Battery fault monitoring and processing system and method based on deep learning

Abstract

The invention discloses a battery fault monitoring and processing system and method based on deep learning, and relates to the technical field of battery monitoring, and the method comprises the steps: collecting the multi-source data of a battery in a battery replacement cabinet in real time, and constructing the analysis characteristics of the battery according to the preprocessed multi-source data; inputting a battery health analysis model for analysis, outputting a health state score, inputting the multi-source data of the battery smaller than a preset threshold value and battery analysis characteristics into a CNN-LSTM hybrid model for real-time fault diagnosis, performing classification identification, and determining a fault category; and acquiring a state vector of the fault battery, updating a state action value function through a reinforcement learning algorithm, and selecting an optimal response action according to strategy distribution to realize adaptive processing of the fault battery. Accurate quantification of battery changes is realized through battery analysis characteristics, and the accuracy of health state scores is improved; a state action value function is updated through the constructed state vector in combination with reinforcement learning, key changes are captured, and the prediction and classification precision of a time sequence model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring, and specifically to a battery fault monitoring and processing system and method based on deep learning. Background Art

[0002] Currently, as a key component of new energy systems, the safe and stable operation of batteries has a decisive impact on the performance and lifespan of the overall system. In application scenarios such as electric vehicles, energy storage systems, and battery swapping cabinets, the health status monitoring and fault warning of batteries are particularly important. Traditional monitoring methods mainly rely on rule-based threshold judgment and simple statistical indicators, such as mean and variance, which are difficult to accurately describe the complex electrochemical and thermodynamic changes inside the battery and cannot effectively capture the abnormal characteristics that gradually evolve during the long-term use of the device.

[0003] In recent years, with the development of Internet of Things, big data, and deep learning technologies, fault monitoring and processing methods based on deep learning have gradually become a research hotspot. Existing technologies have attempted to use advanced algorithms such as convolutional neural networks, long short-term memory networks, and variational autoencoders to extract features, perform time series modeling, and detect anomalies on battery multi-source sensing data, but there are still deficiencies in data preprocessing, feature construction, and response strategy formulation. Traditional methods are difficult to simultaneously consider physical domain features such as internal resistance mapping, dynamic response characteristics, uneven monomer voltage, and temperature gradient of the battery, and these features are of great significance for accurately evaluating the health status of the battery and predicting fault risks.

[0004] In addition, existing fault diagnosis methods usually rely on a single classifier or static threshold, and it is difficult to adaptively adjust the response strategy under dynamic working conditions, resulting in response delays or misjudgments when faults occur. Summary of the Invention

[0005] Based on the above-mentioned disadvantages of the prior art, the purpose of the present invention is to provide a battery fault monitoring and processing system and method based on deep learning to solve the above technical problems.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A battery fault monitoring and processing method based on deep learning, including:

[0007] Real-time collect multi-source data of the batteries in the battery swapping cabinet, preprocess the multi-source data, and construct battery analysis features according to the preprocessed multi-source data;

[0008] Set the battery analysis features as the input, analyze the historical time series data through the battery health analysis model, and output the health status score;

[0009] Input the multi-source data and battery analysis features of batteries with health status scores less than a preset threshold into a CNN-LSTM hybrid model for real-time fault diagnosis, classify and identify battery anomalies, determine the fault category, and generate fault warning information;

[0010] Obtain the state vector of the faulty battery, update the state-action value function through a reinforcement learning algorithm in a predefined action set, and select the optimal response action according to the policy distribution to achieve adaptive processing of the faulty battery.

[0011] The present invention is further configured such that the multi-source data includes open-circuit voltage, load voltage, charging current, discharging current, charging rate, charging temperature, and cumulative depth of discharge times, and the preprocessing includes normalization processing.

[0012] The present invention is further configured such that the battery analysis features include;

[0013] Construct an internal resistance mapping feature based on the open-circuit voltage, load voltage, and discharging current;

[0014] Construct a dynamic response feature based on the charging temperature, charging current, charging rate, and cumulative depth of discharge times.

[0015] The present invention is further configured such that the calculation logic of the internal resistance mapping feature is: where, Z int is the internal resistance mapping feature, V OC is the open-circuit voltage, V L is the load voltage, I DIC is the discharging current, and α, β are power adjustment parameters;

[0016] The calculation logic of the dynamic response feature is: where, Λ is the dynamic response feature, T is the charging temperature, I CH is the charging current, R is the charging rate, U is the cumulative depth of discharge times, δ and ρ are phase adjustment parameters, and γ is a power adjustment parameter.

