Machine learning-based power plant battery pack intelligent operation and maintenance system and method

Abstract

The present application relates to the technical field of battery operation and maintenance, and more particularly to a thermal power plant battery pack intelligent operation and maintenance system and method based on machine learning, which proposes the following scheme: a three-dimensional virtual model of the battery pack is constructed, equipped with a monitoring device group to collect battery monomer and battery pack operation information; the battery monomer fault state and the imbalance phenomenon in the battery pack are identified through clustering analysis and feature extraction; the abnormal state of the battery monomer and the battery pack is detected based on the identification result, a warning signal is sent for potential faults in the three-dimensional virtual model, and an operation and maintenance scheme including charge and discharge optimization and fault monomer isolation is generated. The present application can improve the operation efficiency and safety of the battery pack, prolong its service life, reduce maintenance cost, has the advantages of strong real-time, high adaptability and high intelligence, and is widely applicable to the intelligent operation and maintenance of thermal power plant battery packs.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery operation and maintenance, and particularly relates to an intelligent operation and maintenance system and method for a battery pack of a thermal power plant based on machine learning. BACKGROUND

[0002] As an important energy storage device in thermal power plants, battery packs are widely used for backup power, frequency modulation, peak regulation, and power system stabilization. However, traditional battery pack operation and maintenance methods have several shortcomings, including reliance on periodic manual inspections for operational state monitoring, delayed fault identification, and difficulty in effectively controlling imbalances between individual cells within the pack. These shortcomings can lead to decreased battery pack performance, fault propagation, and even safety incidents. Furthermore, as modern thermal power plants increasingly demand high efficiency and safety, traditional operation and maintenance methods cannot meet the real-time monitoring and precise operation and maintenance needs of large-scale complex systems.

[0003] For example, CN118655476B, an intelligent battery pack operation and maintenance monitoring method and system based on artificial intelligence, is disclosed in a Chinese patent with the publication number CN118655476B, which belongs to the field of battery monitoring. The method includes: configuring an online monitoring device through a parameter configuration interface of a host computer; the online monitoring device collects real-time operational data and environmental temperature of the battery pack, determines whether to issue an alarm, whether to perform active balancing, activation operation, and online sulfur removal; the host computer analyzes and processes the operational data and environmental temperature to obtain visual information; based on different battery pack pressure difference alarm thresholds and corresponding battery pack pressure difference historical alarm frequencies, as well as corresponding battery service life, the best battery pack pressure difference alarm threshold is obtained; the online monitoring device is configured based on the best battery pack pressure difference alarm threshold. This invention can greatly reduce the maintenance cost of manpower and material resources, and improve the safety of battery use.

[0004] The above patents all have the problems mentioned in the background: it is difficult to effectively identify single cell faults and intra-pack imbalances, leading to low operation and maintenance efficiency, decreased battery performance, and increased fault risk. To solve the above problems, the present application designs an intelligent operation and maintenance system and method for a battery pack of a thermal power plant based on machine learning. SUMMARY

[0005] The technical problem to be solved by the present application is to address the shortcomings of the prior art, and to provide an intelligent operation and maintenance system and method for a battery pack of a thermal power plant based on machine learning. A three-dimensional virtual model of the battery pack is constructed, and monitoring devices are equipped to collect operational information of the battery cells and the battery pack. Cluster analysis and feature extraction are used to identify the fault status of the battery cells and the imbalance within the battery pack. Based on the identification results, the abnormal status of the individual cells and the battery pack is detected, a warning signal is issued for potential faults in the three-dimensional virtual model, and an operation and maintenance plan is generated, including charge and discharge optimization and fault cell isolation.

[0006] To achieve the above object, the present application provides the following technical solutions:

[0007] The method comprises the following steps:

[0008] Constructing a three-dimensional virtual model of the battery pack;

[0009] According to the three-dimensional virtual model of the battery pack, a monitoring device group is equipped, wherein the monitoring device group comprises a battery cell monitoring device and a battery pack monitoring device;

[0010] According to the battery cell monitoring device, battery cell operation information is collected, and a battery cell fault state is identified according to the cell operation information;

[0011] According to the battery pack monitoring device, battery pack operation information is collected, and an imbalance phenomenon between batteries is identified according to the battery pack operation information;

[0012] According to the identification results of the two devices, the abnormal state of the battery cell and the battery pack is detected, and a warning signal is sent for a potential fault in the three-dimensional virtual model of the battery pack;

[0013] An operation and maintenance scheme is generated through an operation and maintenance strategy, wherein the operation and maintenance scheme comprises an adjustment group charge and discharge scheme and a fault cell isolation scheme.

[0014] The three-dimensional virtual model of the battery pack comprises:

[0015] Obtaining design information and current structure information of each battery in the battery pack, wherein the design information comprises technical drawings and specification information of the battery, and the current structure information comprises battery images;

[0016] Obtaining the battery main structure in the technical drawings, and according to the battery main structure and the specification information, the battery pack is materialized through modeling software to obtain a battery pack main model;

[0017] The battery images are feature extracted through image processing technology, wherein the features include geometric features and connection features of the battery;

[0018] The geometric features and connection features are feature reconstructed through multi-scale convolution downsampling, and are compensated according to a reconstruction error function, are spliced through morphological operations, are converted into point cloud data, the point cloud data is imported into a point cloud processing software for processing, is pasted on the battery main model, and a three-dimensional virtual model of the battery pack is output.

[0019] The battery cell fault state is identified, comprising:

[0020] acquire battery standard operation information, perform cluster analysis on the battery monomer operation information and the battery standard operation information, and output a cluster result, wherein the cluster result comprises an operation model, a charge-discharge behavior, and a dynamic characteristic group;

[0021] According to the cluster result, the characteristic information of the battery monomer operation information is calculated, wherein the characteristic information comprises electrochemical characteristics, thermal characteristic features, and electrical energy characteristics;

[0022] A fault identification network is constructed, and the characteristic information is taken as an input parameter of the fault identification network, the input parameter is trained through the fault identification network, and a fault state of a monomer battery is output.

[0023] The calculation of the characteristic information of the battery monomer operation information comprises:

[0024] According to the operation model, the voltage of the battery monomer under different operation conditions is extracted, the voltage deviation is calculated, the dynamic current and voltage data in the charge-discharge behavior are used to calculate the internal resistance conversion rate by using Ohm's law, the charge-discharge curve offset of the battery monomer in the charge-discharge process is analyzed, the electrochemical efficiency of the battery is evaluated, and thus the electrochemical characteristics are calculated;

[0025] According to the dynamic characteristic group, the change curve of the surface temperature of the battery monomer in the operation process is extracted, the temperature rise rate is calculated, and the time constant and the thermal diffusion rate of the temperature rise process are calculated, the heat dissipation characteristics of the battery are evaluated, and thus the thermal characteristic features are calculated;

[0026] The charge-discharge energy ratio of the battery monomer in different operation models in the cluster result is extracted, the energy conversion efficiency is evaluated, the power output characteristics of the battery under short-time high-rate discharge are extracted according to the dynamic characteristic group, the capacity utilization rate and the voltage platform attenuation trend under different discharge depths in the cluster result are analyzed, and thus the electrical energy characteristics are calculated.

