Model construction method for multi-scene state monitoring of multi-system energy storage battery

Abstract

The invention discloses a model construction method for multi-system energy storage battery multi-scene state monitoring, and relates to the technical field of battery state monitoring, and the method comprises the steps: collecting the multi-source data of a battery; preprocessing the battery real-time operation data and the battery historical fault data to obtain a battery data set; establishing a battery failure database; carrying out deep learning by using the real-time operation data of the battery, and constructing a battery fault diagnosis initial model; optimizing the battery fault diagnosis initial model to generate a battery fault diagnosis optimization model; operation data of the battery are received in real time, and online monitoring and fault diagnosis are carried out. The technical problem of low battery reliability caused by inaccurate multi-scene state monitoring and poor fault diagnosis effect of the multi-system energy storage battery in the prior art is solved, and accurate monitoring and effective fault diagnosis and early warning of the multi-scene state of the multi-system energy storage battery are achieved. And the safety and the reliability of the operation of the battery system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status monitoring, and in particular to a model building method for multi-scenario status monitoring of multi-system energy storage batteries. Background Art

[0002] In the context of the global active promotion of energy structure transformation, energy storage battery technology plays a key role in improving energy utilization efficiency and ensuring the stability of energy supply. Among them, energy storage batteries of different systems such as lithium-ion batteries, lead-acid batteries, and sodium-ion batteries play irreplaceable advantages in different application scenarios with their unique material properties, electrochemical principles, and structural designs. However, in the actual operation process, the state of energy storage batteries is affected by many factors such as the battery's own characteristics, external environment, and complex working conditions, which brings huge challenges to its state monitoring and management. On the one hand, the real-time operation data of the battery reflects the working status of the battery in real time, and its subtle changes may indicate potential fault hazards of the battery. For example, a slight fluctuation in voltage may indicate a change in the microstructure of the electrode material, an abnormal change in current may be related to an imbalance in the electrochemical reaction inside the battery, and a rise in temperature may be due to problems such as poor heat dissipation or internal short circuits. Inaccurate evaluation of SOC, SOH, SOE, and SOP will directly affect the battery's efficiency and life prediction. On the other hand, various alarm information contained in historical fault data, such as over-temperature alarm, over-voltage alarm, SOC abnormality, etc., contains valuable information on the patterns and causes of faults, but there is currently a lack of effective means to fully mine and utilize this information.

[0003] The existing technology has technical problems such as inaccurate multi-scenario status monitoring of multi-system energy storage batteries and poor fault diagnosis, which leads to low battery reliability. Summary of the invention

[0004] The present application provides a model construction method for multi-scenario status monitoring of multi-system energy storage batteries, which is used to solve the technical problems in the prior art of inaccurate multi-scenario status monitoring of multi-system energy storage batteries, poor fault diagnosis effect, and low battery reliability.

[0005] In view of the above problems, the present application provides a model construction method for multi-scenario state monitoring of multi-system energy storage batteries, the method comprising:

[0006] Collect multi-source data from energy storage batteries of different systems, wherein the multi-source data includes real-time operation data and historical fault data of the battery; pre-process the real-time operation data and historical fault data of the battery to obtain a real-time operation data set and a historical fault data set of the battery respectively; fuse the historical fault data set of the battery with the battery mechanism model, extract the fault data of multi-system energy storage batteries, and establish a battery failure database; adopt an autoencoder structure, use the real-time operation data of the battery for deep learning, and construct an initial model for battery fault diagnosis; obtain multiple battery fault types from the battery failure database, use a gradient descent algorithm to optimize the initial model for battery fault diagnosis, and generate a battery fault diagnosis optimization model; integrate the battery fault diagnosis optimization model into a battery management system, receive the operation data of the battery in real time, and perform online monitoring and fault diagnosis.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Collect multi-source data from different energy storage batteries, where the multi-source data includes real-time battery operation data and historical fault data; pre-process the real-time battery operation data and historical battery fault data to obtain real-time battery operation data set and historical battery fault data set respectively; fuse the historical battery fault data set with the battery mechanism model, extract the fault data of multi-system energy storage batteries, and establish a battery failure database; adopt an autoencoder structure, use the real-time battery operation data for deep learning, and build a battery fault diagnosis initial model; obtain multiple battery fault types in the battery failure database, use the gradient descent algorithm to optimize the initial battery fault diagnosis model, and generate a battery fault diagnosis optimization model; integrate the battery fault diagnosis optimization model into the battery management system, receive the battery operation data in real time, and perform online monitoring and fault diagnosis. The accurate monitoring of the multi-scenario status of multi-system energy storage batteries and the effective fault diagnosis and early warning are achieved, and the safety and reliability of the battery system operation are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A flow chart of a model building method for multi-scenario state monitoring of multi-system energy storage batteries provided in an embodiment of the present application;

[0011] Figure 2A schematic diagram of a flow chart of establishing a battery failure database in a model building method for multi-scenario status monitoring of multi-system energy storage batteries provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application provides a model building method for multi-scenario status monitoring of multi-system energy storage batteries, which is used to solve the technical problems in the prior art of inaccurate multi-scenario status monitoring of multi-system energy storage batteries, poor fault diagnosis effect, and low battery reliability.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0014] Examples, such as Figure 1 As shown, the present application provides a model construction method for multi-system energy storage battery multi-scenario state monitoring, the method comprising:

[0015] Step S100: Collect multi-source data from energy storage batteries of different systems, wherein the multi-source data includes real-time operation data and historical fault data of the batteries.