[0017] The present invention is further configured such that the battery health analysis model uses a long short-term memory network to perform temporal modeling on the internal resistance mapping feature and the dynamic response feature, and generates a health status score through a non-linear fusion function, H = σ(w1·Z int + w2·Λ + b), where, H is the health status score, σ(·) is the activation function, w1 and w2 are weight coefficients, and b is the bias term.

[0018] The present invention is further configured such that the CNN-LSTM hybrid model extracts the key fault fingerprints of multi-source data and battery analysis features through a convolutional neural network, performs temporal dynamic modeling on the features extracted by the convolutional neural network through a long short-term memory network, analyzes the evolution of fault patterns in the time dimension, and classifies fine-grained fault patterns for abnormal situations based on the output of the CNN-LSTM model.

[0019] The present invention is further configured such that the dimension of the state vector includes multi-source data, internal resistance mapping features, dynamic response features, and fault risk aggregation features. Among them, the fault risk aggregation feature is calculated through the monomer voltage difference, charging temperature gradient, and polarization impedance index. where, Ω F is the fault risk aggregation feature, ΔV cell is the monomer voltage difference, G T is the charging temperature gradient, Θ P is the polarization impedance index, and θ1, θ2, and θ3 are power adjustment parameters.

[0020] The present invention is further configured such that a state-action value function is constructed using a deep Q-network according to the state vector, an estimated value is initialized for each state-action pair, and iterative updates are performed using historical data.

[0021] After the state-action value function is updated, the expected value of each action is calculated according to the current state vector, and a policy distribution is constructed. The optimal response action is selected using random sampling or a greedy policy.

[0022] After the optimal response action is selected, a control signal is generated and executed to achieve adaptive processing of the faulty battery.

[0023] The present invention is further configured such that the state-action value function is: Q(s,a)′ = Q(s,a) + η[tanh(r + λ·max a′ Q(s′,a′)) - Q(s,a)], where Q(s,a)′ is the updated state-action value function, Q(s,a) is the initial state-action value function, η is the learning rate, λ is the discount factor, s is the state vector, s′ is the next state vector, a is the selected action, a′ is the next selected action, r is the immediate reward, and r = -ln(1 + Ω F (t));

[0024] The policy distribution is: where π(a∣s) is the policy distribution, κ is the temperature parameter, and a "" is all possible actions under the state vector s.

[0025] The present invention also provides a battery fault monitoring and processing system based on deep learning for implementing the above-mentioned battery fault monitoring and processing method based on deep learning. The system includes:

[0026] A data acquisition module: It collects multi-source data of the batteries in the battery swapping cabinet in real time, preprocesses the multi-source data, and constructs battery analysis features based on the preprocessed multi-source data.

[0027] A time series analysis module: It sets the battery analysis features as the input, analyzes the historical time series data through a battery health analysis model, and outputs a health status score.

[0028] A fault determination module: It inputs the multi-source data and battery analysis features of the batteries with health status scores less than a preset threshold into a CNN-LSTM hybrid model for real-time fault diagnosis, classifies and identifies battery anomalies, determines the fault category, and generates a fault warning message.

[0029] A fault processing module: It obtains the state vector of the faulty battery, updates the state-action value function through a reinforcement learning algorithm in a predefined action set, and selects the optimal response action according to the policy distribution to achieve adaptive processing of the faulty battery.

[0030] The present invention provides a battery fault monitoring and processing system and method based on deep learning. The method collects multi-source data of the batteries in the battery swapping cabinet in real time, preprocesses the multi-source data, constructs battery analysis features based on the preprocessed multi-source data, sets the battery analysis features as the input, analyzes the historical time series data through a battery health analysis model, and outputs a health status score. It inputs the multi-source data and battery analysis features of the batteries with health status scores less than a preset threshold into a CNN-LSTM hybrid model for real-time fault diagnosis, classifies and identifies battery anomalies, determines the fault category, and generates a fault warning message. It obtains the state vector of the faulty battery, updates the state-action value function through a reinforcement learning algorithm in a predefined action set, and selects the optimal response action according to the policy distribution to achieve adaptive processing of the faulty battery. The beneficial effects produced include:

[0031] 1. Through physical domain indicators such as internal resistance mapping features and dynamic response features, it fully reflects the complex changes of the battery in electrochemistry and thermodynamics, realizes accurate quantitative evaluation of the battery health status, improves the accuracy of the health status score, and provides a scientific basis for fault warning;

[0032] 2. By constructing a state vector that includes multi-source data, internal resistance mapping features, dynamic response features, and fault risk aggregation features, the description of the battery operating state becomes more comprehensive and detailed. It not only captures the key changes in the internal electrochemistry, thermodynamics, and structural degradation of the battery but also provides rich input information for the time-series model, thus significantly improving the accuracy of state-of-health prediction and fault mode recognition.

[0033] 3. Combine the reinforcement learning algorithm to construct a state-action value function to achieve an adaptive response to faulty batteries. By dynamically updating the state vector and the optimal response action, this solution can quickly take adjustment measures when a fault occurs, reduce system risks, and continuously optimize the response strategy through closed-loop feedback, thus ensuring the safe and stable operation of the battery system.

[0034] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically gives the specific implementation manners of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0036] Figure 1 is a flowchart of a method for monitoring and processing battery faults based on deep learning shown in an exemplary embodiment of the present invention;

[0037] Figure 2 is a schematic structural diagram of a system for monitoring and processing battery faults based on deep learning shown in an exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The following will describe the embodiments of the present invention with reference to the drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for explaining the present invention and not for limiting the protection scope of the present invention.

[0039] It should be noted that the illustrations provided in the following embodiments only schematically illustrate the basic concept of the present invention. Therefore, only the components related to the present invention are shown in the drawings, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and proportion of each component in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0040] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0041] Embodiment 1

[0042] A battery fault monitoring and processing method based on deep learning, as Figure 1 shown, includes:

[0043] Real-time collect multi-source data of the batteries in the battery swapping cabinet, preprocess the multi-source data, and construct battery analysis features according to the preprocessed multi-source data;

[0044] Set the battery analysis features as the input, analyze the historical time-series data through the battery health analysis model, and output the health status score;

[0045] Input the multi-source data and battery analysis features of the batteries with health status scores less than the preset threshold into the CNN-LSTM hybrid model for real-time fault diagnosis, classify and identify battery anomalies, determine the fault category, and generate a fault warning message;

[0046] Obtain the state vector of the faulty battery, update the state-action value function through the reinforcement learning algorithm in the predefined action set, and select the optimal response action according to the policy distribution to achieve the adaptive processing of the faulty battery.

[0047] The present invention is further configured such that the multi-source data includes open-circuit voltage, load voltage, charging current, discharging current, charging rate, charging temperature, and cumulative deep discharge times, and the preprocessing includes normalization processing. Specifically, the open-circuit voltage refers to the voltage value of the battery terminals under no-load conditions, reflecting the chemical stability and potential level of the battery in a static state; the load voltage refers to the terminal voltage measured when the battery bears an external load, reflecting the voltage drop generated due to the internal resistance and chemical reactions under the action of the load. In the present invention, it is measured by setting a rated load in the electrical cabinet; the charging current refers to the current value flowing into the battery during the battery charging process, and the magnitude of the charging current directly affects the charging rate and the rate of chemical reactions inside the battery; the discharging current refers to the current value flowing out during the battery discharging process, reflecting the current output of the battery under a power supply load; the charging rate refers to the charging current rate relative to the nominal capacity of the battery, which not only reflects the magnitude of the current during the charging process but also relates to the charging efficiency and heat generation. The charging temperature refers to the temperature value monitored by the built-in temperature sensor during the battery charging process. The charging temperature has an important impact on the safe operation and life of the battery because it is directly related to the molecular movement and chemical reaction rate inside the battery; the cumulative deep discharge times refer to the cumulative number of deep discharge cycles completed by the battery during use, usually related to the accumulation of discharge capacity, and is obtained by integrating the discharge depth during each discharge process, reflecting the usage load and aging degree of the battery. In the preprocessing stage, the above-mentioned raw data are normalized, and the data with different dimensions and different value ranges are uniformly converted to the same scale to eliminate the magnitude difference and systematic error between the data. The normalization processing usually adopts interval mapping or linear normalization method based on the maximum and minimum values to ensure that the subsequent deep learning model can effectively utilize each parameter information during the feature extraction and modeling process, and improve the model convergence speed and prediction accuracy.

