistor is L It can be expressed as

[0202]

[0203] Among them, U t It is directly measured, U oc It can be obtained by looking up the table according to SOC (note that at this time, because SOC is changing, the open circuit voltage U oc Changes with time), R0, R1, C1 and R2, C2 are the electrical parameters identified in the above process. L =U L / i can ultimately determine the value of the variable resistor at different times.

[0204] According to R L The expression of the solution and the obtained R L Numerical sequence, using the least squares method, to find the corresponding parameters a and b.

[0205] By introducing second-order RC elements and variable resistance elements, the equivalent circuit model can be used in a wider range to simulate the working state of the lithium battery. It can more accurately simulate the changes in the terminal voltage of the lithium battery under different discharge currents, which is beneficial to the analysis and calculation of complex working conditions of the energy storage system.

[0206] The data-driven energy storage power station operation and maintenance method provided in this embodiment uses an improved equivalent circuit model to perform parameter identification on the model parameters through non-invasive charge and discharge experiments. This method can accurately predict the operating characteristics of lithium batteries under charge and discharge conditions, and can further ensure the prediction accuracy of lithium batteries under fault conditions.

[0207] In step S204, based on the equivalent circuit model, the charge and discharge characteristic curves of the lithium battery under different fault types are simulated, the fault parameters are extracted according to the charge and discharge characteristic curves, and the gradient descent algorithm is used for inversion to obtain the real fault parameters, and corresponding operation and maintenance measures are taken based on the real fault parameters.

[0208] Specifically, the above step S204 includes:

[0209] Step S2041 , simulating the charge and discharge characteristic curves of the lithium battery under different fault types based on the equivalent circuit model.

[0210] Specifically, based on the improved equivalent circuit model, the charge and discharge characteristic curves of lithium batteries under different fault conditions are simulated. Figure 10 As shown in the figure, it is a flow chart of the fault inversion method based on the equivalent circuit. According to the fault characteristics of different lithium battery strings, the parameters such as the internal resistance and polarization capacitance in the equivalent circuit are changed to analyze the change law of the output current and voltage characteristic curves of the lithium battery string.

[0211] Step S2042: extract fault parameters according to the charge-discharge characteristic curve and construct a response equation of the lithium battery.

[0212] Specifically, according to the equivalent circuit model of the lithium battery, the response equation of the lithium battery can be expressed as:

[0213] AX=Y

[0214] Where A is the forward operator for the lithium battery system, X is the system model parameter or the fault parameter to be inverted, and Y is the system response parameter or the measured output variable data. The forward operator A can be derived from the equivalent circuit model of the lithium battery or from the measured charge and discharge curve of the lithium battery.

[0215] Step S2043: Based on the response equation, different fault parameters are simulated to obtain predicted values ​​of the fault parameters.

[0216] Specifically, based on the response equation in step S2042, with the operating current of the energy storage battery as input, different fault parameters are simulated to obtain the output variables of the system (voltage, temperature, etc.) as the predicted values ​​of the fault parameters.

[0217] In step S2044, the mean square error between the predicted value and the measured value is defined as the objective function, and the gradient of the objective function with respect to the parameter to be optimized is calculated using the gradient descent algorithm. The optimization is performed in the opposite direction of the gradient until the objective function converges to obtain the true fault parameter.

[0218] Specifically, the mean square error J is defined as the objective function (loss function) to measure the deviation between the model output variable data and the measured output data:

[0219]

[0220] Among them, y th Represents the model prediction value, y m represents the measured fault data, and N is the number of data points. The goal of the inversion is to minimize this objective function through an optimization algorithm to obtain the true fault parameters.

[0221] To improve the accuracy of fault inversion, a gradient descent algorithm is used to optimize the fault parameters. The core idea of ​​the gradient descent algorithm is to continuously adjust the parameters in the opposite direction of the objective function's gradient, gradually approaching the optimal solution.

[0222] Assume that the objective function J(θ) is a function of θ, where θ is the parameter vector to be optimized: θ = [θ1, θ2, ..., θ n ], calculate the gradient of the objective function J with respect to θ, that is The learning and update of the optimization parameters are performed in the opposite direction of the gradient:

[0223] Since the degree of nonlinearity of the physical process corresponding to each parameter in θ is different, η is taken as a diagonal matrix, representing the learning rate corresponding to these different parameters.

[0224] The above formula is iteratively updated until the Euclidean norm of its gradient is less than a small amount ε, which can be considered as convergence:

[0225] At this point, it can be considered that the optimized parameter solution vector θ that meets the minimization of the objective function J is obtained, that is, the true fault parameter.

[0226] During the fault inversion process, the system establishes a forward operator based on the current, voltage, temperature and other data of the charging and discharging process using an equivalent circuit model, and uses optimization algorithms such as the gradient descent algorithm to determine the actual fault parameters.