[0027] The fault identification network comprises:

[0028] An input layer, which establishes a characteristic sequence according to an input parameter and calculates a fitting value of the characteristic sequence;

[0029] A hidden layer, which converts the fitting value of the characteristic sequence of the input layer to a high-dimensional space and calculates an output value of the hidden layer in the high-dimensional space;

[0030] An output layer, which is applied to output a fault state of a monomer battery of the fault identification network.

[0031] The identification of the imbalance phenomenon between the batteries comprises:

[0032] The remaining energy of the battery in the battery pack is tested, the capacity loss characteristic parameters are extracted, the state of charge SOC of each battery is calculated according to the capacity loss characteristic parameters, the SOC balance is calculated according to the state of charge SOC;

[0033] According to the battery pack operation information, the operation balance of the battery pack is calculated, wherein the operation balance includes voltage balance, power balance and thermal balance;

[0034] The SOC balance and the operation balance are subjected to min-max standardization processing to generate characteristic parameters, the imbalance level is calculated according to the characteristic parameters, and the imbalance identification result of the battery pack is output.

[0035] The state of charge SOC of each battery is calculated, including:

[0036] Under standard room temperature conditions, the battery is discharged at a current of 1 / 4C size, and the discharge is stopped when the battery voltage is equal to the discharge cutoff voltage, the data is recorded and rested;

[0037] Under standard room temperature conditions, the battery is charged at a current of 1 / 4C size, and when the battery voltage is equal to the charge cutoff voltage, it is converted to constant voltage charging, and the charging current is detected, and when the charging current is equal to the charge cutoff current, the charging is stopped, the data is recorded and rested;

[0038] Under standard room temperature conditions, the battery is discharged at a current of 1 / 4C size, and the discharge is stopped when the battery voltage is equal to the discharge cutoff voltage, the data is recorded and rested;

[0039] Under standard room temperature conditions, the battery is discharged at a current of 1 / 4C size, and the discharge is stopped when the battery voltage is equal to the discharge cutoff voltage, the data is recorded and rested;

[0040] According to the mode, a nonlinear optimization algorithm is searched to fit the recorded data, and the fitting result is optimized by a loss function, and the capacity loss characteristic parameters are output;

[0041] An SOC evaluation model is constructed, the kernel parameters and penalty factors of the SOC evaluation model are optimized by a firefly optimization algorithm, the test set is evaluated according to the optimized kernel parameters and penalty factors, and the state of charge SOC of the battery is output.

[0042] The characteristic parameters are generated, including:

[0043] The boundary parameters of the SOC balance and the operation balance after standardization are optimized according to the simulated annealing algorithm, and the characteristic parameter boundary is determined;

[0044] According to the characteristic parameter boundary, the SOC uniformity and the operation uniformity, curve fitting is carried out, the SOC uniformity and the operation uniformity are converted into signals through wavelet transform, the signals are translated, the positions of discontinuous points are changed, and the characteristic parameters are displayed in the curve and extracted.

[0045] The specific calculation formula of the imbalance level is:

[0046] ,

[0047] Among them, The imbalance level is represented by n, the total number of characteristic parameters, f represents a single characteristic parameter, The weight of the fth characteristic parameter is represented by f, The normalized value of the fth characteristic parameter is represented by f, The dynamic sensitivity coefficient of the fth characteristic parameter is represented by f, The dynamic deviation of the fth characteristic parameter is represented by f, The smoothing correction coefficient is represented by f, The total running time of the battery is represented by d, and m represents the total number of batteries, The running deviation value of the dth battery is represented by d, The standard deviation of the dth battery in the running process is represented by d, The upward rounding function is represented by d.

[0048] The intelligent operation and maintenance system for the battery pack of the thermal power plant based on machine learning, the system comprises a three-dimensional modeling module, a battery monitoring module, a data processing module and a maintenance suggestion module;

[0049] The three-dimensional modeling module is used to obtain the design information and current structure information of each battery in the battery pack, and construct a three-dimensional virtual model of the battery pack;

[0050] The battery monitoring module is used to equip a monitoring device group according to the three-dimensional virtual model of the battery pack, and collect battery operation information according to the monitoring device group;

[0051] The data processing module is used to identify the fault state of the battery monomer and the imbalance phenomenon between the batteries;

[0052] The maintenance suggestion module is used to detect the abnormal state of the monomer battery and the battery pack according to the identification results of the two devices, issue a warning signal for potential failure in the three-dimensional virtual model of the battery pack, and generate an operation and maintenance scheme.

[0053] Compared with the prior art, the beneficial effects of the present application are:

[0054] 1.The application combines real-time monitoring, intelligent analysis and optimization algorithm, realizes accurate identification of battery monomer fault state and battery pack imbalance phenomenon, and generates targeted operation and maintenance scheme, significantly improves the operation efficiency and safety of the battery pack. BRIEF DESCRIPTION OF DRAWINGS

[0055] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0056] Figure 1 A flowchart of the machine learning-based intelligent operation and maintenance method for the battery pack of the thermal power plant in embodiment 1 of the application is shown in the figure.

[0057] Figure 2 A flowchart of the battery operation information clustering analysis in embodiment 1 of the application is shown in the figure.

[0058] Figure 3 A network structure diagram of fault identification in embodiment 1 of the application is shown in the figure.

[0059] Figure 4 A module diagram of the machine learning-based intelligent operation and maintenance system for the battery pack of the thermal power plant in embodiment 2 of the application is shown in the figure. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments.

[0061] Embodiment 1:

[0062] Please refer to Figure 1 An embodiment provided by the application is a machine learning-based intelligent operation and maintenance method for the battery pack of the thermal power plant, and the specific steps are as follows:

[0063] S1: Construct a three-dimensional virtual model of the battery pack.

[0064] In this step, the three-dimensional virtual model is generated by collecting the design information and the current actual structure information of the battery pack, combined with modeling technology. Specifically, the accuracy and visualization effect of the model are ensured through point cloud processing and mapping technology, so that the operation and maintenance personnel can intuitively understand the physical structure and state of the battery pack.

[0065] S2: Equip the monitoring device group according to the three-dimensional virtual model of the battery pack.

[0066] In this step, the key monitoring positions are determined by analyzing the three-dimensional virtual model, and appropriate monitoring device groups are equipped. The monitoring device groups include battery cell monitoring devices and battery pack monitoring devices, which improve the scientificity of monitoring device arrangement, reduce redundancy and energy consumption, and improve the monitoring efficiency of the system.

[0067] S3: Collecting battery cell operation information to identify the fault state of the battery cell;

[0068] In this step, real-time operation information is obtained by battery cell monitoring devices, including voltage, current, internal resistance, and temperature data. Through clustering analysis method, the battery operation data is divided into different operation models (such as high-rate operation, low-rate operation, and static state). Further, the electrochemical characteristics, thermal characteristics, and electrical energy characteristics of the battery are extracted to identify the fault state of the battery cell, such as overcharge, overdischarge, high internal resistance, or thermal runaway. Potential faults of the battery cell are detected in advance to prevent the spread of faults to the entire battery pack.