[0048] The present invention is further configured such that the battery analysis features include;

[0049] Construct an internal resistance mapping feature based on the open-circuit voltage, load voltage, and discharging current; the present invention is further configured such that the calculation logic of the internal resistance mapping feature is: Where Z int is the internal resistance mapping feature, V OC is the open-circuit voltage, V L is the load voltage, I DIC is the discharging current, and α, β are power adjustment parameters; specifically, the above calculation logic is based on the open-circuit voltage V OC , load voltage V L and discharging current I DIC These three parameters are used to construct the internal resistance mapping feature Z int by non-linear mapping, and use V OC -V LThe reciprocal reflects the voltage drop phenomenon caused by the internal resistance of the battery under load conditions: when the internal resistance of the battery is high, the difference between the load voltage and the open-circuit voltage is small, so its reciprocal is large, and this change can be sensitively captured after being amplified by a power function; at the same time, by applying an exponential mapping to the discharge current I DIC the internal resistance effect caused by high current under the discharge state is further amplified. Finally, the results of the two parts are multiplied and then added by 1, and the natural logarithm is taken to obtain a smooth and comparable internal resistance mapping feature; the power adjustment parameter α is used to control the sensitivity of this part, and its value range is [1, 3]. The power adjustment parameter β is used to adjust the exponential mapping of the discharge current I DIC . This parameter determines the degree of non-linear amplification of the current change on the internal resistance mapping feature, and its value range is [0.5, 3]. Through the above logic, the voltage drop caused by the internal resistance of the battery under load and the dynamic response under high current conditions can be effectively captured. The generated internal resistance mapping feature is smooth and sensitive, and can be used as an important indicator to evaluate the internal health status of the battery. Through non-linear amplification and normalization processing, this feature not only improves the accuracy of fault warning, but also provides a scientific and quantitative basis for subsequent battery health analysis and fault diagnosis based on deep learning, thus helping to timely identify and handle potential battery fault risks;

[0050] Construct a dynamic response feature based on the charging temperature, charging current, charging rate, and cumulative deep discharge times; the calculation logic of the dynamic response feature is: where Λ is the dynamic response feature, T is the charging temperature, I CHLet \(I\) be the charging current, \(R\) be the charging rate, \(U\) be the cumulative deep discharge times, \(\delta\) and \(\rho\) be the phase adjustment parameters, and \(\gamma\) be the power adjustment parameter. Specifically, the above calculation logic aims to comprehensively consider multiple key factors reflecting the dynamic response of the battery during charging. By using the periodic characteristics of the charging temperature, charging current, and charging rate, the signal is corrected through the phase adjustment parameters, and then combined with the amplitude contribution of the reference charging voltage, a comprehensive index is constructed. This index is normalized by dividing by the power of the cumulative deep discharge times to offset the influence brought by the long-term use of the battery. Finally, the arctangent function is used to map the result to a bounded interval to obtain the dynamic response characteristic, ensuring that the output value is stable and convenient for subsequent deep models to use; the phase adjustment parameters \(\delta\) and \(\rho\) are respectively used to correct the phase differences of the charging temperature and charging rate signals, so that the signals are aligned in the periodic function, and the value range is \([-\pi,\pi]\); the power adjustment parameter \(\gamma\) is used to adjust the normalization effect of the cumulative deep discharge times on the dynamic response characteristic, and its role is to balance the long-term aging effect and the instantaneous response during charging, and the value range is \([0.5,2]\); by respectively performing phase correction on the charging temperature and charging rate signals and multiplying them with the reference charging voltage and charging current, this formula can carefully reflect the instantaneous changes in the battery response during charging and effectively reveal the dynamic characteristics of the internal reaction of the battery.