[0016] Specifically, through specially designed data acquisition networks and interfaces, rich multi-source data is widely obtained from energy storage batteries of various systems. For real-time battery operation data, high-precision sensors and data acquisition cards are used to accurately record voltage data. These voltage data reflect the real-time changes in the electrochemical potential difference inside the battery. The fluctuation of its value is closely related to the battery's charge and discharge status and health. The current data presents the current size and direction of the battery during the charge and discharge process in detail. By monitoring the current data, the battery's power output and load conditions can be analyzed. Temperature data can be obtained with the help of temperature sensors. The change in temperature during the operation of the battery has a great impact on its performance and safety. Too high or too low temperature may cause battery performance to decline or even cause failure. SOC (state of charge) data intuitively shows the proportion of the current remaining power of the battery to the total power, which is a key indicator for measuring battery endurance. SOH (state of health) data comprehensively reflects the battery's aging degree, capacity attenuation, etc., which is of great significance for evaluating the battery's service life and reliability. SOE (state of energy) data gives the energy information currently stored in the battery. SOP (state of power) data clarifies the power range that the battery can output or accept under different working conditions. At the same time, the historical fault data of the battery is also collected comprehensively. These data are records of problems that occurred during the operation of the battery in the past. The over-temperature alarm data records that the battery temperature has exceeded the normal range, which is caused by poor heat dissipation, excessive charging and discharging, or excessively high ambient temperature. The over-voltage alarm data indicates that the battery voltage has exceeded the safety threshold, causing damage to the internal structure of the battery. The SOC abnormal data covers sudden and significant changes in SOC and inaccurate SOC display. These abnormalities may indicate problems such as battery management system failure and battery cell performance differences. By collecting these real-time operation data and historical fault data, a solid data foundation has been laid for the subsequent in-depth analysis of the status of energy storage batteries and the construction of accurate monitoring models.

[0017] Step S200: pre-processing the battery real-time operation data and the battery historical fault data to obtain a battery real-time operation data set and a battery historical fault data set respectively.

[0018] Specifically, the real-time battery operation data is cleaned, each data point is carefully identified, and a reasonable threshold range is determined using data statistical analysis methods. Abnormal data that is obviously deviated from the normal range due to sensor failure, electromagnetic interference and other factors, such as instantaneous ultra-high or ultra-low voltage values, abnormal current fluctuation values, etc., are eliminated to ensure the purity and reliability of the data. After cleaning, standardization is performed. For data of different physical dimensions such as voltage, current, and temperature, standardized formulas are used to uniformly convert them into data forms with the same magnitude distribution characteristics, eliminating data analysis deviations that may be caused by dimensional differences, so that a standardized battery real-time operation data set is obtained.

[0019] For the battery historical fault data, the cleaning process is also performed first. Carefully check each record, correct the inaccurate fault time and fault type errors caused by manual recording errors and data transmission errors, delete the redundant information of duplicate records, and use a reasonable interpolation algorithm to supplement the key information that is missing due to data loss to ensure the integrity and accuracy of the data. After that, the cleaned historical fault data is standardized so that different types of fault data can reach consistency in data format, expression form, etc., so as to facilitate subsequent classification and in-depth analysis, and then construct a high-quality battery historical fault data set, providing solid data support for the construction of the multi-scenario status monitoring model of the entire multi-system energy storage battery.

[0020] Step S300: Fusing the battery historical fault data set and the battery mechanism model, extracting fault data of multi-system energy storage batteries, and establishing a battery failure database.

[0021] Specifically, the battery historical fault data set is first comprehensively analyzed and feature extracted to obtain key information such as the time when the fault occurred, the duration of the fault, and the trend of the change of fault-related parameters. Based on these information features, the thermal management model, electrochemical model or mechanical stress model are matched in a targeted manner. Taking the thermal management model as an example, after the introduction of the battery historical fault data set, the model builds a partial differential equation model of battery heat transfer based on Fourier's law of heat conduction, Newton's law of cooling, and the thermal physical parameters such as the heat capacity and thermal conductivity of each component inside the battery. The temperature anomaly data points in the fault data set are substituted into the model as boundary conditions or initial conditions, and the evolution of the temperature field distribution inside the battery over time is accurately simulated by the numerical solution method finite difference method. In the simulation process, key thermal indicators such as temperature gradient changes, heat flux density distribution, and thermal resistance changes on the heat dissipation path are analyzed, and temperature-related fault feature data are extracted, such as the coordinate range of the local overheating area, the overheating peak temperature and its corresponding battery state of charge (SOC), charging and discharging current and other operating parameters.

[0022] For the electrochemical model, based on the information related to electrochemical reactions in the battery historical fault data, such as voltage mutation, capacity drop and other data features, combined with the electrochemical reaction kinetics described by the Butler-Volmer equation and the relationship between electrode potential and ion concentration determined by the Nernst equation, an electrochemical coupling model including electrode kinetics, electrolyte ion transport and solid phase diffusion is constructed. Using the voltage and current curves in the fault data as input excitation, the nonlinear equations in the electrochemical model are iteratively solved to obtain the change rules of key electrochemical parameters such as charge transfer resistance, ion diffusion coefficient and electrode active material utilization at the interfaces of each phase in the battery during the fault process, and extract characteristic data reflecting the electrochemical fault mechanism, such as the time and amplitude of abnormal increase in electrochemical reaction polarization overpotential, and the abnormal change range of key ion concentration in the electrolyte.