[0069] S4: Collecting battery pack operation information to identify the imbalance between batteries;

[0070] In this step, battery pack monitoring devices are used to collect group operation data including voltage, current, temperature, etc. Combined with SOC balance and operation balance (including voltage, power, and thermal balance), the imbalance between batteries in the battery pack is analyzed. Standardization and optimization algorithms are used to further generate imbalance level results and locate problem battery cells. Avoiding local battery overloading or early degradation caused by group imbalance, prolonging the overall life of the battery pack.

[0071] S5: Detecting abnormal states of battery cells and battery packs and issuing a warning signal for potential faults in the three-dimensional virtual model of the battery pack;

[0072] S6: Generating an operation and maintenance scheme through operation and maintenance strategies;

[0073] In this step, the operation and maintenance scheme includes charge and discharge optimization strategies and fault cell isolation strategies. For example, through the equalization charge and discharge strategy to improve the performance difference between the batteries in the group, or isolate the fault cell to prevent it from affecting the normal operation of other batteries.

[0074] The specific steps of S1 are as follows:

[0075] S1.1: Obtain the design information and current structure information of each battery in the battery pack, wherein the design information includes technical drawings and specification information of the battery, and the current structure information includes battery images;

[0076] Traditional battery pack modeling often relies only on design information, ignoring the changes that may occur in the actual operation of the battery. By introducing current structure information, especially image and three-dimensional scanning data, the actual state of the battery can be reflected, and the accuracy and reliability of the model can be improved.

[0077] Specifically, the design information includes technical drawings of battery monomers, specification information, and assembly logic. These data can be extracted from technical materials provided by manufacturers or related databases, covering the physical size of the battery, the location of the connection interface, the internal structure description, and the standard operating parameters. Current structure information is collected through industrial imaging devices, mainly including appearance images of battery monomers.

[0078] S1.2: Obtain the main structure of the battery in the technical drawing, and according to the main structure of the battery and the specification information, use modeling software to visualize the battery pack and obtain the main model of the battery pack.

[0079] The main structure includes the shape of the battery, the location of the wiring terminal, and the assembly sequence, etc. Combined with the specification information (such as rated voltage, capacity, etc.), the geometric structure and electrical connection relationship of the battery pack are visualized through modeling software to generate a preliminary main model of the battery pack.

[0080] S1.3: Feature extraction of the battery image through image processing technology, where the features include geometric features and connection features of the battery.

[0081] Geometric features include the appearance of the battery, and connection features include the location of the battery wiring terminal and the state of the connector. Through edge detection, the contour and geometric boundary of the battery are extracted to correct the geometric deviation in the design model. Then the YOLO network is used to identify the location of the battery wiring terminal and the connector to ensure the accuracy of the connection features.

[0082] S1.4: Feature reconstruction of the geometric features and connection features through multi-scale convolution downsampling, compensation according to the reconstruction error function, splicing through morphological operations, converting to point cloud data, importing the point cloud data into point cloud processing software for processing, pasting on the main model of the battery, and outputting a three-dimensional virtual model of the battery pack.

[0083] Specifically, different scale convolution kernels are used to extract multi-level features, preserving global structural features while highlighting local details. Morphological algorithms are used to repair point cloud data, filling in missing areas in the sampling, and texture mapping on the surface of the point cloud model to accurately superimpose the battery image on the three-dimensional model, improving the realism of the model.

[0084] The specific steps of S3 are as follows:

[0085] S3.1: Obtain battery standard working operation information, perform cluster analysis on the battery monomer operation information and the battery standard working operation information, and output a cluster result, wherein the cluster result includes an operation model, a charge-discharge behavior, and a dynamic characteristic group;

[0086] First, real-time operation information of each battery monomer is collected by a monitoring device, including voltage, current, internal resistance, and temperature parameters. At the same time, design reference values (such as rated voltage, rated capacity, charge-discharge rate limit, and temperature range) are extracted from the battery standard working operation information. The real-time operation information is compared and normalized with the standard working operation information to eliminate the dimensional differences between the characteristics and ensure the accuracy of the subsequent analysis results.

[0087] Referring to Figure 2 The specific steps of the cluster analysis are as follows:

[0088] S3.1.1: According to the DBSCAN clustering algorithm, the battery standard working operation information is preliminarily clustered to obtain core points, and the number of core points is used to determine the number of cluster centers.

[0089] S3.1.2: The core points after DBSCAN clustering are used as the initial center clusters of the K-means clustering algorithm, and the Euclidean distances of the battery monomer operation information to each initial center cluster are calculated.

[0090] S3.1.3: According to the Euclidean distance, the cluster is divided, and the cluster average error of the initial center cluster is calculated. The center cluster is updated according to the cluster average error.

[0091] S3.1.4: Define a convergence condition to determine whether the updated center cluster meets the convergence condition. If the convergence condition is not met, continue to update the center cluster. If the convergence condition is met, output the cluster result.

[0092] Specifically, by combining the DBSCAN and K-means clustering algorithms, the battery standard working operation information is clustered and analyzed, the center cluster is iteratively optimized, and a more accurate cluster result is obtained, so that the working state of the monomer battery is better understood, which helps to improve the operation efficiency of the battery pack and reduce energy loss. DBSCAN is a density clustering algorithm used to identify closely connected data points in space and divide them into clusters. DBSCAN can be used for preliminary clustering to obtain core points. A core point is a core region that contains at least a specified number of data points within a given radius. Since each core point may represent a cluster, the number of core points can be used to preliminarily determine the number of cluster centers.

[0093] The kernel points obtained by the preliminary clustering by DBSCAN are used as the initial center clusters of the K-means clustering algorithm, and an initial cluster center position is provided for the K-means algorithm; the Euclidean distance is used for clustering division, and each working operation information is distributed to the nearest center cluster by calculating the Euclidean distance of the battery monomer operation information to each initial center cluster. The average error of the cluster refers to the average value of the Euclidean distance of all data points contained in the center, and the average error of the cluster can be used to update the initial center cluster to improve the clustering effect. According to the calculated average error of the cluster, the position of the center cluster can be adjusted to update the center cluster, and a convergence condition is defined to determine whether the updated center cluster meets the condition to determine whether to end the iteration. If the convergence condition is met, the clustering result is output, including the running model, the charge and discharge behavior and the dynamic characteristic group.

[0094] The running model is used to describe the current running state of the battery, such as static mode, high-rate charge and discharge mode, low-rate charge and discharge mode, etc., and reflects the performance of the battery under different working conditions. The charge and discharge behavior is used to describe the charge and discharge curve characteristics of the battery, such as current response time, voltage change curve slope, etc., which is used to identify whether there is an abnormality in the charge and discharge process. The dynamic characteristic group is classified based on the feature changes of the battery in different time windows (such as temperature change trend, internal resistance growth rate, etc.), and identifies potential degradation problems that may exist.

