Battery energy storage power station state evaluation and intelligent sensing method and system

Abstract

The invention discloses a battery energy storage power station state evaluation and intelligent sensing method and system, and the method comprises the steps: collecting the core monitoring quantity of a battery energy storage power station, and carrying out the data cleaning and preprocessing; and then, through multi-order feature extraction and complex relation fusion, multi-dimensional monitoring quantity information is mapped to a state evaluation space, and comprehensive quantification of the battery operation state is realized. Thirdly, modeling the evolution trajectory of the state points of the battery pack in a state evaluation space by using a depth time feature convolutional network; and finally, through a fuzzy expert state analysis system, outputting the health state and abnormal event information of the battery energy storage power station, and providing a basis for situation awareness of the battery energy storage power station. The method can solve the problem that the battery energy storage power station is insufficient in operation situation perception, health state evaluation and operation trend prediction capability, and provides technical support for intelligent operation and maintenance of the battery energy storage power station.

Description

Technical Field

[0001] The present invention relates to the technical field of battery state assessment, and specifically to a method and system for battery energy storage power station state assessment and intelligent perception. Background Art

[0002] With the rapid development of new energy technologies, battery energy storage technology has gradually become an important means to solve the stability problems of power systems. Battery energy storage power stations achieve the dynamic balance of grid load by adjusting the charge and discharge states of batteries, thereby ensuring the stable operation of the power grid. However, during the long-term operation of battery energy storage power stations, due to factors such as environmental impacts, battery aging, and complex operating conditions, performance degradation or even failures of varying degrees may occur. Therefore, it is necessary to perceive the operating status of battery energy storage power stations in real time and comprehensively, and conduct quantitative assessment and trend prediction of their health status.

[0003] In the prior art, the assessment methods for the health status of battery energy storage power stations are mostly based on the analysis of single parameters or traditional machine learning algorithms. However, due to the complexity of battery energy storage power stations, these methods usually can only evaluate the status of a single battery cluster, and it is difficult to capture the complex coupling relationships of the overall operation of the energy storage power station by integrating multi-dimensional monitoring data, and cannot accurately reflect the overall operating status of the energy storage power station. The patent with the publication number CN119024195A proposes a method for online assessment of digital energy storage battery packs, which collects the operating parameter data of energy storage battery packs in real time through a digital data acquisition and transmission system, and constructs a data analysis model system using deep learning algorithms to automatically identify the abnormal status of energy storage battery packs, predict the change trend of their future performance indicators, and generate an assessment report, thereby realizing the intelligent monitoring and management of energy storage battery packs. The patent with the publication number CN114578251A proposes a method and device for assessing the safety status of battery modules based on convolutional neural networks. The method includes extracting the output voltage, current, and surface temperature of battery cells in different states, forming a database and dividing it into a training set and a test set, training the convolutional neural network with them until the output accuracy reaches the standard, and then inputting real-time data to obtain the safety and health status information of battery cells. The problems of these methods are: (1) The data dimension is low, and key aging indicators such as internal resistance and capacity are not fused, resulting in one-sided state assessment; (2) The complex coupling relationships between multiple parameters are not explicitly modeled, and it is difficult to reflect the overall operating status of the battery pack; (3) The comprehensive analysis of each battery pack in the energy storage power station is not carried out, and the long-term dynamic evolution law of the energy storage power station cannot be predicted. All in all, the existing methods only conduct local assessments on battery packs and cannot achieve the quantitative assessment and intelligent perception of the overall operating status of battery energy storage power stations, with the core defect of "seeing only the trees but not the forest".

[0004] Due to the lack of intelligent health state assessment and trend prediction methods, energy storage power stations still face challenges in terms of safety, reliability, and operation efficiency, which restricts the popularization and application of the "condition-based maintenance" mode and affects the large-scale deployment of energy storage power stations in the new power system. Therefore, there is an urgent need for an intelligent perception and state assessment method for the operation status of battery energy storage power stations based on multi-parameter fusion analysis that can comprehensively capture complex coupling relationships, so as to improve the intelligent operation and maintenance level and operation safety of energy storage power stations and promote the efficient development of the new power system. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the present invention provides a state assessment and intelligent perception method and system for a battery energy storage power station based on multi-parameter fusion analysis that can comprehensively capture complex coupling relationships. The present invention can solve the problems of insufficient ability in the perception of the operation status, health state assessment, and operation trend prediction of battery energy storage power stations, and provide technical support for the intelligent operation and maintenance of battery energy storage power stations.

[0006] The present invention adopts the following technical solutions.