[0023] In terms of mechanical stress model, based on the information reflecting mechanical structural problems in the historical battery failure data, such as battery shell deformation data, internal component displacement data, etc., combined with the elastic theory and finite element analysis method in material mechanics. The battery structure is discretized into a finite number of units, and based on the mechanical properties of each unit such as the elastic modulus and Poisson's ratio, the stress-strain relationship matrix of the battery when it is subjected to mechanical load during the charging and discharging process is constructed. The structural deformation data in the fault data is substituted into the model as the boundary constraint condition, and the stress distribution cloud map and strain distribution of each component inside the battery are solved. The key mechanical parameter data such as the stress tensor components, the direction and magnitude of the principal stress, and the strain energy density in the mechanical stress concentration area are extracted as the characteristic data of mechanical stress-related faults.

[0024] Through the above-mentioned model fusion and feature extraction process, the obtained multi-system energy storage battery fault feature data in thermal, electrical, mechanical and other aspects are integrated and standardized. Clustering algorithms, such as K-Means clustering or hierarchical clustering algorithms, are used to divide the data into different fault categories based on the similarity of fault features. Each type of fault data is given a unique coding label, and a structured database record containing fault labels, fault feature parameter vectors, fault probability statistics, and corresponding battery operating condition information is established. Finally, a comprehensive, accurate and scalable battery failure database is constructed to provide strong data support and decision-making basis for subsequent battery fault diagnosis and prediction.

[0025] Step S400: adopting an autoencoder structure, using the real-time operation data of the battery to perform deep learning, and constructing an initial model for battery fault diagnosis.

[0026] Specifically, the real-time battery operation data is subjected to deep feature extraction, focusing on the time series features in the data. Through means such as sliding window technology, the changing rules, trends, and periodicity of the data in the time series are mined, and these time series features are used as the input data of the autoencoder. The autoencoder is a neural network structure composed of an encoder and a decoder. When constructing the initial model of battery fault diagnosis, the encoder and decoder are constructed using deep learning algorithms such as the multi-layer perceptron (MLP) architecture. The encoder maps high-dimensional time series feature data to a low-dimensional potential representation space to learn the key information and internal structure in the data; the decoder attempts to reconstruct the original input data from the low-dimensional representation and optimize the network parameters by minimizing the reconstruction error. During the training process, the back propagation algorithm is used to calculate the gradient and update the network weights based on the reconstruction error. After a large number of rounds of training, the autoencoder gradually grasps the characteristic patterns of the real-time battery operation data, thereby constructing an initial battery fault diagnosis model that can perform preliminary characterization and analysis of the battery operation status, laying the foundation for subsequent accurate fault diagnosis.

[0027] Step S500: obtaining a plurality of battery fault types from the battery failure database, optimizing the initial battery fault diagnosis model using a gradient descent algorithm, and generating a battery fault diagnosis optimization model.

[0028] Specifically, a variety of different types of battery failure data are carefully selected from the constructed battery failure database. These fault types cover various abnormal conditions that may occur during the operation of the battery, such as thermal runaway, short circuit, and plate aging. Based on these rich and diverse fault data, a highly targeted training set is constructed to ensure that the training set contains data samples of various typical fault modes. Then, based on the label information corresponding to each fault type in the database, the advanced deep learning framework TensorFlow is used to input the training set into the previously constructed battery fault diagnosis model for deep training. During the training process, the gradient descent algorithm is used to calculate the error between the model output and the true label. The error measures the degree of deviation between the model prediction result and the actual fault condition. Based on the calculated error, the gradient direction of the error function relative to the model parameters (such as weights and biases) is determined. This gradient direction indicates the direction of parameter adjustment, so that the model parameters can be updated in the direction of reducing the error. At the same time, the step size is determined in combination with the preset learning rate to control the amplitude of each parameter update. Then, the error is propagated back from the output layer to each layer of the model through the back-propagation algorithm, and the model parameters such as weights and bias are accurately updated according to the gradient direction and step size. This iterative process is repeated multiple times, the model parameters are continuously adjusted, and the model is continuously optimized until the model error converges to a stable minimum value, and finally a battery fault diagnosis optimization model with excellent performance and higher diagnostic accuracy is successfully generated, enabling it to more accurately identify and diagnose various types of battery faults.

[0029] Step S600: Integrate the battery fault diagnosis optimization model into the battery management system, receive the battery operation data in real time, and perform online monitoring and fault diagnosis.

[0030] Specifically, the carefully optimized battery fault diagnosis optimization model is seamlessly integrated into the battery management system (BMS) through a software interface or data interaction protocol. As the core control unit of the battery system, the BMS is responsible for real-time monitoring of various operating states of the battery. After the integration is completed, the BMS uses its built-in data acquisition module to continuously receive real-time operating data from the battery, which covers key parameters such as voltage, current, temperature, and SOC. The BMS transmits the received real-time operating data to the integrated battery fault diagnosis optimization model in real time. Based on its pre-trained complex algorithms and precise parameters, the model instantly analyzes and processes the input real-time data, and uses the powerful pattern recognition ability of the deep learning model to quickly determine whether the current operating state of the battery is normal. Once it is detected that the data features match the pre-defined fault mode or abnormal fluctuations occur, the model immediately performs fault diagnosis to determine key information such as the fault type, severity, and possible fault location. Subsequently, the BMS will quickly take corresponding measures based on the diagnostic results output by the model. If a fault is diagnosed, it will promptly output detailed fault diagnosis results and generate a clear and unambiguous fault warning signal. This signal can be notified to relevant operators or monitoring systems through sound and light alarms, remote communications, etc., so that repair measures can be taken in time to ensure that the battery system is always in a safe, stable and efficient operating state, effectively improving the reliability and overall performance of the energy storage battery system.