[0051] The present invention is further configured such that the battery health analysis model uses a long short-term memory network to perform temporal modeling on the internal resistance mapping feature and the dynamic response feature, and generates a health state score through a non-linear fusion function, \(H = \sigma(w_1\cdot Z int +w_2\cdot\Lambda + b)\), where \(H\) is the health state score, \(\sigma(\cdot)\) is the activation function, \(w_1\) and \(w_2\) are weight coefficients, and \(b\) is the bias term. The above logic uses a long short-term memory network (LSTM) to perform temporal modeling on the internal resistance mapping feature and the dynamic response feature formed during the charging and discharging process of the battery, so as to capture the dynamic characteristics of the battery health state changing with time. Subsequently, through the non-linear fusion function, these two features are fused in the form of a weighted sum, and then mapped to a predetermined interval through the activation function, and finally a health state score is generated. This score can be used to reflect the overall health status of the battery. The higher the value, the better the state, and a low score indicates that there may be a risk of failure. The long short-term memory network (LSTM) is a recurrent neural network that can capture long-term dependence relationships and is suitable for processing temporal data. LSTM selectively retains or forgets information through a gating mechanism (input gate, forget gate, output gate), which helps to model the changing trend of the battery state; by performing temporal modeling on the internal resistance mapping feature and the dynamic response feature, it can accurately capture the changes in the internal electrochemistry, thermodynamics, and dynamic response of the battery, providing a scientific quantitative basis for the health state score.

[0052] The present invention is further configured such that the CNN-LSTM hybrid model extracts the key fault fingerprints of multi-source data and battery analysis features through a convolutional neural network, performs temporal dynamic modeling on the features extracted by the convolutional neural network through a long short-term memory network, analyzes the evolution of fault patterns in the time dimension, and classifies fine-grained fault patterns for abnormal situations based on the output of the CNN-LSTM model. Specifically, the CNN-LSTM hybrid model is used to extract and perform temporal modeling on the fault information contained in the battery multi-source data and its analysis features. The specific construction logic is as follows: First, a convolutional neural network (CNN) is used to perform local feature extraction on the original multi-source data and battery analysis features, including internal resistance mapping features and dynamic response features. The CNN automatically learns local patterns and fault fingerprints in the data through multiple layers of convolutional filters, and can capture feature information caused by minute fluctuations and local abnormal changes. While effectively suppressing noise, this process extracts key feature maps reflecting changes in the internal state of the battery; Next, the feature maps extracted by the CNN are fed into a long short-term memory network (LSTM) module as input. The LSTM is a recurrent neural network that can capture the long-term dependencies of time series data. It selectively retains and forgets information in the time series through a gating mechanism (input gate, forget gate, output gate), thereby dynamically modeling the evolution of fault patterns in the time dimension. The main purpose of this stage is to capture the changing rules of local features over time and reveal the development trajectory of battery faults from initial anomalies to mature fault patterns; Finally, based on the output of the CNN-LSTM model, a fully connected layer and classifiers such as Softmax are used to perform fine-grained pattern classification on battery faults. The classifier maps the high-dimensional features after temporal modeling to the probability distributions of each fault category, including overcharge, over-discharge, overheat, internal short circuit, or consistency faults, thereby achieving accurate classification of abnormal situations and generating corresponding fault warning information. The above logic makes full use of the advantages of deep learning in local feature extraction and temporal modeling, significantly improving the accuracy and response speed of fault detection, and providing a more intelligent and adaptive fault monitoring and maintenance solution for the battery management system.