[0007] The present invention proposes a state assessment and intelligent perception method for a battery energy storage power station, including: Step S10, collecting historical data of each monitored quantity in the battery energy storage power station, including the voltage , current , capacity , internal resistance and temperature of each battery pack, and performing data cleaning and preprocessing; Step S20, inputting the preprocessed data into a time-frequency domain hybrid feature extraction module to construct a multi-order feature vector of the monitored quantity; extracting the complex coupling relationship between the monitored quantities in the multi-order feature vector through a relationship fusion module; generating a state assessment vector based on the complex coupling relationship to quantify the operation status of the battery pack in the battery energy storage power station; Step S30, constructing a state assessment space based on the state assessment vector; in the state assessment space , modeling the evolution trajectory of the state points of the battery pack in the battery energy storage power station through a deep time feature convolutional network to obtain a high-order state vector in the evolution trajectory of the state points; Step S40, designing a fuzzy rule base in combination with the feature dimension of the high-order state vector to construct a fuzzy expert state analysis system for the energy storage power station; using the fuzzy expert state analysis system for the energy storage power station to perform fusion analysis on the evolution trend of the state points of the battery pack in the battery energy storage power station, and outputting the prediction results of the health state and abnormal event information of the battery energy storage power station.

[0008] Further, in step S10, the data cleaning includes removing abnormal data and repairing missing data; preprocessing the historical data specifically includes: For the collected voltage , current , capacity and internal resistance , perform compression smoothing processing on their original values respectively, and the processing formula is as follows: ; where, represents the value after compression smoothing processing; represents the compression smoothing hyperparameter, which is adjusted according to the statistical characteristics of the monitored quantity to make obey the target distribution; represents the original value of voltage , current , capacity and internal resistance ; represents the maximum value among the original values of voltage , current , capacity and internal resistance ; For the original value of the collected temperature , the processing formula is as follows: ; where, represents the processed temperature sampling value; represents the upper threshold of temperature ; represents the weight coefficient of the temperature anomaly factor, which is adjusted and optimized according to the actual data and requirements.

[0009] Further, the specific process of step S20 includes: S201: Extract the time-frequency domain hybrid features from the preprocessed data respectively to obtain the time-domain features and frequency-domain features of each monitored quantity, and construct the multi-order feature vector ( , , , , ) through the weighted fusion of the time-domain features and frequency-domain features, where represents the voltage feature vector, represents the current feature vector, represents the capacity feature vector, represents the internal resistance feature vector, represents the temperature feature vector; Among them, the calculation formula for weighted fusion is as follows: ; Among them, represents specific monitored quantities, including voltage, current, capacity, internal resistance, and temperature; and represent weighted coefficients, and the weights of each feature type are adjusted according to actual requirements.

[0010] Furthermore, the specific process of step S20 further includes: S202: Input the multi-order feature vector ( , , , , ) into the relationship fusion module, and calculate the coupling degree parameter between each parameter in the multi-order feature vector. The calculation formula for the coupling degree parameter is: ; Among them, and represent two different parameters in the multi-order feature vector ( , , , , ), including , , , , ; represents the coupling degree parameter between parameter and parameter , briefly written as ; and represent the th element of the parameter, represents the number of elements; represents a second-order nonlinear transformation function, and the formula is , where , and represent the global sensitivity hyperparameter, local nonlinear hyperparameter, and local exponential hyperparameter, which are adjusted and optimized according to actual data and requirements, and can dynamically adjust the sensitivity of the battery parameter hysteresis effect, is the logarithmic compensation intensity factor, which is preset as a fixed constant through calibration experiments and is used to adapt to different battery systems; Based on the coupling degree parameter, calculate the first evaluation index vector , and the calculation formula is as follows: ; Among them, represents a multi-layer perceptron with 5 input channels and 1 output channel; represents the parameter and the parameter the coupling degree parameter value between them.

[0011] Furthermore, the specific process of step S20 further includes: S203: First, through the linear transformation matrices , map the first evaluation index vector and the aging degree of the batteries in the battery energy storage power station battery pack to the same dimensional space, then splice them into a second evaluation index vector by feature dimension, and finally use the second evaluation index vector as the battery pack state evaluation vector.

[0012] Furthermore, the specific process of step S30 includes: S301: Use the state evaluation vectors at several time points to construct the state evaluation space of the battery energy storage power station battery pack, where represents the state evaluation vector at time t.

[0013] Furthermore, the specific process of step S30 further includes: S302: The deep time feature convolutional network DTFCN is composed of 3 improved ResNet12s to form convolutional layers for feature extraction, and the output of each convolutional layer immediately enters the corresponding activation function NLA-AF for non-linear mapping processing; a LSTM network is connected immediately after the last ResNet12 in the DTFCN to further enhance the representation ability of long time series features; is the activation function applied by the NLA-AF at the th scale, and automatically selects a suitable mapping according to the data fluctuation characteristics. The formula of the activation function is: ; Among them, is the time scale adjustment factor at scale ; In the DTFCN, the features in the state evaluation vector at each time point are extracted through multi-layer convolution operations, and at the same time, the multi-scale non-linear adaptive activation function NLA-AF is applied to the output of each layer for non-linear mapping. The formula is: ; wherein, represents the state evaluation vector of the NLA-AF at time t corresponding output; represents the th scale of the NLA-AF; and represents the state evaluation vector at time t weights and biases of the corresponding convolutional layer.