[0227] The data-driven energy storage power station operation and maintenance method provided by the present invention simulates various faults based on an improved equivalent circuit model, establishes a response equation, takes the mean square error of measured data and model predicted data as the objective function, adopts the gradient descent method to minimize the objective function, approximates the optimal solution optimization of the fault parameters, obtains the actual fault parameters corresponding to the fault when it occurs, effectively improves the fault parameter inversion accuracy, is suitable for online monitoring and early warning, provides a reliable basis for lithium battery health assessment, fault diagnosis and life prediction, and enhances system safety and management efficiency.

[0228] In this embodiment, a data-driven energy storage power station operation and maintenance device is also provided. The device is used to implement the above-mentioned embodiments and preferred embodiments. The details that have been described will not be repeated here. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0229] This embodiment provides a data-driven energy storage power station operation and maintenance device, the energy storage power station includes multiple lithium batteries, such as Figure 11 Shown, including:

[0230] The data processing module 1101 is used to obtain historical feature data of the energy storage power station, pre-process the historical feature data, filter sensitive feature data from the historical feature data using a preset algorithm, and perform dimensionality reduction processing on the sensitive feature data to obtain sensitive reduced dimensionality feature data.

[0231] The fault classification module 1102 is used to filter out normal feature data from the sensitive dimensionality reduction feature data, use the normal feature data to train an autoencoder, and use the trained autoencoder to detect whether the actual operating data of the energy storage power station is fault data, perform type analysis on the fault data, and determine the fault type corresponding to the fault data.

[0232] The equivalent circuit model parameter determination module 1103 is used to construct an equivalent circuit model of the lithium battery and perform pulse charge and discharge experiments under different residual capacities to obtain parameters of the equivalent circuit model under different residual capacities.

[0233] The fault inversion module 1104 is used to simulate the charge and discharge characteristic curves of the lithium battery under different fault types based on the equivalent circuit model, extract the fault parameters according to the charge and discharge characteristic curves, and use the gradient descent algorithm to perform inversion to obtain the real fault parameters, and take corresponding operation and maintenance measures based on the real fault parameters.

[0234] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0235] The data-driven energy storage power station operation and maintenance device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0236] The embodiment of the present invention also provides a computer device having the above Figure 11 The data-driven energy storage power station operation and maintenance device shown.

[0237] See also Figure 12 , Figure 12 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 12As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 12 A processor 10 is taken as an example.

[0238] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0239] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0240] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0241] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0242] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0243] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0244] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A data-driven energy storage power station operation and maintenance method, wherein the energy storage power station includes multiple lithium batteries, characterized in that: The method comprises: Acquire historical characteristic data of the energy storage power station, preprocess the historical characteristic data, filter sensitive characteristic data from the historical characteristic data using a preset algorithm, perform dimensionality reduction processing on the sensitive characteristic data, and obtain sensitive reduced dimensionality characteristic data; Normal feature data is screened out from the sensitive dimensionality reduction feature data, an autoencoder is trained using the normal feature data, and the trained autoencoder is used to detect whether the actual operating data of the energy storage power station is fault data, and a type analysis is performed on the fault data to determine the fault type corresponding to the fault data; Construct an equivalent circuit model of the lithium battery and conduct pulse charge and discharge experiments at different residual capacities to obtain the parameters of the equivalent circuit model at different residual capacities. Based on the equivalent circuit model, the charge and discharge characteristic curves of lithium batteries under different fault types are simulated. The fault parameters are extracted according to the charge and discharge characteristic curves, and the gradient descent algorithm is used for inversion to obtain the true fault parameters. Corresponding operation and maintenance measures are taken based on the true fault parameters.

2. The method according to claim 1, characterized in that Using a preset algorithm to filter sensitive feature data from the historical feature data, performing dimensionality reduction processing on the sensitive feature data to obtain sensitive dimensionality reduction feature data, including: Normalize the original values ​​of each historical feature data to obtain the standard value of each historical feature data; Calculate the information gain value of each historical feature data using the information gain algorithm, where the information gain value is used to characterize the contribution of the historical feature data to fault determination; Based on the sliding window algorithm, the classification performance of the sliding window is calculated after the classifier is used for classification. The optimal classification threshold of the information gain value is selected according to the classification performance. The sensitive feature data is filtered out from the historical feature data based on the optimal classification threshold. By analyzing the similarity of sensitive feature data, the sensitive feature data is subjected to dimensionality reduction mapping to obtain sensitive dimensionality reduction feature data.