[0031] In one possible implementation, Figure 2 As shown, step S300 also includes:

[0032] Step S310: using the battery historical fault data set, selecting the battery mechanism model, wherein the battery mechanism model includes an electrochemical model, a thermal management model and a mechanical stress model.

[0033] Step S320: inputting the battery historical fault data set into the battery mechanism model to generate fault prediction data.

[0034] Step S330: classify the fault data by a clustering algorithm, label different fault data and fault modes, and construct the battery failure database.

[0035] Specifically, we will deeply explore the information contained in the battery historical fault data set, and accurately select the appropriate battery mechanism model through detailed analysis of various characteristics such as fault type, fault frequency, and fault-related parameter change trends in the data set. If temperature-related abnormal records frequently appear in the historical fault data set, such as battery overheating leading to performance degradation or fault alarms, then the thermal management model will become the first choice. The thermal management model is based on an in-depth understanding of the internal heat conduction, convection heat transfer, and heat dissipation mechanisms of the battery, and comprehensively considers the thermal physical properties of battery materials such as thermal conductivity and specific heat capacity, as well as the heat generation rate of the battery under different working conditions, to construct a mathematical model that can accurately simulate the distribution and change law of the battery temperature field. When the historical failure data set shows many abnormal phenomena related to the electrochemical reaction process of the battery, such as voltage instability and rapid capacity decay, the electrochemical model becomes a key tool. Based on the principle of electrochemical kinetics, the electrochemical model deeply studies the charge transfer process between the electrode and the electrolyte, the ion diffusion process, and the phase change of the electrode material, and combines the experimental data such as the open circuit voltage characteristics and polarization curve of the battery to build a model framework that can accurately describe the dynamic process of the electrochemical reaction inside the battery, thereby revealing the intrinsic connection between the electrochemical performance of the battery and the failure. If there are many records reflecting the mechanical structural integrity problems of the battery in the historical failure data set, such as deformation of the battery shell, loosening or fracture of internal components, the mechanical stress model plays an important role. The mechanical stress model is based on the theory of material mechanics, considering the various mechanical loads that the battery bears during manufacturing, assembly and use, such as expansion force, external vibration or impact load during charging and discharging, etc., analyzes the stress-strain relationship of the battery structural material, and builds a model that can simulate the distribution and evolution of mechanical stress inside the battery, providing a theoretical basis for exploring battery failures caused by mechanical stress.

[0036] The screened battery historical fault data set is input into the selected battery mechanism model. Taking the thermal management model as an example, the battery operating current, voltage, ambient temperature and battery temperature change curve recorded in the historical fault data set are used as input parameters. The model performs complex calculations based on mathematical relationships such as heat conduction equations and heat dissipation boundary conditions to predict the distribution of internal battery temperature under different fault scenarios and the possible location and time of temperature peaks, and generate heat-related fault prediction data, such as the critical temperature threshold of battery thermal runaway under different fault conditions, the rate of increase of battery temperature when the heat dissipation system fails, etc. For the electrochemical model, the voltage and current transient data in the historical fault data set and the number of battery charge and discharge cycles are input. The model predicts the consumption rate of active substances inside the battery, the impedance change at the electrode interface, and the change trend of electrolyte concentration by solving the electrochemical kinetic equation, thereby revealing the law of battery electrochemical performance degradation and potential fault points. In the mechanical stress model, the deformation data of the battery casing, the displacement data of the internal components, and the historical data of the mechanical loads borne by the battery under different working conditions are input. The model calculates the stress concentration area inside the battery structure, the stress change curve over time, and the critical stress value that may cause structural failure and other fault prediction data, providing a basis for evaluating the reliability of the battery mechanical structure.

[0037] Clustering algorithms, such as K-means clustering algorithm or density clustering algorithm, are used to classify the generated fault prediction data. According to the similarity characteristics of the data, fault data with similar fault modes and parameter change trends are classified into the same category. For example, all fault prediction data related to thermal runaway are clustered into one category, and data related to internal short circuit of the battery are clustered into another category. After the classification is completed, for each type of fault data, combined with its corresponding fault mode, a label with a clear meaning is assigned to it. For thermal runaway fault data, the label details the triggering conditions of thermal runaway, the possible degree of harm, and the corresponding preventive measures and recommendations; for internal short circuit fault data, the label contains the possible range of short circuit location, the estimated value of short circuit resistance, and the evaluation of the impact on battery performance. At the same time, the working status information in the fault data, such as the battery charging and discharging status, SOC range, etc., as well as environmental information, such as working environment temperature and humidity, etc., are extracted, and this information is integrated into the label to generate a multi-dimensional label. In this way, the characteristics and background information of each type of fault data are described comprehensively and in detail, and a battery failure database with rich information and clear structure is constructed. This database not only provides valuable data resources for subsequent battery fault diagnosis and prediction, but also provides important reference basis for battery design optimization, safety management strategy formulation, etc., which helps to improve the overall performance and reliability of multi-system energy storage batteries, reduce the risk of failure, and ensure the safe and stable operation of energy storage systems.