[0014] Furthermore, the specific process of step S30 further includes: S303: Predict the high-order state vectors of the state point evolution trajectories of each battery pack in the battery energy storage power station through the DTFCN, and model the high-order state vectors. The formula is: ; wherein, represents the high-order state vector predicted for the th time step; represents the DTFCN; represents the state evaluation vector corresponding to the th time step; represents the local trajectory adaptive adjustment term, which is updated through the following mechanism: ; wherein, represents the high-order state vector predicted for the current th time step; and represent the prediction error correction hyperparameter and the state transformation inertia hyperparameter, which are adjusted and optimized according to actual data and requirements; the high-order state vector is expressed as: , wherein the comprehensive health index describes the overall health state of the battery pack; the long-term trend feature reflects the time trend of state evolution; the abnormal behavior feature is a non-linear expression for sudden changes; the dynamic coupling feature is a quantization index for the complex correlation between monitored quantities.

[0015] Furthermore, the specific process of step S40 is: S401: Design a fuzzy rule base in combination with the feature dimensions of the high-order state vector ; S402: Perform fuzzy reasoning according to the rule base and membership functions: According to the rule base and membership function, fuzzify the eigenvalues of the high-order state vector ; Conduct rule reasoning: Use fuzzy rules to reason about the evolution of all battery pack state points, and combine multiple said high-order state vectors to generate the health state and abnormal risk of the battery energy storage power station; Defuzzify: Use the weighted average method to convert the fuzzy result into a specific numerical output; S403: Output the evaluation results, including the overall health state of the battery energy storage power station and the prediction results of abnormal event information.

[0016] The present invention also proposes a battery energy storage power station state evaluation and intelligent perception system, including a data acquisition module, a data preprocessing module, a feature extraction module, a deep time feature convolutional network module, and a fuzzy expert system analysis module.

[0017] The data acquisition module is used to collect the monitoring data of each battery pack in the battery energy storage power station, including parameters such as the voltage, current, capacity, internal resistance, and temperature of the battery pack; The data preprocessing module is used to clean and normalize the monitoring data to obtain preprocessed data; The feature extraction module is used to calculate the multi-order feature vectors of each monitoring data based on the preprocessed data, and extract the complex coupling relationship between the monitoring data through a relationship fusion module to construct a state evaluation vector for quantifying the operation trend of the battery packs in the battery energy storage power station; The deep time feature convolutional network module is used to construct a state evaluation space for the battery packs in the battery energy storage power station based on the state evaluation vector, and within the state evaluation space, model the evolution trajectory of the state points of the battery packs in the battery energy storage power station through a deep learning network; The fuzzy expert system analysis module is used to perform fusion analysis on the evolution trajectories of the state points of all battery packs in the battery energy storage power station, and based on fuzzy logic reasoning, predict and output the overall health state information of the battery energy storage power station, including the health state, output power, and prediction results of abnormal events.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Improve the operation trend perception ability: The present invention comprehensively extracts the time-series features and frequency-domain features of each monitored quantity (voltage, current, capacity, internal resistance, temperature) in the battery energy storage power station through a multi-order feature vector construction method based on time-series - frequency-domain hybrid feature extraction, and captures the complex coupling relationship between each monitored quantity through a relationship fusion module, so as to achieve accurate quantification of the operation trend of the battery packs in the battery energy storage power station.

[0019] 2. Enhance the accuracy of health status assessment: By constructing a status assessment space and combining a deep temporal feature convolutional network, the present invention can comprehensively capture the dynamic evolution law of each battery pack status point in the battery energy storage power station, accurately evaluate the health status of the battery energy storage power station, and significantly improve the accuracy and robustness of status assessment.

[0020] 3. Promote intelligent operation and maintenance: By realizing intelligent perception, accurate status assessment, and operation trend prediction of the battery energy storage power station, the present invention can significantly improve the intelligent operation and maintenance level of the energy storage power station, promote the application of the "condition-based maintenance" mode, ensure the safe and efficient operation of the energy storage power station, and contribute to the development of the new power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a flowchart of the method for status assessment and intelligent perception of the battery energy storage power station of the present invention; Figure 2 is a functional block diagram of the monitoring quantity feature extraction module of the battery energy storage power station of the present invention; Figure 3 is a block diagram of the structure of the deep temporal feature convolutional network of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] The present invention proposes a method for status assessment and intelligent perception of a battery energy storage power station. The specific steps include: Step S10, collecting historical data of each monitoring quantity in the battery energy storage power station, including the voltage , current , capacity , internal resistance , and temperature of each battery pack, and performing data cleaning and preprocessing on the data to obtain preprocessed data.