[0053] The present invention is further configured such that the dimension of the state vector includes multi-source data, internal resistance mapping features, dynamic response features, and fault risk aggregation features, wherein the fault risk aggregation feature is calculated through the cell voltage difference, charging temperature gradient, and polarization impedance index. wherein, Ω F is the fault risk aggregation feature, ΔV cell is the cell voltage difference, G T is the charging temperature gradient, Θ Pis the polarization impedance index, and θ1, θ2, and θ3 are power adjustment parameters. Specifically, by constructing a fault risk aggregation feature to quantify the potential fault risk existing inside the battery, this feature is obtained from the non-linear combination of three key indicators: the monomer voltage difference, the charging temperature gradient, and the polarization impedance index. Its basic idea is as follows: the monomer voltage difference reflects the balance among the individual cells in the battery pack, the charging temperature gradient reflects the thermal distribution during the charging process of the battery, and the polarization impedance index reflects the internal resistance change of the battery caused by electrochemical reaction polarization. After amplifying or suppressing these three indicators with preset power adjustment parameters respectively, then taking the natural logarithm of their product, the fault risk aggregation feature is obtained, which can more intuitively reveal the comprehensive risk level of battery faults. The monomer voltage difference refers to the voltage difference between different individual cells in the battery pack. A large voltage difference usually indicates the existence of imbalance inside the battery pack, which may lead to local overcharging, over-discharging, or inconsistent aging. The calculation method is to collect the voltages of all individual cells in the battery pack in real time and calculate the difference between the maximum value and the minimum value as the quantitative indicator of the monomer voltage difference. The charging temperature gradient refers to the temperature distribution difference between different parts or different individual cells during the charging process. A large temperature gradient may reflect uneven heat dissipation or the existence of local hot spots, thus increasing the fault risk. The calculation method is based on the difference in temperature data of each monitoring point or each individual cell during the charging process. The polarization impedance index represents the internal resistance change index of the battery caused by the polarization effect (polarization phenomenon formed during the electrochemical reaction process) during charge and discharge. This index reflects the comprehensive influence of ion transport and reaction kinetics inside the battery and is calculated by applying a short-time current pulse or an alternating current signal, measuring the change in the battery response voltage, and then using non-linear mapping. ΔV pulse represents the instantaneous drop in the battery voltage during the short-time pulse, and I pulse represents the pulse current. The power adjustment parameters θ1, θ2, and θ3 are respectively used to adjust the influence intensity of the monomer voltage difference, the charging temperature gradient, and the polarization impedance index during aggregation. The value range is [0.5, 3]; through the non-linear combination of the monomer voltage difference, the charging temperature gradient, and the polarization impedance index, the generated fault risk aggregation feature can comprehensively and intuitively reflect the balance problem, uneven thermal distribution, and polarization phenomenon existing inside the battery, not only improving the quantization accuracy of the battery fault risk, but also serving as an important part of the state vector, providing richer and more recognizable information for subsequent health state prediction and fault mode recognition based on deep learning, thereby improving the accuracy of fault warning and the safety of the system.

[0054] The present invention is further configured to construct a state-action value function using a deep Q-network according to the state vector, initialize an estimated value for each state-action pair, and perform iterative updates using historical data; the present invention is further configured that the state-action value function is: Q(s,a)′ = Q(s,a) + η[tanh(r + λ·max a′ Q(s′,a′)) - Q(s,a)], where Q(s,a)′ is the updated state-action value function, Q(s,a) is the initial state-action value function, η is the learning rate, λ is the discount factor, s is the state vector, s′ is the next state vector, a is the selected action, a′ is the next selected action, r is the immediate reward, and r = -ln(1 + Ω F (t)); specifically, a deep Q-network (DQN) is used to construct the state-action value function. Through iterative updates, the expected cumulative reward estimates for each state-action pair can be continuously corrected according to historical data, effectively integrating immediate and future reward information, realizing sensitive response and adaptive control of battery failure risks, and providing an accurate and stable decision-making basis for battery failure monitoring;

[0055] After the state-action value function is updated, calculate the expected value of each action according to the current state vector, construct a policy distribution, and select the optimal response action using random sampling or a greedy strategy; the policy distribution is: where π(a∣s) is the policy distribution, κ is the temperature parameter, and a ＂＂ are all possible actions under the state vector s; specifically, after the state-action value function is updated, the estimated values of the expected cumulative rewards of each action in the current state have been obtained. To select the optimal response action from these expected values, the Softmax strategy is used to non-linearly map Q(s,a) to a probability distribution, that is, to construct the policy distribution π(a∣s). Specifically, the κ-th power of Q(s,a) is mapped to the positive number space using the exponential function, and through normalization, the sum of the selection probabilities of all actions is 1. This policy distribution can not only reflect the advantages of each action relative to other actions but also be adjusted between exploration and exploitation through the temperature parameter κ. Finally, the optimal response action is selected using random sampling or a greedy strategy according to this distribution. The policy distribution is adjusted in real-time as the state-action value function is continuously updated, ensuring that the response strategy is always based on the latest battery operation information, improving decision-making accuracy and adaptability;

[0056] After the optimal response action is selected, generate and execute a control signal to achieve adaptive processing of the faulty battery.