[0038] In a possible implementation, step S330 further includes:

[0039] Step S331: Combine the fault data and the fault mode, identify the fault mode corresponding to each data point, and assign a label to each data point.

[0040] Step S332: extracting the working status and environmental information from the fault data, identifying the fault manifestations of the battery under different working conditions and environments, and generating a multi-dimensional label.

[0041] Step S333: Based on the combination of historical fault data sets and battery mechanism models, the conditions for the occurrence of battery faults are analyzed, and labels and fault modes are dynamically adjusted.

[0042] Step S334: construct a battery failure database based on the labeled historical fault data set.

[0043] Specifically, through feature extraction and pattern recognition technology of fault data, the unique pattern characteristics presented by each data point are carefully identified. For example, the characteristics of sudden changes in voltage data, abnormal fluctuations in current data, and sharp rises in temperature data are accurately matched with various pre-defined fault modes, so as to accurately identify the fault mode corresponding to each data point. Once the fault mode is determined, a unique and representative label is assigned to each data point. The label covers the basic information of the fault type, such as overvoltage fault, overcurrent fault, overheating fault, etc., so that the fault data can be quickly identified and classified during subsequent data processing and analysis.

[0044] Comprehensively mine the fault data to extract the working status information and environmental information. The working status information includes key parameters such as the current charge and discharge status of the battery (charging, discharging, and static), the state of charge (SOC) level, and the power output or input. These parameters reflect the operating characteristics of the battery under different working stages and load conditions. Environmental information covers environmental factors such as the temperature, humidity, and air pressure of the battery, because these environmental conditions have an important impact on the performance of the battery and the probability of failure. Combine these working status and environmental information with the fault manifestations to generate multidimensional labels. For example, a label may contain information such as "overvoltage fault-discharge state-high SOC-high temperature environment". Such multidimensional labels can more comprehensively and meticulously describe the fault characteristics of the battery under specific working conditions and environments, and provide richer information basis for subsequent fault analysis and prediction.

[0045] Based on the close combination of historical fault data sets and battery mechanism models, the conditions for the occurrence of battery failures are deeply explored. By inputting historical fault data into the battery mechanism model, the model's simulation and analysis capabilities for the internal electrochemical process, thermal management mechanism, and mechanical stress distribution of the battery are used to reveal the deep-seated causes and conditions behind the failure. For example, through electrochemical model analysis, it is found that some battery materials are prone to side reactions within a specific temperature and SOC range, resulting in battery performance degradation and failure; the thermal management model may reveal that poor heat dissipation in a high temperature environment will accelerate battery aging and failure. Based on these analysis results, dynamically adjust labels and failure modes. If new fault characteristics or fault occurrence conditions are found to be inconsistent with existing labels and modes, update and improve the label system in a timely manner, redefine the failure mode, ensure that the labels and failure modes can accurately reflect the actual situation of the battery failure, and improve the accuracy of fault diagnosis and prediction.

[0046] Based on the historical fault data set that has been labeled in the previous steps, a battery failure database is constructed. The fault data with detailed labels are organized and stored according to a certain data structure and storage method. A relational database management system is used to establish multiple data tables to store fault data, label information, fault mode descriptions, and related battery working status and environmental information. By rationally designing the database architecture, the integrity, consistency, and scalability of the data are ensured to facilitate subsequent data query, retrieval, and analysis operations. The constructed battery failure database will become a valuable resource library, providing strong data support for battery fault diagnosis, prediction, optimization design, and safety management, which will help improve the reliability and performance of the entire multi-system energy storage battery system.

[0047] In a possible implementation, step S400 further includes:

[0048] Step S410: extracting time series features from the real-time running data and using them as input data for the autoencoder, and using a deep learning algorithm to train the autoencoder model.

[0049] Step S420: Optimize the hyperparameters of the encoder model to construct the initial battery fault diagnosis model.

[0050] Specifically, the real-time operation data of the battery is carefully extracted for time series features, and professional data processing technologies such as sliding window analysis and Fourier transform are used to mine representative time series features from data sequences that continuously change over time, such as voltage, current, and temperature. For example, by calculating the mean, variance, slope change, and periodicity of the data in different time windows, these features that reflect the dynamic change trend and law of the data are extracted to form a feature vector set containing rich time series information. These feature vectors will be used as input data for the autoencoder model, with the aim of enabling the model to learn the inherent patterns and laws of the battery's operating status over time. Subsequently, the training journey of the autoencoder model is started using a deep learning algorithm. The autoencoder consists of an encoder and a decoder. During the training process, the encoder maps the high-dimensional time series feature vector to a low-dimensional potential representation space. By continuously adjusting the connection weights between neurons in the encoder, the key information of the input data can be retained as much as possible in the low-dimensional space, thereby achieving data compression and feature extraction. The decoder attempts to reconstruct the original input data from the low-dimensional representation, and reversely adjusts the parameters of the encoder and decoder by minimizing the reconstruction error. During the training process, the stochastic gradient descent optimization algorithm is used to calculate the gradient based on the reconstruction error, and the model parameters are gradually updated according to the set learning rate. After a large number of iterative training, the model can gradually grasp the timing feature patterns in the real-time operation data of the battery, thereby constructing an autoencoder model with preliminary feature learning capabilities, providing a basic framework for subsequent battery fault diagnosis.