[0024] Specifically, Figure 1 is a flowchart of the method for status assessment and intelligent perception of the battery energy storage power station. As Figure 1 shown, in Step S10, the historical data is cleaned by a data processing module to remove "dirty" data and repair missing data; then the historical data is preprocessed to obtain the preprocessed data, and the specific processing method is as follows: For the collected voltage , current , capacity and the internal resistance are respectively subjected to compression smoothing processing, and the processing formula is as follows: ; wherein, represents the value after compression smoothing processing; represents the compression smoothing hyperparameter, which is adjusted according to the statistical characteristics (skewness, kurtosis) of the monitored quantity (voltage / current / capacity / internal resistance) so that obeys the target distribution (such as Gaussian or uniform distribution); represents the voltage , current , capacity and the internal resistance of the original value; represents the logarithmic function with base 10; represents the voltage , current , capacity and the internal resistance of the maximum value in the original value; The empirical value range of the hyperparameter is obtained through extensive experiments, as shown in the following table: Table 1 Example of the empirical value range of the hyperparameter

[0025] Preferably, in this embodiment, the hyperparameter is dynamically adjusted within the range of the above table according to the data input by each monitored quantity, so that the monitored data voltage and current after compression smoothing obey the Gaussian distribution, and the capacity and internal resistance obey the uniform distribution.

[0026] For the original value of the collected temperature , the processing formula is as follows: ; wherein, represents the processed temperature sampling value; represents the upper threshold of the temperature ; represents the weight coefficient of the temperature anomaly factor, which is adjusted and optimized according to the actual data and requirements, and its optimization process follows the general hyperparameter tuning paradigm in machine learning. In this embodiment, first, according to prior experiments, the initial search interval is determined to be [0.1, 5.0]; when there are fewer abnormal practices in the temperature data, it is recommended to increase to strengthen the capture of signals near the threshold; on the contrary, if there is more data noise, then needs to be reduced to suppress overfitting.​

[0027] Collecting the core monitoring quantities of the battery energy storage power station and performing data cleaning and preprocessing can provide an accurate and reliable data basis for subsequent analysis. Eliminating abnormal data can avoid its interference with the evaluation results and ensure the authenticity and effectiveness of the data. The data after cleaning and preprocessing helps to more accurately reflect the actual operating status of the battery, provides strong support for comprehensively quantifying the battery operating status, accurately evaluating the health status, and predicting the operating trend, and improves the credibility of the entire evaluation and perception process.

[0028] Figure 2 It is a functional block diagram of the monitoring quantity feature extraction module 600 of the battery energy storage power station.

[0029] Step S20: Input the preprocessed data into the time-frequency domain hybrid feature extraction module to construct a multi-order feature vector of the monitoring quantity; extract the complex coupling relationship between the monitoring quantities in the multi-order feature vector through the relationship fusion module; generate a state evaluation vector based on the complex coupling relationship to quantify the operating situation of the battery pack of the battery energy storage power station.

[0030] The specific process of step S20 includes: S201: Extract the time-frequency domain hybrid features from the preprocessed data respectively to obtain the time domain features and frequency domain features of each monitoring quantity, and construct the multi-order feature vector ( , , , , ) through the weighted fusion of the time domain features and the frequency domain features, where represents the voltage feature vector, represents the current feature vector, represents the capacity feature vector, represents the internal resistance feature vector, represents the temperature feature vector; Among them, the calculation formula of the weighted fusion is as follows: ; Among them, represents the specific monitoring quantity, including voltage, current, capacity, internal resistance, and temperature; and represent the weighted coefficients, and the weights of each feature type are adjusted according to actual needs; here, the normalized weights are adopted: and . Preferably, in this embodiment, the weighted coefficient calculation adopts the feature importance scoring method. First, the importance scores of the time domain features are calculated using the random forest or XGBoost model And the importance scores of frequency domain features , and then allocate weights proportionally: , ; Fuse the data of each monitored quantity to obtain the multi-order feature vector ( , , , , ).

[0031] S202: Input the multi-order feature vector ( , , , , ) into the relationship fusion module, and calculate the coupling degree parameter between the parameters in the multi-order feature vector. The coupling degree parameter calculation formula is: ; Among them, and represent two different parameters in the multi-order feature vector ( , , , , ), including , , , , ; represents the coupling degree parameter between parameter and parameter , simply written as ; and represent the th element of the parameter, represents the number of elements; represents the second-order nonlinear transformation function, and the formula is , where , and represent the global sensitivity hyperparameter, local nonlinear hyperparameter and local exponential hyperparameter, which are adjusted and optimized according to actual data and requirements, and can dynamically adjust the sensitivity of the battery parameter hysteresis effect, is the logarithmic compensation intensity factor, which is preset as a fixed constant through calibration experiments for adapting different battery systems; represents the absolute value operation; Obtain the hyperparameters , and The empirical value ranges of are shown in the following table: Table 2 Hyperparameters 、 and Examples of empirical value ranges

[0032] Preferably, in this embodiment, the hyperparameters 、 and are determined by the following method: (a) Select initial empirical values based on the battery type. For LFP batteries = 0.8, = 2.5, = 0.05; (b) Use the gradient descent method with regularization for joint optimization. The objective function is the mean square error between the predicted coupling degree and the measured value; (c) Use the grid search method to perform joint search within the range of the above table. The 、 and values corresponding to the optimal coupling degree are the final parameter values.