[0095] S3.2: According to the clustering result, the feature information of the battery monomer operation information is calculated, wherein the feature information includes electrochemical characteristics, thermal characteristics and electrical energy characteristics;

[0096] Specifically, the electrochemical characteristics include voltage characteristics, internal resistance characteristics and charge and discharge offset characteristics. The voltage characteristics are used to extract the voltage platform values (such as charge voltage platform and discharge voltage platform) under different working conditions in the running model group, calculate the voltage fluctuation amplitude and voltage recovery time, and evaluate the voltage stability of the battery. The internal resistance characteristics are calculated by using Ohm's law based on the dynamic current and voltage data, and the dynamic internal resistance value of the battery is further analyzed, and the internal resistance growth rate and internal resistance fluctuation range are used as the core indicators of battery aging and failure. The charge and discharge offset characteristics are used to extract the charge and discharge curve offset of the battery in combination with the data of the charge and discharge behavior group, and to evaluate the energy conversion efficiency and electrochemical activity change.

[0097] The thermal characteristics include temperature rise characteristics, thermal balance and thermal stability. The temperature rise characteristics are based on the dynamic characteristic group data, and the temperature change curve of the battery surface is analyzed to calculate the temperature rise rate and thermal diffusion rate, which is used to evaluate the heat dissipation performance of the battery. The thermal balance extracts the maximum deviation and mean deviation of the temperature of each monomer battery in the group, and analyzes whether the battery has a risk of thermal runaway. The thermal stability calculates the thermal diffusion time constant through the temperature change curve with time, and evaluates the thermal dynamic performance of the battery.

[0098] The electrical energy characteristics include energy conversion efficiency, power output characteristics and capacity utilization rate. The energy conversion efficiency extracts the ratio of charge and discharge energy in the operation model group, and evaluates the energy conversion efficiency of the battery under different rates. The power output characteristics combine the dynamic characteristics group to analyze the peak power output under the condition of short-time high-rate discharge, which is used to evaluate the transient performance of the battery. The capacity utilization rate analyzes the capacity attenuation curve based on the depth of discharge to identify the impact of deep discharge on the capacity of the battery.

[0099] The specific steps of feature information extraction are as follows:

[0100] S3.2.1: Extract the voltage of the battery monomer under different operating conditions according to the operation model, calculate the voltage deviation, calculate the internal resistance conversion rate by Ohm's law through the dynamic current and voltage data in the charge and discharge behavior, analyze the charge and discharge curve offset of the battery monomer in the charge and discharge process, and evaluate the electrochemical efficiency of the battery, thereby calculating the electrochemical characteristics. The calculation formula of electrochemical characteristics is:

[0101] ,

[0102] wherein, represents the electrochemical characteristics, represents the time-varying function of open-circuit voltage during the operation of the battery, represents the dynamic current curve of the battery during the charge and discharge process, represents the time-varying function of internal resistance, represents the voltage offset of the charge and discharge curve, and the calculation method is the difference between the average values of the charge and discharge curves, represents the average value of the open-circuit voltage, represents the conversion rate of the internal resistance, represents the average value of the internal resistance, and represents the start time and end time of the charge and discharge behavior;

[0103] Specifically, according to the operation model, real-time voltage data of the battery monomer under different operating conditions such as high rate, low rate, and static is collected. During the charging and discharging process, the curve of voltage change over time is recorded, and key parameters such as charging platform voltage, discharging platform voltage, peak voltage, and minimum voltage are extracted. The voltage deviation is calculated through these data, specifically the difference between the monomer battery voltage and the average voltage within the group. Using dynamic current and voltage data, the rate of change of internal resistance is calculated through Ohm's law, focusing on the nonlinear change of internal resistance under different current intensities. Further, the internal resistance transformation rate is combined with the voltage curve to analyze the degree of curve shift during the charging and discharging process, such as voltage hysteresis. When evaluating the electrochemical efficiency of the battery monomer, the ratio of energy lost during charging and discharging to the total input energy is combined to determine whether the electrochemical performance meets the design requirements.

[0104] Battery voltage, internal resistance, and curve shift are core parameters that reflect electrochemical characteristics, which can reveal the degree of battery aging, the uniformity of electrochemical reaction rate, and the difference in charging and discharging efficiency. Through the analysis of voltage deviation and internal resistance transformation rate, battery monomers with potential problems can be quickly identified, such as batteries that may have excessive polarization, effectively improving the fault prediction capability.

[0105] S3.2.2: According to the dynamic characteristic group, the change curve of the surface temperature of the battery monomer during operation is extracted, the temperature rise rate is calculated, and the time constant and thermal diffusion rate of the temperature rise process are calculated to evaluate the heat dissipation characteristics of the battery, thereby calculating the thermal characteristic feature, and the calculation formula of the thermal characteristic feature is:

[0106] ,

[0107] wherein, represents the thermal characteristic feature, represents the maximum temperature difference of the battery surface temperature rise, represents the temperature rise rate, represents the Laplace operator of the temperature field, represents the thermal diffusion coefficient of the battery, represents the second time derivative of temperature, C represents the specific heat capacity of the battery, represents the density of the battery, represents the heat generated by the battery, represents the heat source coefficient of the battery, represents the temperature rise influence coefficient of the battery, represents the time constant of temperature rise in the thermal diffusion process, and represent the start time and end time of the battery temperature rise process;

[0108] The thermal characteristics of a battery in operation are directly related to its safety and long-term performance, especially under high-rate discharge or high ambient temperature, the risk of thermal runaway increases significantly.

[0109] Specifically, the surface temperature data of the battery cell during operation is collected, and the temperature curve over time is recorded. Especially during the high-rate charging and discharging stage, the temperature rise rate, i.e. the rate of temperature rise, is extracted. Combined with the temperature change data, the time constant of the temperature rise process is calculated, reflecting the speed of heat diffusion from the battery surface to the surrounding environment. Further, by analyzing the distribution of temperature gradient, the heat diffusion rate is evaluated, and whether the heat dissipation capacity inside the battery is uneven is identified. Under specific environmental conditions (such as different heat dissipation designs or changes in external temperature), the fluctuation trend of the temperature rise rate and the time constant is combined to establish a thermal characteristic evaluation model for predicting the risk of thermal runaway.

[0110] Compared with the existing method relying on peak analysis of surface temperature, this method can more accurately locate the source of thermal runaway risk (such as local hot spots or insufficient heat dissipation) by comprehensive calculation of time constant and heat diffusion rate, effectively identify battery cells with poor heat dissipation, and thus prevent battery aging or failure caused by heat accumulation, improving the operation safety of the battery pack.

[0111] S3.2.3: Extract the charge-discharge energy ratio of the battery cell in different operation models in the clustering result, evaluate the energy conversion efficiency, according to the dynamic characteristic group, extract the power output characteristics of the battery under short-time high-rate discharge, analyze the capacity utilization rate and voltage platform decay trend under different discharge depths in the clustering result, and calculate the electric energy characteristics, the calculation formula of the electric energy characteristics is:

[0112]

[0113] wherein E represents the electric energy characteristics, and represents the start time and end time of the battery in the operation model, represents the voltage change function of the battery in the operation model, represents the energy conversion function of the battery during charging and discharging, represents the influence factor of discharge depth, represents the instantaneous capacity decay of the battery, represents the rated capacity of the battery, and represents the time interval of short-time high-rate discharge of the battery, represents the instantaneous peak transformation function of the battery, represents the internal resistance sensitive coefficient of the battery, represents the instantaneous resistance of the battery, represents the average power in the short-time discharge process, ​Weight factor representing short-time power characteristics.