[0051] The trained encoder model is optimized for hyperparameters, including learning rate, number of hidden layers, number of neurons, number of iterations, etc. The selection of these hyperparameters directly affects the performance of the model. Through grid search, random search or heuristic search based on model evaluation indicators, a systematic combination test is conducted within a certain range of parameter values. For example, for the learning rate, try different small values ​​(such as 0.001, 0.01, 0.1, etc.) to observe the convergence speed and final performance of the model under different learning rates; for the number of hidden layers, gradually increase or decrease the number of layers (such as 1 layer, 2 layers, 3 layers, etc.) to evaluate the model's learning ability and generalization ability for complex data patterns. After each adjustment of the hyperparameters, the model is evaluated using the validation data set, and the optimal hyperparameter combination is selected based on the evaluation results, thereby further improving the performance of the model and building a more accurate and efficient initial model for battery fault diagnosis, laying a solid foundation for subsequent model optimization and practical application.

[0052] In a possible implementation, step S500 further includes:

[0053] Step S510: Acquire data of various battery fault types from the battery failure database, construct a training set, and train the initial battery fault diagnosis model using a deep learning framework based on the label of each fault type.

[0054] Step S520: using a gradient descent algorithm to calculate the error between the output of the battery fault diagnosis model and the label, and updating parameters according to the error to generate the battery fault diagnosis optimization model.

[0055] Specifically, first, data covering a variety of battery failure types are screened out from the battery failure database to ensure that the selected data is diverse and representative, and can fully reflect the various types of faults that the battery may encounter in actual operation. These data include the change curves of key operating parameters such as voltage, current, temperature, SOC, etc. of the battery under different fault states, as well as the corresponding information such as the time of occurrence of the fault and the severity of the fault. According to the predefined labels of each fault type, these data are classified and organized to construct a data set for training, one part of which is used as a training set and the other part can be reserved as a validation set. Subsequently, with the help of advanced deep learning frameworks such as TensorFlow, the constructed training set is input into the previously constructed battery fault diagnosis model. The deep learning framework provides a wealth of tools and functions to facilitate the construction of complex neural network structures and realize an efficient training process. During the training process, the model automatically learns the intrinsic mapping relationship between different fault types and battery operating parameters based on the characteristics of the input data, and continuously adjusts the connection weights and biases between neurons within the model to improve the ability to accurately identify fault types. Through multiple iterative training on the training set, the model gradually optimizes its parameters, so that the gap between the fault diagnosis results it outputs and the true labels gradually narrows, thereby improving the accuracy and generalization ability of the model.

[0056] During the model training process, the gradient descent algorithm is used to accurately calculate the error between the output of the battery fault diagnosis model and the true label. The gradient descent algorithm determines the direction and step size of parameter adjustment based on the partial derivatives of the error function with respect to the model parameters, calculates the loss function (such as mean square error, cross entropy loss, etc.) between the model output and the label, and then calculates the gradient of the loss function with respect to the model parameters (weights and biases). According to the calculated gradient direction, the parameters are adjusted in the direction that reduces the value of the loss function, and the step size is determined by the pre-set learning rate. By repeating this process continuously, the error is calculated and the parameters are updated based on the new input data in each iteration, so that the model gradually approaches the optimal solution, thereby continuously improving the model's diagnostic accuracy for battery faults. After multiple iterative trainings, until the performance indicators of the model on the validation set (such as accuracy, recall rate, F1 value, etc.) reach a stable and satisfactory level, a battery fault diagnosis optimization model with excellent performance and accurate diagnosis is finally generated, providing a strong technical guarantee for the safe and reliable operation of the battery system.

[0057] In a possible implementation, step S520 further includes:

[0058] Step S521: Determine the gradient direction and step size of the error through the gradient descent algorithm.

[0059] Step S522: According to the gradient direction and step size, use the back propagation algorithm to update the weights and biases of the model and adjust the battery fault diagnosis model parameters.

[0060] Step S523: Repeat multiple iterations and update the model parameters according to the error until the error converges, and finally generate the battery fault diagnosis optimization model.

[0061] Specifically, the gradient descent algorithm is used to accurately determine the gradient direction and step size of the error. First, a suitable error function needs to be selected to measure the degree of difference between the model output and the true label. Taking the mean square error function as an example, for a data set containing multiple samples, the mean square error is calculated by summing the squares of the difference between the predicted value and the true value of each sample and taking the average. After calculating the error function, the gradient of the error function relative to the model parameters (weights and biases) is obtained by derivative operation. In a multi-layer neural network, due to the complex model structure involving multiple hidden layers and connections between neurons, the chain rule needs to be used to gradually calculate the gradient from the output layer to the input layer. Starting from the output layer, the gradient of the error of the output layer neurons to their input (that is, the result of the weighted sum of the output of the neurons in the previous layer and the activation function) is calculated first, and then multiplied by the partial derivative of the input to the weights and biases, and so on, back propagated to each layer. The direction of the calculated gradient vector indicates the direction in which the error increases fastest, and the goal is to minimize the error, so the opposite direction of the gradient is the parameter update direction we need. After determining the gradient direction, in order to control the amplitude of each parameter update, a step size, that is, the learning rate, needs to be set. The learning rate is a hyperparameter that needs to be set in advance before training, and its value directly affects the training effect of the model. If the learning rate is set too large, the model will skip the optimal solution when updating the parameters, and may even make the error function value larger and larger, causing the model to fail to converge; conversely, if the learning rate is set too small, the model parameter update will be very slow, and a large number of iterations will be required to achieve a good effect, which will greatly increase the training time and computing resource consumption. Therefore, it is necessary to determine a suitable learning rate through multiple experiments and tuning according to the specific model and data set.