[0033] Based on the coupling degree parameter, calculate the first evaluation index vector , and the calculation formula is as follows: ; where represents a multi-layer perceptron with 5 input channels and 1 output channel; represents the coupling degree parameter value between the parameter and the parameter .

[0034] S203: First, through the linear transformation matrices 、 map the first evaluation index vector and the aging degree of the batteries in the battery energy storage power station battery pack to the same-dimensional space. Preferably, in this embodiment, the linear transformation matrices 、 are generated through the following steps: (a) Initialize the matrix based on the Xavier distribution or a pre-trained autoencoder; (b) With the goal of minimizing the battery pack state evaluation error, jointly optimize the matrix elements through the gradient descent method; (c) Use L2 regularization constraints to prevent matrix overfitting.

[0035] Then, splice them according to the feature dimension to form a second evaluation index vector , and finally use the second evaluation index vector as the battery pack state evaluation vector.

[0036] Multi-order feature extraction and complex relationship fusion can fully exploit the useful information in multi-dimensional monitoring quantities. By comprehensively extracting the features of monitoring quantities such as voltage and current and fusing the complex relationships between them, this information can be mapped to the state evaluation space to achieve a comprehensive quantification of the battery operating state. This can not only depict the battery state in more detail but also capture potential connections that are difficult to discover through single-parameter analysis, providing a richer and more valuable quantitative basis for accurately evaluating the operating situation of the battery energy storage power station and improving the comprehensiveness and accuracy of the evaluation.

[0037] Figure 3 It is a block diagram of the deep time feature convolutional network structure.

[0038] Step S30: Construct a state evaluation space based on the state evaluation vector ; In the state evaluation space , model the state point evolution trajectory of the battery pack of the battery energy storage power station through a deep time feature convolutional network to obtain the high-order state vector in the state point evolution trajectory.

[0039] The specific process of step S30 includes: S301: Use the state evaluation vectors at several time points to construct the state evaluation space of the battery pack of the battery energy storage power station , where represents the state evaluation vector at time t.

[0040] S302: The deep time feature convolutional network DTFCN consists of 3 improved ResNet12s to form a convolutional layer for feature extraction. The output of each convolutional layer immediately enters the corresponding activation function NLA-AF for non-linear mapping processing to adapt to the different fluctuation characteristics of time series data and better extract deep features.; After the last ResNet12 in the DTFCN, an LSTM network is connected in series to further enhance the representation ability of long time series features; is the activation function applied at the th scale of the NLA-AF, which automatically selects a suitable mapping according to the data fluctuation characteristics. The formula of the activation function is: ; Among them, is the time scale adjustment factor at scale ; represents the natural constant; Preferably, in this embodiment According to the 3 convolutional layers of DTFCN, it is divided into 3 scales, The initial values at each scale are set to 0.5, 1.2, and 2.5 respectively, and its dynamic adjustment strategy is: dynamically adjust according to the Hurst exponent H of the input sequence of the activation function, and control the bending degree of the output feature space. The specific formula is as follows: ; In the DTFCN, the state evaluation vector at each time point is extracted through multi-layer convolution operations The features in are extracted, and at the same time, a multi-scale non-linear adaptive activation function NLA-AF is applied to the output of each layer for non-linear mapping. The formula is: ; Among them, represents the state evaluation vector of the NLA-AF at time t corresponding output; represents the th scale of the NLA-AF; and represent the weights and biases of the convolutional layer corresponding to the state evaluation vector at time t .

[0041] S303: The high-order state vector of the evolution trajectory of each battery pack state point of the battery energy storage power station is predicted through the DTFCN, and the high-order state vector is modeled. The formula is: ; Among them, represents the high-order state vector predicted for the th time step; represents the DTFCN; represents the th time step corresponding state evaluation vector; represents the local trajectory adaptive adjustment term, which is updated through the following mechanism: ; Among them, represents the high-order state vector predicted for the current th time step; and represent the prediction error correction hyperparameter and the state transformation inertia hyperparameter, which are adjusted and optimized according to actual data and requirements; characterizes the sensitivity of the model to short-term fluctuations, and the larger the value, the stronger the online update ability of the model; It reflects the degree of dependence of the system on the inertia of historical trends. The larger the value, the better the trajectory smoothness. and need to satisfy the stability condition: and ; Based on the empirical data of the prediction scenario, the initial value range of the hyperparameters , is shown in the following table: Table 3 Initial value range examples of hyperparameters ,

[0042] Preferably, in this embodiment, the hyperparameters , are determined by the following method: (a) Select the initial value based on the prediction time scale. In this embodiment, short-term prediction is selected = 0.6, = 0.1; (b) Use the rolling time window joint optimization algorithm to jointly optimize and . The optimization formula is as follows: , ; where is the exponentially decaying weight; is the regularization hyperparameter, a non-negative real number, dynamically adjusted according to the prediction error. The initial value and dynamic adjustment rules are as follows: If the signal-to-noise ratio (SNR) of the data < 10 dB, the initial value is taken as 0.5, otherwise 0.1; When the prediction error of rises continuously for 10 steps, increase to enhance regularization. The dynamic update formula is: where = 0.1 is the adjustment rate; (c) Verify the stability of the adjustment process through the Lyapunov exponent.