3. The method according to claim 2, characterized in that Based on the sliding window algorithm, the classification performance of the sliding window is calculated after classification using the classifier. The optimal classification threshold of the information gain value is selected based on the classification performance. Sensitive feature data is filtered out from the historical feature data based on the optimal classification threshold, including: Sort the information gain values ​​of each historical feature data in ascending order, and determine at least one candidate threshold value for each historical feature data; Using a sliding window to traverse all candidate thresholds, and using a classifier to perform classification, the classification performance of the classification results is calculated, wherein the classification performance includes accuracy and learning calculation time; Select the candidate threshold with the highest accuracy and the shortest learning calculation time as the optimal classification threshold; Based on the optimal classification threshold and the information gain value of each historical feature data, a classifier is used to classify each historical feature data to determine sensitive feature data.

4. The method according to claim 1, wherein The autoencoder is trained using normal feature data, and the trained autoencoder is used to detect whether the actual operating data of the energy storage power station is fault data, including: Input the normal feature data into the automatic encoder and perform encoding and decoding in sequence to obtain the reconstructed data; Calculating the error between the reconstructed data and the corresponding normal feature data, and optimizing the parameters of the autoencoder according to the error until the error is within a preset error range, thereby obtaining a trained autoencoder; Inputting the actual operation data into the trained autoencoder to obtain actual reconstructed data, and calculating the actual error between the actual reconstructed data and the actual operation data; If the actual error is within the preset error range, the actual operation data is not fault data; if the actual error is not within the preset error range, the actual operation data is fault data.

5. The method according to claim 1, wherein Performing type analysis on the fault data to determine the fault type corresponding to the fault data includes: Clustering the fault data to obtain multiple clusters; Perform fault simulation using an equivalent circuit model to obtain multiple simulated fault data, and label the fault type of each cluster based on the simulated fault data and the fault data; The labeled cluster data are used to train a neural network model to obtain a neural network model for identifying fault types.

6. The method according to claim 1, characterized in that The equivalent circuit model of the lithium battery includes: internal resistance, first resistor, first capacitor, second resistor, second capacitor, and variable resistor, wherein: an internal resistor, one end of which is connected to the positive electrode of the lithium battery and the other end of which is connected to the first end of the first resistor; a first resistor, connected in parallel with the first capacitor, with a second end of the first resistor connected to the first end of the second resistor; a second resistor, connected in parallel with the second capacitor, with a second end connected to the first end of the variable resistor; A variable resistor, wherein an equivalent output voltage of the lithium battery is formed between the second end of the variable resistor and the negative electrode of the lithium battery; Pulse charge and discharge experiments were performed at different residual capacities to obtain the parameters of the equivalent circuit model at different residual capacities, including: Conduct pulse charge and discharge experiments on lithium batteries at different remaining capacities, and collect voltage and time data during the pulse charge and discharge process. According to the collected voltage data and time data, the parameters of the equivalent circuit model are identified to obtain the parameter values ​​in the equivalent circuit model under different residual capacities.

7. The method according to claim 1, characterized in that The gradient descent algorithm is used for inversion to obtain the true fault parameters, including: Based on the equivalent circuit model, simulate the charge and discharge characteristic curves of lithium batteries under different fault types; Extract fault parameters based on the charge and discharge characteristic curve and construct the response equation of the lithium battery: Based on the response equation, different fault parameters are simulated to obtain predicted values ​​of the fault parameters; The mean square error between the predicted value and the measured value is defined as the objective function. The gradient of the objective function to the parameter to be optimized is calculated using the gradient descent algorithm. The optimization is performed in the opposite direction of the gradient until the objective function converges to obtain the true fault parameter.

8. A data-driven energy storage power station operation and maintenance device, wherein the energy storage power station includes multiple lithium batteries, characterized in that: The device comprises: A data processing module is used to obtain historical characteristic data of the energy storage power station, pre-process the historical characteristic data, filter sensitive characteristic data from the historical characteristic data using a preset algorithm, and perform dimensionality reduction processing on the sensitive characteristic data to obtain sensitive reduced dimensionality characteristic data; A fault classification module is used to filter out normal feature data from the sensitive dimensionality reduction feature data, use the normal feature data to train an autoencoder, and use the trained autoencoder to detect whether the actual operating data of the energy storage power station is fault data, perform type analysis on the fault data, and determine the fault type corresponding to the fault data; An equivalent circuit model parameter determination module is used to construct an equivalent circuit model of a lithium battery and perform pulse charge and discharge experiments at different remaining capacities to obtain the parameters of the equivalent circuit model at different remaining capacities; The fault inversion module is used to simulate the charge and discharge characteristic curves of lithium batteries under different fault types based on the equivalent circuit model, extract fault parameters based on the charge and discharge characteristic curves, and use the gradient descent algorithm to perform inversion to obtain the true fault parameters. Corresponding operation and maintenance measures are taken based on the true fault parameters.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images