[0062] After determining the gradient direction and step size, the back propagation algorithm is used to update the model parameters. The back propagation algorithm is based on the chain rule. Starting from the output layer of the model, the error gradient calculated by the output layer is passed back to each layer of the network layer by layer in the opposite direction of the forward propagation of the data. For each neuron in each layer, its connection weights and biases are calculated and updated according to a specific formula based on the gradient direction and step size. The weight update is the product of the current weight value minus the learning rate (step size) and the gradient of the error to the weight, and the bias update is the same. In this way, the model can adjust the connection strength (weight) between neurons in each layer and the offset (bias) of the neurons themselves by an appropriate amplitude based on error feedback, so that the model parameters are gradually optimized in the direction of reducing errors and improving performance, thereby improving the recognition accuracy and generalization ability of the battery fault diagnosis model for various fault modes.

[0063] Repeat the process of S521 and S522 for multiple iterations, and continuously update the model parameters according to the error until the error converges. In each iteration, the model calculates the error based on the new input data, then determines the gradient direction and step size, and updates the weights and biases. As the number of iterations increases, the model gradually deepens its learning of the features and patterns in the data, and the error will show a trend of gradually decreasing. In order to determine whether the error has converged, some convergence conditions need to be set. By monitoring the performance indicators of the model on the validation set (such as accuracy, recall rate, F1 value), if these indicators no longer have a significant improvement in multiple consecutive iterations, it is used as a basis for judging the convergence of the error. When the error finally converges, the model obtained at this time is an optimized battery fault diagnosis optimization model. This model has been trained with a large amount of data and multiple parameter adjustments, and can diagnose and predict battery faults more accurately, providing reliable support for the safe operation of the battery system. In practical applications, the optimization model can also be further evaluated and verified according to specific needs to ensure that it has good performance in different scenarios.

[0064] In a possible implementation, step S600 further includes:

[0065] Step S610: collect the operating data of the battery in real time, use the battery fault diagnosis optimization model to perform fault diagnosis, and if a fault is found, output the fault diagnosis result and generate a fault warning signal.

[0066] Specifically, using specialized data acquisition equipment and systems, various types of battery operation data are collected in real time with high frequency and high precision. These data cover key parameters such as voltage, current, temperature, SOC (state of charge), and form a continuous data stream in the time dimension, accurately reflecting the current working status of the battery. Subsequently, the collected real-time operation data is immediately input into the pre-built and fully trained and optimized battery fault diagnosis optimization model. Based on its complex neural network structure and precise parameter settings, the model quickly conducts in-depth analysis and processing of the input data, and uses its powerful pattern recognition capabilities to comprehensively compare the input data with various predefined fault mode characteristics in the model. If, after analysis and judgment, it is found that the characteristics of the input data match a certain fault mode, or the data shows obvious abnormal fluctuations and conforms to the potential law of fault occurrence, the model immediately outputs detailed and accurate fault diagnosis results, clearly indicating key information such as the fault type, the location of the fault (such as the specific battery cell or battery pack area), and the severity of the fault. At the same time, a clear and unambiguous fault warning signal is automatically generated. The signal can be presented in a variety of forms, such as the sound and light alarm device emitting a strong sound and light prompt so that on-site operators can detect the fault at the first time; the fault warning information can also be sent to the remote monitoring center through the communication module, so that relevant technical personnel can take corresponding measures in time, such as adjusting battery operating parameters, arranging maintenance personnel to carry out inspections, etc., so as to ensure the safe and stable operation of the battery system and effectively avoid further deterioration of the fault and cause greater losses.

[0067] In a possible implementation, step S200 further includes:

[0068] Step S210: cleaning and standardizing the real-time battery operation data and the historical battery fault data.

[0069] Step S220: extract key features from the battery real-time operation data and the battery historical fault data through a feature selection algorithm, and obtain the battery real-time operation data set and the battery historical fault data set respectively.

[0070] Specifically, for real-time battery operation data and historical fault data, data cleaning is the primary task. In terms of real-time operation data, there may be abnormal values ​​caused by sensor failure, electromagnetic interference or transmission errors, such as instantaneous voltage spikes or abnormal current fluctuations. By setting a reasonable data range threshold and combining statistical analysis methods, such as calculating the mean and standard deviation of the data, identify and eliminate those data points that are obviously deviated from the normal range. At the same time, check the consistency of the data to ensure that the same type of data collected by different sensors are logically consistent. For example, the temperature values ​​measured at different locations at the same time should be within a reasonable error range. For historical fault data, it is necessary to focus on the integrity and accuracy of the data. There may be missing or erroneous records, such as inaccurate records of the time of the fault, incorrect labeling of the fault type, etc. Compare and verify with relevant record documents or other data sources, supplement missing information, and correct erroneous data. After cleaning, standardization is performed. For different physical quantities in real-time operation data, such as voltage, current, temperature, etc., they have different magnitudes and dimensions. Use standardization methods, such as normalization or z-score standardization, to convert these data into a unified numerical range for subsequent calculations and model training. Taking normalization as an example, the voltage value is mapped to a specific interval (such as between 0 and 1) so that the data in different battery systems or different measurement environments are comparable.