[0043] The high-order state vector is expressed as: , where the comprehensive health index describes the overall health status of the battery pack; the long-term trend feature reflects the time trend of state evolution; the abnormal behavior feature ​Nonlinear expression for sudden changes; dynamic coupling characteristics Quantitative indicators for complex correlations between monitored quantities.

[0044] Using a deep time feature convolutional network to model the evolution trajectory of battery pack state points can effectively capture the changing rules of battery state over time. Through multi-layer convolutional operations and multi-scale nonlinear adaptive activation functions, deep features in the time series can be extracted to adapt to different fluctuation characteristics of the data. Combining with a long short-term memory network to enhance the representation ability of long time series features can better predict the future state changes of the battery pack. This helps to detect potential battery problems in advance, provide a scientific basis for formulating reasonable operation and maintenance strategies, and ensure the stable operation of the battery energy storage power station.

[0045] Step S40: Design a fuzzy rule base in combination with the feature dimensions of the high-order state vector, and construct a fuzzy expert state analysis system for the energy storage power station; use the fuzzy expert state analysis system for the energy storage power station to perform fusion analysis on the evolution trend of the state points of the battery pack in the battery energy storage power station, and output the prediction results of the health state and abnormal event information of the battery energy storage power station.

[0046] The specific process of step S40 is as follows: S401: Design a fuzzy rule base in combination with the feature dimensions of the high-order state vector The membership function is defined as follows: Comprehensive health index, and the membership functions are Healthy, Warning, and Fault.

[0047] Long-term trend change, and the membership functions are Stable, Rising, and Declining.

[0048] Abnormal behavior characteristics, and the membership functions are NoAnomaly, MinorAnomaly, and SevereAnomaly.

[0049] Dynamic coupling characteristics, and the membership functions are Weak, Moderate, and Strong.

[0050] Examples of fuzzy rules: Rule 1: If it is "Fault" and "SevereAnomaly", then the health state is "Fault" and the abnormal risk is "High".

[0051] Rule 2: If it is "Declining" and "Strong", then the health state is "Warning" and the abnormal risk is "Medium".

[0052] Rule 3: If it is "healthy" and "no abnormality", then the health status is "healthy" and the abnormal risk is "low".

[0053] S402: Conduct fuzzy reasoning according to the rule base and membership function. First, according to the rule base and membership function, fuzzify the eigenvalue of the high-order state vector . For example, for the value of the comprehensive health index, its membership degrees may be: Healthy: 0.8; Warning: 0.2; Fault: 0; Then, conduct rule reasoning: Use the fuzzy rules to reason about the evolution of all battery pack state points, and combine multiple said high-order state vectors to generate the health status and abnormal risk of the battery energy storage power station; Finally, conduct defuzzification: Use the weighted average method to convert the fuzzy result into a specific numerical output, including: Health status score (range 0 - 100, the higher the value, the healthier); Abnormal risk score (range 0 - 100, the higher the value, the greater the risk).

[0054] S403: Output the evaluation result. The final output of the fuzzy expert state analysis system for the energy storage power station includes: (1) The overall health status of the battery energy storage power station: Output the health score (0 - 100), and divide the health status into: Healthy (80 - 100); Warning (50 - 79); Fault (0 - 49).

[0055] (2) Possible abnormal event information: Combine the abnormal risk score to output possible abnormal events and risk levels, for example: Abnormal event 1: The comprehensive health index of a certain battery pack decreases, and the long-term trend shows attenuation.

[0056] Abnormal event 2: The dynamic coupling characteristics of a certain battery pack are abnormal, and there may be problems with temperature control or internal resistance.

[0057] Through the fuzzy expert state analysis system, fuse and analyze the evolution trend of the battery pack state points, and output the health status of the battery energy storage power station and possible abnormal event information. Based on the fuzzy rule base and reasoning mechanism, complex battery state information can be converted into intuitive health scores and abnormal risk assessments. This enables the operation and maintenance personnel to quickly understand the overall situation of the energy storage power station, timely discover potential abnormalities, conduct targeted maintenance and management, improve the intelligent operation and maintenance level of the battery energy storage power station, and reduce the operation risk.

[0058] The present invention also proposes a battery energy storage power station state evaluation and intelligent perception system, including a data acquisition module, a data preprocessing module, a feature extraction module, a deep time feature convolution network module, and a fuzzy expert system analysis module.