[0114] The charge-discharge energy ratio, voltage platform change and power characteristics are the core indicators for evaluating the energy performance and dynamic response capability of the battery. These indicators not only reflect the current performance of the battery, but also can predict its future performance degradation.

[0115] Specifically, by analyzing the input energy and output energy of the battery under different operating models (such as high-rate charging and low-rate discharging) in the clustering results, the energy conversion efficiency is calculated. Under short-time high-rate discharging, the instantaneous power output characteristics are extracted, and the peak power, response time and power decay curve of the battery are analyzed to evaluate the transient performance of the battery. Further, the influence of different discharge depths on the capacity utilization and voltage platform is analyzed in combination with the clustering results, such as extracting the voltage platform drop rate and the change trend of the capacity utilization, reflecting the aging process of the battery in long-term cycling.

[0116] Compared with the traditional method which only relies on a single power curve analysis, the present method can more comprehensively reflect the long-term energy characteristics and dynamic response capability of the battery through the comprehensive evaluation of the charge-discharge ratio combined with the discharge depth. The battery cells with low energy efficiency are identified, providing a reliable basis for the charge-discharge optimization of the battery pack, and the capacity degradation trend of the battery in long-term cycling can be predicted.

[0117] S3.3: Construct a fault identification network, and input the feature information into the fault identification network as input parameters of the fault identification network, train the input parameters through the fault identification network, and output the fault state of the single battery cell.

[0118] Referring to Figure 3 , the fault identification network structure diagram of the embodiment of the present application, the fault identification network comprises:

[0119] An input layer establishes a feature sequence according to input parameters and calculates a fitting value of the feature sequence.

[0120] The input layer collects the operating information of the battery cell, including electrochemical characteristics, thermal characteristics and electrical energy characteristics, and uses a standardization processing method to normalize the feature data of different dimensions, so as to ensure that the numerical ranges of the features are consistent and eliminate the deviation caused by the range difference of the feature values. At the same time, in order to improve the data fitting capability, a time sequence sliding window mechanism is introduced, and the dynamic change of each feature is taken as a time sequence feature to form a complete feature sequence.

[0121] After the feature sequence is normalized, the least square method is further used to fit the feature sequence to generate a fitting value sequence. The fitting value is used to describe the change trend of each feature in a specific time period, and provides an optimized data basis for the calculation of the high-dimensional space of the hidden layer.

[0122] a hidden layer that converts the fitted values of the sequence of features of the input layer to a high-dimensional space in which output values of the hidden layer are calculated;

[0123] The hidden layer receives the sequence of fitted values of the input layer and maps them to a high-dimensional space through an activation function. The hidden layer adopts a stacked structure and contains multiple layers of nonlinear activation units. The specific activation function is ReLU (Rectified Linear Unit), which can capture complex feature interaction relationships while avoiding the problem of gradient disappearance.

[0124] In the high-dimensional space, the hidden layer performs deep feature mining on the input data, focusing on the coupling relationships between the following features:

[0125] Coupling of electrochemical features and thermal characteristics: such as the effect of internal resistance growth rate on temperature rise rate.

[0126] Coupling of thermal characteristics and electrical energy characteristics: such as the dynamic effect of temperature change on power output efficiency.

[0127] Joint analysis of cross-dimensional features: such as the relevance of voltage platform change and energy conversion efficiency.

[0128] Specifically, the input features are weighted using a weight matrix, and the batch normalization technique is used to improve the training efficiency and generalization ability of the model. Finally, the output values of the hidden layer serve as intermediate results for further calculations by the network, containing deep information of multi-dimensional features.

[0129] The output layer is applied to output the fault state of the single battery of the fault recognition network.

[0130] The output layer receives the output values of the hidden layer and maps the high-dimensional features to the fault state of the single battery through a classifier. The classifier uses the Softmax function to process the probabilities of each fault category, and finally outputs the fault classification result.

[0131] The fault state includes:

[0132] Normal state: the operating parameters of the single battery are within the normal range, and there is no obvious abnormality.

[0133] Minor fault: such as slight overcharge, overdischarge, or slight increase in internal resistance, which has limited impact on battery performance.

[0134] Serious fault: such as internal resistance significantly increased due to overcharge and overdischarge, voltage collapse, or risk of thermal runaway.

[0135] To improve the robustness of fault identification, the output layer adds confidence analysis to the classification results of fault states. Classification results with low confidence will be returned to the hidden layer for recalculation, and the weights of the activation function will be updated according to the confidence analysis results to ensure the reliability of the output results. At the same time, the output layer performs correlation analysis on the fault classification results and the feature sequence to generate an explanatory report of the fault cause and location.

[0136] Specifically, the fault identification network is provided with an input layer, a hidden layer and an output layer. The input layer is used to establish a feature sequence according to the input parameters of the fault identification network, accumulate the feature sequence, establish a gray differential equation, calculate the least square parameters of the gray differential equation, and calculate the fitting value of the feature sequence by discretizing the least square parameters. The hidden layer is used to convert the fitting value of the feature sequence of the input layer to a high-dimensional space through mapping according to the activation function between the input layer and the hidden layer, sum the feature sequence fitting values in the high-dimensional space after nonlinear weighting by the full Chinese matrix, and calculate the output value of the hidden layer. The output layer is used to obtain the output layer input, classify the state through the classifier and confidence analysis, and output.

[0137] The specific steps of S4 are as follows:

[0138] S4.1: Perform a residual capacity test on the batteries in the battery pack, extract a capacity loss characteristic parameter, and calculate the state of charge SOC of each battery according to the capacity loss characteristic parameter, and calculate the SOC uniformity according to the state of charge SOC;

[0139] Specifically, by performing a residual capacity test on each battery monomer, the actual capacity information during the discharge process is comprehensively collected. The specific operation is to use a standard charge and discharge process: a fixed current is used to charge and discharge the battery, the charge and discharge curve and the actual output capacity of the battery are recorded, and the actual capacity is compared with the rated capacity of the battery to extract the capacity loss characteristic parameter of the battery. At the same time, combined with the recorded discharge termination voltage, discharge depth and residual capacity, a nonlinear optimization algorithm is used to curve fit these data to further optimize the accuracy of the capacity loss characteristic parameter. According to the extracted capacity loss characteristic parameter, the SOC of each battery monomer is calculated by establishing a state of charge evaluation model. The calculation of SOC not only considers the charge and discharge curve characteristics of the current battery, but also combines the historical operation data and capacity attenuation trend of the battery to ensure the accuracy of the evaluation. Subsequently, by statistically analyzing the SOC of all monomer batteries in the battery pack, the SOC uniformity index is calculated, including the standard deviation, maximum deviation and mean deviation of the SOC. These indexes comprehensively reflect the charge state distribution of each monomer battery in the battery pack and identify possible imbalance problems.

[0140] Compared with the prior art, the method improves the accuracy of SOC calculation by combining the capacity loss characteristic parameter and the nonlinear optimization algorithm, is especially suitable for the battery pack with serious capacity attenuation after long-term operation, and solves the problem that the traditional SOC evaluation method is difficult to handle the attenuation error.