[0071] Based on the cleaned and standardized data, the feature selection algorithm is used to mine the key information. For the real-time operation data of the battery, the feature selection algorithm analyzes the correlation between different parameters and their importance in characterizing the battery status. For example, the rate of change of voltage and current may better reflect the health status of the battery than its instantaneous value. By calculating the feature importance indicators, such as information gain and Gini index, the features that are key to battery fault diagnosis and performance evaluation, such as voltage change trend, current ripple coefficient, temperature change rate, etc., are screened out to construct a real-time operation data set for the battery. For historical fault data, key features are also extracted according to the needs of fault diagnosis and analysis. In addition to the basic operating parameters of the battery when the fault occurs, it may also include information such as fault duration, fault frequency, and battery working mode before the fault. After determining these key features through the feature selection algorithm, a battery historical fault data set is constructed. These data sets will provide targeted data support for the subsequent construction of battery failure databases, fault diagnosis models, etc., which will help improve the accuracy and effectiveness of the multi-scenario state monitoring model of the entire multi-system energy storage battery.

[0072] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0074] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A model construction method for multi-system energy storage battery multi-scenario status monitoring, characterized in that: The method comprises: Collect multi-source data from energy storage batteries of different systems, wherein the multi-source data includes real-time operation data and historical fault data of the batteries; Preprocessing the battery real-time operation data and the battery historical fault data to obtain a battery real-time operation data set and a battery historical fault data set respectively; The battery historical failure data set and the battery mechanism model are integrated to extract the failure data of multi-system energy storage batteries and establish a battery failure database; An autoencoder structure is adopted to perform deep learning using the real-time operation data of the battery to build an initial model for battery fault diagnosis; Acquire multiple battery fault types from the battery failure database, optimize the initial battery fault diagnosis model using a gradient descent algorithm, and generate a battery fault diagnosis optimization model; The battery fault diagnosis optimization model is integrated into the battery management system to receive the battery operation data in real time and perform online monitoring and fault diagnosis.

2. A model building method for multi-system energy storage battery multi-scenario state monitoring according to claim 1, characterized in that: The battery historical fault data set and the battery mechanism model are integrated to extract the fault data of multi-system energy storage batteries and establish a battery failure database. The method includes: Using the battery historical fault data set, selecting the battery mechanism model, wherein the battery mechanism model includes an electrochemical model, a thermal management model, and a mechanical stress model; Inputting the battery historical fault data set into the battery mechanism model to generate fault prediction data; The fault data is classified by a clustering algorithm, different fault data and fault modes are labeled, and the battery failure database is constructed.

3. A model building method for multi-system energy storage battery multi-scenario state monitoring according to claim 2, characterized in that: Labeling different fault data and fault modes to construct the battery failure database, the method comprising: Combine the fault data and fault mode, identify the fault mode corresponding to each data point, and assign a label to each data point; Extracting the working state and environment information from the fault data, identifying the fault manifestations of the battery under different working conditions and environments, and generating a multi-dimensional label; Based on the combination of historical failure data sets and battery mechanism models, the conditions for battery failures are analyzed and labels and failure modes are dynamically adjusted; A battery failure database is constructed based on the labeled historical failure data set.

4. A model building method for multi-scenario state monitoring of multi-system energy storage batteries according to claim 1, characterized in that: An autoencoder structure is adopted, and the real-time operation data of the battery is used for deep learning to construct an initial model for battery fault diagnosis. The method includes: Extracting time series features from real-time running data and using them as input data of the autoencoder, and using a deep learning algorithm to train the autoencoder model; The encoder model is hyper-parameter optimized to construct the initial battery fault diagnosis model.

5. A model building method for multi-scenario state monitoring of multi-system energy storage batteries according to claim 4, characterized in that: Acquire multiple battery fault types from the battery failure database, optimize the initial battery fault diagnosis model using a gradient descent algorithm, and generate a battery fault diagnosis optimization model. The method includes: Acquire data of various battery fault types from the battery failure database, construct a training set, and train the battery fault diagnosis preliminary model using a deep learning framework according to the label of each fault type; The error between the output of the battery fault diagnosis model and the label is calculated using a gradient descent algorithm, and parameters are updated according to the error to generate the battery fault diagnosis optimization model.

6. A model building method for multi-scenario state monitoring of multi-system energy storage batteries according to claim 5, characterized in that: The error between the output of the battery fault diagnosis model and the label is calculated using a gradient descent algorithm, and parameters are updated according to the error until the optimization process converges. The method includes: Determine the gradient direction and step size of the error through the gradient descent algorithm; According to the gradient direction and step size, using a back propagation algorithm to update the weights and biases of the model and adjust the battery fault diagnosis model parameters; Repeat multiple iterations and update the model parameters according to the error until the error converges, and finally generate the battery fault diagnosis optimization model.

7. A model building method for multi-system energy storage battery multi-scenario state monitoring according to claim 6, characterized in that: Receiving the operating data of the battery in real time and performing online monitoring and fault diagnosis, the method comprises: The operating data of the battery is collected in real time, and the battery fault diagnosis optimization model is used to perform fault diagnosis. If a fault is found, the fault diagnosis result is output and a fault warning signal is generated.

8. A model building method for multi-scenario state monitoring of multi-system energy storage batteries according to claim 1, characterized in that: Preprocessing the battery real-time operation data and the battery historical fault data to obtain a battery real-time operation data set and a battery historical fault data set, respectively, the method comprising: Cleaning and standardizing the real-time operation data of the battery and the historical fault data of the battery; By using a feature selection algorithm, key features are extracted from the real-time operation data and the historical fault data of the battery, and the real-time operation data set and the historical fault data set of the battery are obtained respectively.

Images