[0059] The data acquisition module is used to collect the monitoring data of each battery pack in the battery energy storage power station, including parameters such as the voltage, current, capacity, internal resistance and temperature of the battery pack; The data preprocessing module is used to clean and normalize the monitoring data to obtain preprocessed data; The feature extraction module is used to calculate the multi-order feature vectors of each monitoring data based on the preprocessed data, and extract the complex coupling relationships between the monitoring data through the relationship fusion module to construct a state evaluation vector for quantifying the operation status of the battery packs in the battery energy storage power station; The deep time feature convolution network module is used to construct a state evaluation space for the battery packs in the battery energy storage power station based on the state evaluation vector, and within the state evaluation space, model the evolution trajectory of the state points of the battery packs in the battery energy storage power station through a deep learning network; The fuzzy expert system analysis module is used to perform fusion analysis on the evolution trajectories of the state points of all the battery packs in the battery energy storage power station, and based on fuzzy logic reasoning, predict and output the overall health status information of the battery energy storage power station, including the health status, output power and abnormal event prediction results.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still modifications or equivalent replacements can be made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. Battery energy storage power station status assessment and intelligent perception method, characterized in that: include: Step S10, collecting historical data of various monitoring quantities in the battery energy storage power station, including the voltage of each battery pack , Current ,capacity 、Internal resistance and temperature , and perform data cleaning and preprocessing; Step S20, inputting the preprocessed data into a time series-frequency domain hybrid feature extraction module to construct a multi-order feature vector of the monitoring quantity; extracting the complex coupling relationship between the monitoring quantities in the multi-order feature vector through a relationship fusion module; generating a state evaluation vector based on the complex coupling relationship to quantify the operating status of the battery group of the battery energy storage power station; Step S30: construct a state evaluation space based on the state evaluation vector ; In the state evaluation space In the embodiment, the state point evolution trajectory of the battery pack of the battery energy storage power station is modeled by a deep time feature convolutional network to obtain a high-order state vector in the state point evolution trajectory; Step S40, designing a fuzzy rule base in combination with the characteristic dimension of the high-order state vector, and constructing a fuzzy expert state analysis system for an energy storage power station; using the fuzzy expert state analysis system for the energy storage power station to perform a fusion analysis on the state point evolution trend of the battery group in the battery energy storage power station, and outputting the health status and abnormal event information prediction results of the battery energy storage power station.

2. The battery energy storage power station status assessment and intelligent perception method according to claim 1, characterized in that: In step S10, the data cleaning includes removing abnormal data and repairing missing data; preprocessing the historical data specifically includes: The voltage collected , Current ,capacity and internal resistance The original values ​​of are compressed and smoothed respectively, and the processing formula is as follows: ； in, Indicates the value after compression and smoothing; represents the compression smoothing hyperparameter, according to the monitoring quantity The statistical characteristics of Obey the target distribution; Indicates voltage , Current ,capacity and internal resistance The original value of Indicates voltage , Current ,capacity and internal resistance The maximum value among the original values ​​of ; The temperature collected The original value of is processed as follows: ； in, Indicates the processed temperature sampling value; Indicates temperature The upper threshold of The weight coefficient of the temperature anomaly factor is adjusted and optimized according to actual data and needs.

3. The battery energy storage power station status assessment and intelligent perception method according to claim 1 is characterized in that: The specific process of step S20 includes: S201: Extracting time series-frequency domain mixed features from the preprocessed data to obtain time domain features of each monitoring quantity and frequency domain characteristics , through the weighted fusion of time domain features and frequency domain features, the multi-order feature vector is constructed ( , , , , ),in, represents the voltage eigenvector, represents the current eigenvector, represents the capacity feature vector, represents the internal resistance eigenvector, represents the temperature eigenvector; Among them, the calculation formula of weighted fusion is as follows: ； in, Indicates specific monitored quantities, including voltage, current, capacity, internal resistance and temperature; and Represents the weighting coefficient, which adjusts the weight of each feature type according to actual needs.

4. The battery energy storage power station status assessment and intelligent perception method according to claim 3 is characterized in that: The specific process of step S20 also includes: S202: The multi-order feature vector ( , , , , ) is input into the relationship fusion module to calculate the coupling degree parameters between the parameters in the multi-order feature vector. The coupling degree parameter calculation formula is: ； in, and Represents the multi-order eigenvector ( , , , , ) contains two different parameters, including , , , , ; Representation parameter and parameters The coupling parameter between ; and Indicates the parameter elements, Indicates the number of elements; represents the second-order nonlinear transformation function, and the formula is ,in , and Represents global sensitivity hyperparameters, local nonlinear hyperparameters, and local exponential hyperparameters, which are adjusted and optimized according to actual data and requirements. and Ability to dynamically adjust the sensitivity of battery parameter hysteresis effect, is the logarithmic compensation intensity factor, which is preset as a fixed constant through calibration experiments and is used to adapt to different battery systems; Based on the coupling degree parameter, calculate the first evaluation index vector , the calculation formula is as follows: ； in, Represents a multilayer perceptron with 5 input channels and 1 output channel; Indicates the parameter and parameters The coupling parameter value between them.