[0141] The specific steps of the state of charge SOC calculation are as follows:

[0142] S4.1.1: Discharge the battery at a current of 1 / 4C size under standard room temperature conditions, stop discharging when the battery voltage is equal to the discharge cutoff voltage, record the data and stand still;

[0143] S4.1.2: Charge the battery at a current of 1 / 4C size under standard room temperature conditions, when the battery voltage is equal to the charge cutoff voltage, change to constant voltage charging, and detect the charging current, when the charging current is equal to the charge cutoff current, stop charging, record the data and stand still;

[0144] S4.1.3: Discharge the fully charged battery at a current of 1 / 4C size under standard room temperature conditions, stop discharging when the battery voltage is equal to the discharge cutoff voltage, record the data and stand still;

[0145] S4.1.4: Charge the discharged battery at a current of 1 / 4C size under standard room temperature conditions, stop charging when the battery voltage is equal to the charge cutoff voltage, record the data and stand still;

[0146] S4.1.5: According to the mode search nonlinear optimization algorithm, the recorded data is curve fitted, the fitting result is optimized by the loss function, and the capacity loss characteristic parameter is output, and the calculation formula of the loss function is:

[0147] ,

[0148] Wherein, L represents the loss function, i represents the measurement time point in the residual energy test, J represents the total measurement time point in the residual energy test, G{•} represents the annealing genetic function, represents the Boltzmann constant, represents the mode search output function, represents the positive electrode potential, represents the positive electrode initial conversion rate, represents the current at the i th measurement time point, represents the positive electrode initial capacity, represents the negative electrode potential, represents the negative electrode initial conversion rate, represents the negative electrode initial capacity, represents the model parameter matrix, represents the voltage at the i th measurement time point;

[0149] The collected voltage, current and capacity data are curve fitted using a pattern search nonlinear optimization algorithm, and the fitted curve describes the nonlinear relationship between SOC and voltage. The fitting results are optimized by designing a loss function to minimize the fitting error, and finally the capacity loss characteristic parameters of the battery (such as capacity decay rate, dynamic response characteristic) are obtained.

[0150] S4.1.6: Constructing the SOC evaluation model, optimizing the kernel parameters and penalty factors of the SOC evaluation model through the firefly optimization algorithm, evaluating the test set according to the optimized kernel parameters and penalty factors, and outputting the state of charge SOC of the battery.

[0151] Specifically, before evaluation, training set and test set data need to be prepared, and the data sources include battery charge-discharge cycle test results and dynamic information such as voltage, current, temperature and time collected under different working conditions. The SOC evaluation model is a nonlinear regression model, which uses support vector regression method to capture the complex nonlinear relationship between input variables and SOC. The kernel function, penalty factor and kernel parameter are set in the model. The kernel function is used to map the input variables to a high-dimensional feature space, the penalty factor is used to balance the model complexity and fitting accuracy, and to control the overfitting and underfitting phenomenon, and the kernel parameter determines the feature mapping effect in the high-dimensional space, which affects the prediction ability of the model. The firefly optimization algorithm is a population-based intelligent optimization algorithm, which simulates the behavior of fireflies searching for fire, and is used to optimize the kernel parameters and penalty factors of the SOC evaluation model. First, a group of solutions is randomly generated, each solution representing a group of kernel parameters and penalty factors. For each firefly's parameter combination, construct the SOC evaluation model, calculate the MSE value on the training set and validation set as the fitness, adjust the next flight direction of the firefly according to the position of the flame (i.e. the current optimal solution), repeat the update of the firefly position until the set iteration number or error threshold is reached, and output the optimal kernel parameters and penalty factors.

[0152] S4.2: According to the battery pack operation information, calculate the operation balance of the battery pack, wherein the operation balance includes voltage balance, power balance and thermal balance;

[0153] The battery pack monitoring device collects the operation information of the battery pack in real time, including the real-time voltage, charge-discharge current, temperature and output power of each battery monomer. Based on these collected data, the multi-dimensional balance index of the battery pack is calculated.

[0154] Voltage balance is used to analyze the real-time voltage data of all battery monomers in the group, calculate the maximum, minimum and average voltage of the battery group, and further calculate the standard deviation and deviation range of the voltage. Voltage balance can reflect the balance degree of voltage distribution in the battery group, and large voltage deviation may cause overcharge or overdischarge of part of the battery monomers.

[0155] Power balance calculates the output power of each battery monomer through real-time current and voltage data, and evaluates the power balance of the battery group according to the unevenness of power distribution (such as power deviation rate and maximum power difference). Uneven power distribution may cause some battery loads to be too heavy, accelerating their aging.

[0156] Thermal balance obtains the surface temperature data of each battery monomer, calculates the maximum temperature difference, temperature rise rate and thermal diffusion rate of the battery group. Thermal balance is used to identify heat dissipation problems in the battery group and prevent local overheating from causing thermal runaway.

[0157] Specifically, through the balance calculation of voltage, power and heat, the balance of the running state in the battery group is comprehensively evaluated. Considering the dynamic change characteristics of thermal balance, through the calculation of temperature rise rate and thermal diffusion rate, the early signals of thermal runaway can be effectively captured, thereby improving the safety evaluation ability of the running state of the battery group.

[0158] S4.3: min-max standardization processing is performed on the SOC balance and the running balance to generate feature parameters, an imbalance level is calculated according to the feature parameters, and an imbalance identification result of the battery group is output.

[0159] The generating feature parameters comprises:

[0160] According to the simulated annealing algorithm, the boundary parameters of the SOC balance and the running balance after standardization processing are optimized to determine the feature parameter boundary;

[0161] Simulated annealing algorithm is a global optimization algorithm that gradually finds the global optimal solution by simulating the energy change in the physical annealing process. In this embodiment, according to the distribution characteristics of the SOC balance and the running balance, the upper and lower limit boundaries of the feature parameters are adaptively adjusted, so as to dynamically define the effective interval of the data. This optimization method can avoid the misjudgment problem caused by fixed boundary parameters. By iteratively adjusting the boundary parameters, the influence of noise and outliers is filtered out, and the robustness and adaptability of feature extraction are improved.

[0162] Specifically, compared with the traditional fixed boundary parameter setting method, the simulated annealing algorithm can dynamically adjust the boundary range according to real-time data, significantly improving the accuracy of feature parameter calculation, especially when dealing with complex and irregular data distribution. In addition, through the combination of standardization and boundary optimization, the difference in battery pack operating state can be better captured.

[0163] According to the feature parameter boundary, SOC uniformity and operation uniformity, curve fitting is performed, the SOC uniformity and operation uniformity are converted into signals through wavelet transform, the signals are translated, the positions of discontinuous points are changed, and the feature parameters are displayed in the curve and extracted.

[0164] The fitting process uses high-order polynomial or piecewise linear fitting method to convert discrete original data into continuous curve expression form. The main purpose of curve fitting is to eliminate the nonlinear characteristics and local discontinuity in the data, so as to more intuitively reflect the change of the uniformity state of the battery pack. At the same time, in order to further analyze the correlation between the SOC uniformity and the operation uniformity, the wavelet transform technology is used to convert the fitted curve into signal form. Wavelet transform is a mathematical tool that decomposes a signal into multiple scale wavelet bases, which can effectively capture local features and abrupt points in the signal. In this embodiment, the local abrupt points and fluctuation characteristics in the SOC uniformity and operation uniformity curves are obtained, which are important manifestations of the imbalance of the battery pack. The signals after wavelet decomposition are translated to adjust the positions of the discontinuous points, so that they are presented in a prominent way in the curve. This processing method can strengthen the visualization of the imbalance state. After curve fitting and wavelet transform processing, key feature parameters are extracted from the signal. The specific extraction contents include:

[0165] Amplitude feature: extract the maximum value, minimum value and difference value of the SOC uniformity and operation uniformity signal as the main index reflecting the difference of the uniformity state.

[0166] Frequency domain feature: frequency domain analysis is performed on the signal to extract the energy distribution in a certain frequency band, which is used to identify the periodic fluctuation characteristics in the battery pack uniformity state.

[0167] Time domain feature: calculate the time distribution characteristics of the signal, including mean, variance and skewness coefficient, etc., to quantify the overall uniformity level of the battery pack operating state.

[0168] The specific calculation formula of the imbalance level is:

[0169] ,

[0170] Wherein, represents the imbalance level, n represents the total number of feature parameters, f represents a single feature parameter, a weight of the fth feature parameter, a normalized value of the fth feature parameter, a dynamic sensitivity coefficient of the fth feature parameter, a dynamic deviation degree of the fth feature parameter, a smoothing correction coefficient, a total battery running time, d represents a single battery, and m represents a total number of batteries, a running deviation value of the dth battery, a standard deviation of the dth battery in the running process, a rounding up function.

[0171] Specifically, the imbalance level can be divided into normal, slight imbalance and serious imbalance by dividing the calculation result by a pre-set threshold, and a detailed analysis report is generated by outputting the imbalance identification result of the battery pack, specifically including the imbalance level of the battery pack as a whole, the specific values of the SOC deviation, voltage deviation and temperature deviation of each single battery and the main source of the imbalance problem.

[0172] The operation and maintenance strategy is generated by intelligent analysis and optimization algorithm based on the aforementioned collected single battery running information and battery pack running information. The operation and maintenance strategy includes adjusting the group charge and discharge strategy and the fault single isolation scheme, specifically including:

[0173] By real-time monitoring and historical data analysis of the SOC (state of charge), voltage, internal resistance, temperature and other key parameters of each single battery in the battery pack, the key factors leading to imbalance in the group are identified, such as charge and discharge rate difference, uneven temperature rise and internal resistance mismatch;

[0174] Specifically, it includes charge phase optimization and discharge phase optimization;

[0175] For the detected fault single battery (such as the battery with high internal resistance, rapid capacity decay or high risk of thermal runaway), the fault single battery is physically and electronically isolated to ensure that it does not affect the normal operation of the battery pack.

[0176] After generating the charge and discharge optimization strategy and the fault isolation scheme, the scheme is intuitively presented to the operation and maintenance personnel in combination with the three-dimensional virtual model and the intelligent operation and maintenance platform, and specific execution guidance is provided.

[0177] Embodiment 2:

[0178] Please refer to Figure 4 The present application provides an embodiment: a machine learning-based intelligent operation and maintenance system for a thermal power plant battery pack, which includes a three-dimensional modeling module, a battery monitoring module, a data processing module and a maintenance suggestion module;

[0179] The three-dimensional modeling module is used to obtain the design information and current structural information of each battery in the battery pack and to construct a three-dimensional virtual model of the battery pack.

[0180] The battery monitoring module is used to equip a monitoring device group according to the three-dimensional virtual model of the battery pack, and to collect battery operation information according to the monitoring device group.

[0181] The data processing module is used to identify the fault status of individual battery cells and the imbalance between batteries.

[0182] The maintenance suggestion module is used to detect abnormal states of individual cells and battery packs based on the identification results of the two devices, issue early warning signals for potential faults in the three-dimensional virtual model of the battery pack, and generate an operation and maintenance plan.

[0183] The 3D modeling module includes:

[0184] The information acquisition unit is used to acquire battery design information and current structural information;

[0185] The feature extraction unit is used to process the battery image and extract the battery's geometric and connectivity features.

[0186] The modeling unit is used to reconstruct battery features using multi-scale convolution and morphological processing to generate a 3D model of a single battery cell.

[0187] The data processing module includes:

[0188] The fault identification unit is used to analyze the operating data of individual battery cells and identify abnormal features through clustering algorithms and feature extraction techniques.

[0189] The balance analysis unit is used to calculate SOC balance and analyze the operational balance of the battery pack.

[0190] The anomaly classification unit is used to output the fault status of individual battery cells and the imbalance level of the battery pack based on the fault characteristics and balance analysis results.

[0191] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for intelligent operation and maintenance of a battery pack of a thermal power plant based on machine learning, characterized in that, The method comprises: constructing a three-dimensional virtual model of the battery pack; equipping a monitoring device group according to the three-dimensional virtual model of the battery pack, wherein the monitoring device group comprises a battery cell monitoring device and a battery pack monitoring device; collecting battery cell operation information according to the battery cell monitoring device, and identifying a battery cell fault state according to the cell operation information; collecting battery pack operation information according to the battery pack monitoring device, and identifying an imbalance phenomenon between batteries according to the battery pack operation information; detecting abnormal states of the battery cell and the battery pack according to the identification results of the two devices, and issuing a warning signal for a potential fault in the three-dimensional virtual model of the battery pack; generating an operation and maintenance scheme through an operation and maintenance strategy, wherein the operation and maintenance scheme comprises an intra-group charge-discharge adjustment scheme and a fault cell isolation scheme; the identification of the battery cell fault state comprises: obtaining battery standard working operation information, performing cluster analysis on the battery cell operation information and the battery standard working operation information, and outputting a cluster result, wherein the cluster result comprises a running model, a charge-discharge behavior, and a dynamic characteristic group; calculating feature information of the battery cell operation information according to the cluster result, wherein the feature information comprises electrochemical characteristics, thermal characteristic features, and electrical energy features; constructing a fault identification network, taking the feature information as input parameters of the fault identification network, training the input parameters through the fault identification network, and outputting a fault state of the battery cell; the calculation of the feature information of the battery cell operation information comprises: extracting the voltage of the battery cell under different operating conditions according to the running model, calculating the voltage deviation, calculating the internal resistance conversion rate by using Ohm's law through the dynamic current and voltage data in the charge-discharge behavior, analyzing the charge-discharge curve offset of the battery cell in the charge-discharge process, evaluating the electrochemical efficiency of the battery, and thus calculating the electrochemical characteristics, the calculation formula of the electrochemical characteristics being: ， wherein, represents the electrochemical signature, represents the time variation function of the open circuit voltage of the battery during operation, represents the dynamic current profile of the battery during charge and discharge, represents the time variation function of the internal resistance, represents the voltage offset of the charge and discharge profile, calculated as the difference between the average of the charge profile and the average of the discharge profile, represents the average of the open circuit voltage, represents the rate of change of the internal resistance, represents the average of the internal resistance, and represents the start time and end time of the charge and discharge behavior; extracting the surface temperature change curve of the battery cell during operation according to the dynamic characteristic group, calculating the temperature rise rate, and calculating the time constant and thermal diffusion rate of the temperature rise process to evaluate the heat dissipation characteristics of the battery, and thus calculating the thermal characteristic features, the calculation formula of the thermal characteristic features being: ， wherein, represents a thermal characteristic feature, represents a maximum temperature difference of a battery surface temperature rise, represents a temperature rise rate, represents a Laplacian of a temperature field, represents a thermal diffusivity of a battery, represents a second time derivative of temperature, C represents a specific heat capacity of a battery, represents a battery density, represents a heat generated by a battery, represents a heat source correlation coefficient of a battery, represents a temperature rise influence coefficient of a battery, represents a time constant of temperature rise in a thermal diffusion process, and represents a start time and an end time of a battery temperature rise process; extracting the charge-discharge energy ratio of different running models in the cluster result, evaluating the energy conversion efficiency, extracting the power output characteristics of the battery under short-time high-rate discharge according to the dynamic characteristic group, analyzing the capacity utilization rate and voltage platform attenuation trend under different discharge depths in the cluster result, and thus calculating the electrical energy features, the calculation formula of the electrical energy features being: ， wherein E represents an electrical energy characteristic, and denotes the start time and end time of the battery in the operation model, denotes a voltage change function of the battery in the operation model, denotes an energy conversion function in the charging and discharging process of the battery, denotes an influence factor of the depth of discharge, denotes an instantaneous capacity attenuation of the battery, denotes a rated capacity of the battery, and denotes a time interval of short-time high-rate discharging of the battery, denotes an instantaneous peak conversion function of the battery, denotes a battery internal resistance sensitive coefficient, denotes an instantaneous resistance of the battery, denotes an average power in the short-time discharging process, denotes a weight factor of the short-time power characteristic.