5. The battery energy storage power station status assessment and intelligent perception method according to claim 4 is characterized in that: The specific process of step S20 also includes: S203: First, through the linear transformation matrix , The first evaluation indicator vector The aging degree of the batteries in the battery pack of the battery energy storage power station Map to the same dimensional space, and then concatenate into the second evaluation index vector according to the feature dimension , and finally the second evaluation index vector As the battery pack state evaluation vector.

6. The battery energy storage power station status assessment and intelligent perception method according to claim 5 is characterized in that: The specific process of step S30 includes: S301: Using state evaluation vectors at several time points Constructing the state evaluation space of the battery group of the battery energy storage power station ,in represents the state evaluation vector at time t.

7. The battery energy storage power station status assessment and intelligent perception method according to claim 6 is characterized in that: The specific process of step S30 also includes: S302: The deep temporal feature convolutional network DTFCN is composed of three improved ResNet12 convolutional layers for feature extraction, and the output of each convolutional layer immediately enters the corresponding activation function NLA-AF for nonlinear mapping processing; an LSTM network is immediately followed by the last ResNet12 in the DTFCN to further enhance the characterization capability of long time series features; The NLA-AF is The activation function applied on each scale automatically selects the appropriate mapping according to the data fluctuation characteristics. The formula of the activation function is: ； in, For scale The time scale adjustment factor under ; In the DTFCN, the state evaluation vector at each time point is extracted through multi-layer convolution operations. The features in the layer are used, and the multi-scale nonlinear adaptive activation function NLA-AF is applied to the output of each layer for nonlinear mapping. The formula is: ； in, represents the state evaluation vector of the NLA-AF at time t The corresponding output: Indicates the NLA-AF scale; and Represents the state evaluation vector at time t The corresponding convolutional layer weights and biases.

8. The battery energy storage power station status assessment and intelligent perception method according to claim 7 is characterized in that: The specific process of step S30 also includes: S303: The high-order state vector of the evolution trajectory of the state points of each battery group in the battery energy storage power station is obtained through the DTFCN prediction, and the high-order state vector is modeled. The formula is: ； in, Indicates The high-order state vector obtained by the time step prediction; represents the DTFCN; Indicates The state evaluation vector corresponding to the time step; Represents the local trajectory adaptive adjustment term, which is updated through the following mechanism: ； in, Indicates the current The high-order state vector obtained by the time step prediction; and Represents the prediction error correction hyperparameter and the state transformation inertia hyperparameter, which are adjusted and optimized according to actual data and requirements; The high-order state vector It is expressed as: , among which, the comprehensive health index Describes the overall health status of the battery pack; long-term trend characteristics Reflects the time trend of state evolution; abnormal behavior characteristics Nonlinear expression for sudden changes; dynamic coupling characteristics Quantitative indicators of complex relationships between monitored quantities.

9. The battery energy storage power station status assessment and intelligent perception method according to claim 1, characterized in that: The specific process of step S40 is as follows: S401: Combining the high-order state vector Design fuzzy rule base based on feature dimensions; S402: Perform fuzzy reasoning based on the rule base and membership function: According to the rule base and membership function, the high-order state vector Fuzzification of eigenvalues; Perform rule reasoning: use fuzzy rules to reason about the evolution of all battery group state points, and combine multiple high-order state vectors to generate the health status and abnormal risk of the battery energy storage power station; Defuzzification: Use weighted average method to convert fuzzy results into specific numerical outputs; S403: Output the evaluation results, including the overall health status of the battery energy storage power station and the abnormal event information prediction results.

10. A battery energy storage power station state assessment and intelligent perception system using the method according to any one of claims 1 to 9, comprising a data acquisition module, a data preprocessing module, a feature extraction module, a deep temporal feature convolutional network module and a fuzzy expert system analysis module, characterized in that: The data acquisition module is used to collect monitoring data of each battery pack in the battery energy storage power station, including parameters such as voltage, current, capacity, internal resistance and temperature of the battery pack; The data preprocessing module is used to clean and normalize the monitoring data to obtain preprocessed data; The feature extraction module is used to calculate the multi-order feature vectors of each monitoring data based on the pre-processed data, and extract the complex coupling relationship between the monitoring data through the relationship fusion module to construct a state evaluation vector for quantifying the operating status of the battery group of the battery energy storage power station; The deep temporal feature convolutional network module is used to construct a state evaluation space of the battery pack of the battery energy storage power station based on the state evaluation vector, and to model the state point evolution trajectory of the battery pack of the battery energy storage power station through a deep learning network in the state evaluation space; The fuzzy expert system analysis module is used to perform a fusion analysis on the state point evolution trajectories of all battery groups of the battery energy storage power station, and based on fuzzy logic reasoning, predict and output the overall health status information of the battery energy storage power station, including health status, output power and abnormal event prediction results.

Images