2. The method of claim 1, wherein the method further comprises: the construction of the three-dimensional virtual model of the battery pack comprises: obtaining design information and current structure information of each battery in the battery pack, wherein the design information comprises technical drawings and specification information of the battery, and the current structure information comprises battery images; obtaining the main structure of the battery in the technical drawings, and visualizing the battery pack through modeling software according to the main structure of the battery and the specification information to obtain a main model of the battery pack; The battery image is subjected to feature extraction by image processing technology, wherein the features include geometric features and connection features of the battery; The geometric features and the connection features are subjected to feature reconstruction by multi-scale convolution downsampling, compensation is performed according to a reconstruction error function, splicing is performed by morphological operation, the point cloud data is converted into point cloud processing software for processing, and a map is pasted on the battery pack main body model to output a three-dimensional virtual model of the battery pack.

3. The intelligent operation and maintenance method for battery packs in thermal power plants based on machine learning according to claim 1, characterized in that, The fault identification network comprises: an input layer, which establishes a feature sequence according to input parameters and calculates a fitting value of the feature sequence; a hidden layer, which converts the fitting value of the feature sequence of the input layer to a high-dimensional space and calculates an output value of the hidden layer in the high-dimensional space; an output layer, which is applied to output a fault state of a single battery of the fault identification network.

4. The method of claim 1, wherein the method further comprises: The identification of the imbalance between the batteries comprises: energy remaining test is performed on the batteries in the battery pack, a capacity loss feature parameter is extracted, a state of charge (SOC) of each battery is calculated according to the capacity loss feature parameter, and SOC uniformity is calculated according to the state of charge (SOC); operation uniformity of the battery pack is calculated according to the battery pack operation information, wherein the operation uniformity includes voltage uniformity, power uniformity and thermal uniformity; min-max standardization processing is performed on the SOC uniformity and the operation uniformity to generate a feature parameter, an imbalance level is calculated according to the feature parameter, and an imbalance identification result of the battery pack is output.

5. The method of claim 4, wherein the method further comprises: The calculation of the state of charge (SOC) of each battery comprises: discharge is performed on the battery at a current of 1 / 4C under standard room temperature conditions, the discharge is stopped when the battery voltage is equal to the discharge cutoff voltage, data is recorded and the battery is left to stand; charge is performed on the battery at a current of 1 / 4C under standard room temperature conditions, the charge is changed to constant voltage charging when the battery voltage is equal to the charge cutoff voltage, the charging current is detected, the charge is stopped when the charging current is equal to the charge cutoff current, data is recorded and the battery is left to stand; discharge is performed on the fully charged battery at a current of 1 / 4C under standard room temperature conditions, the discharge is stopped when the battery voltage is equal to the discharge cutoff voltage, data is recorded and the battery is left to stand; charge is performed on the discharged battery at a current of 1 / 4C under standard room temperature conditions, the charge is stopped when the battery voltage is equal to the charge cutoff voltage, data is recorded and the battery is left to stand; a curve fitting is performed on the recorded data according to a pattern search nonlinear optimization algorithm, a fitting result is optimized by a loss function, and a capacity loss feature parameter is output; an SOC evaluation model is constructed, kernel parameters and a penalty factor of the SOC evaluation model are optimized by a firefly optimization algorithm, a test set is evaluated according to the optimized kernel parameters and the penalty factor, and a state of charge (SOC) of the battery is output.

6. The method of claim 5, wherein the method further comprises: The generation of the feature parameter comprises: boundary parameter optimization is performed on the standardized SOC uniformity and the operation uniformity according to a simulated annealing algorithm to determine a feature parameter boundary. According to the characteristic parameter boundary, the SOC uniformity and the operation uniformity, curve fitting is performed, the SOC uniformity and the operation uniformity are converted into signals through wavelet transform, the signals are translated, the positions of discontinuous points are changed, the discontinuous points are displayed in the curve, and characteristic parameters are extracted.

7. The method of claim 6, wherein the method further comprises: The specific calculation formula of the imbalance level is: , wherein, denotes an unbalance level, n denotes a total number of characteristic parameters, f denotes a single characteristic parameter, denotes a weight of the fth characteristic parameter, denotes a normalized value of the fth characteristic parameter, denotes a dynamic sensitivity coefficient of the fth characteristic parameter, denotes a dynamic deviation degree of the fth characteristic parameter, denotes a smoothing correction coefficient, denotes a total sum of battery runtimes, d denotes a single battery, m denotes a total number of batteries, denotes a runtime deviation value of the dth battery, denotes a standard deviation of the dth battery during runtime, denotes a ceiling function.

8. The intelligent operation and maintenance system for the battery bank of the thermal power plant based on machine learning, which is implemented based on the intelligent operation and maintenance method for the battery bank of the thermal power plant based on machine learning according to any one of claims 1-7, characterized in that, The system comprises a three-dimensional modeling module, a battery monitoring module, a data processing module and a maintenance suggestion module. The three-dimensional modeling module is configured to acquire design information and current structure information of each battery in the battery pack, and construct a three-dimensional virtual model of the battery pack. The battery monitoring module is configured to equip a monitoring device group according to the three-dimensional virtual model of the battery pack, and acquire battery operation information according to the monitoring device group. The data processing module is configured to identify a fault state of a battery monomer and an imbalance phenomenon between batteries. The maintenance suggestion module is configured to detect an abnormal state of the battery monomer and the battery pack according to the identification results of the two devices, issue a warning signal for a potential fault in the three-dimensional virtual model of the battery pack, and generate an operation and maintenance scheme.

Images