[0057] Embodiment 2

[0058] Please refer to Figure 2, an exemplary battery fault monitoring and processing system based on deep learning is used to implement the above-mentioned battery fault monitoring and processing method based on deep learning. The system includes:

[0059] Data acquisition module: Real-time collect multi-source data of the batteries in the battery swapping cabinet, preprocess the multi-source data, and construct battery analysis features based on the preprocessed multi-source data;

[0060] Time series analysis module: Set the battery analysis features as the input, analyze the historical time series data through the battery health analysis model, and output the health status score;

[0061] Fault determination module: Input the multi-source data and battery analysis features of the batteries with health status scores less than the preset threshold into the CNN-LSTM hybrid model for real-time fault diagnosis, classify and identify battery anomalies, determine the fault category, and generate fault warning information;

[0062] Fault handling module: Obtain the state vector of the faulty battery, update the state-action value function through the reinforcement learning algorithm in the predefined action set, select the optimal response action according to the policy distribution, and realize the adaptive processing of the faulty battery.

[0063] It should be noted that the above-mentioned battery fault monitoring and processing system based on deep learning provided by the above embodiment and the above-mentioned battery fault monitoring and processing method based on deep learning belong to the same concept. The specific ways in which each module and unit perform operations have been described in detail in the method embodiment, and will not be repeated here. In practical applications, the above-mentioned battery fault monitoring and processing system based on deep learning provided by the above embodiment can, according to needs, allocate the above functions to different functional modules, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above. This is not limited here either.

[0064] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0065] It should be understood that the term "and / or" in this document is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this document generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood by referring to the context before and after.

[0066] In this application, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or a similar expression means any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0067] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0068] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0069] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0070] In several embodiments provided in this application, it should be understood that the disclosed systems can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be electrical, mechanical, or other forms.

[0071] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0072] In addition, the functional units in each embodiment of this application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0073] When the above-mentioned function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0074] As described above, the above are only specific implementation manners of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A battery fault monitoring and processing method based on deep learning, characterized in that: include: Collect multi-source data of batteries in the battery swap cabinet in real time, pre-process the multi-source data, and construct battery analysis features based on the pre-processed multi-source data; The battery analysis feature is set as input, the historical time series data is analyzed through the battery health analysis model, and the health status score is output; The multi-source data and battery analysis features of batteries with health status scores less than the preset threshold are input into the CNN-LSTM hybrid model for real-time fault diagnosis, which classifies and identifies battery anomalies, determines the fault category, and generates fault alarm information; The state vector of the faulty battery is obtained, the state-action value function is updated through the reinforcement learning algorithm in the predefined action set, and the optimal response action is selected according to the strategy distribution to achieve adaptive processing of the faulty battery.

2. A battery fault monitoring and processing method based on deep learning according to claim 1, characterized in that: The multi-source data include open circuit voltage, load voltage, charging current, discharging current, charging rate, charging temperature and cumulative deep discharge times, and the preprocessing includes normalization processing.

3. A battery fault monitoring and processing method based on deep learning according to claim 2, characterized in that: Battery analysis features include; Constructing internal resistance mapping characteristics based on open circuit voltage, load voltage and discharge current; The dynamic response characteristics are constructed based on the charging temperature, charging current, charging rate and cumulative deep discharge times.

4. A battery fault monitoring and processing method based on deep learning according to claim 3, characterized in that: The calculation logic of the internal resistance mapping feature is: Among them, Z int is the internal resistance mapping characteristic, V OC is the open circuit voltage, V L is the load voltage, I DIC is the discharge current, α, β are power adjustment parameters; The calculation logic of the dynamic response characteristics is: Where Λ is the dynamic response characteristic, T is the charging temperature, I CH is the charging current, R is the charging rate, U is the cumulative number of deep discharges, δ and ρ are phase adjustment parameters, and γ is the power adjustment parameter.

5. A battery fault monitoring and processing method based on deep learning according to claim 4, characterized in that: The battery health analysis model uses a long short-term memory network to perform time series modeling on the internal resistance mapping characteristics and dynamic response characteristics, and generates a health status score through a nonlinear fusion function, H = σ(w1·Z int +w2·Λ+b), where H is the health status score, σ(·) is the activation function, w1 and w2 are weight coefficients, and b is the bias term.

6. The battery fault monitoring and processing method based on deep learning according to claim 1 is characterized in that: The CNN-LSTM hybrid model extracts key fault fingerprints from multi-source data and battery analysis features through convolutional neural networks, performs temporal dynamic modeling on the features extracted by the convolutional neural network through long short-term memory networks, analyzes the evolution of fault modes in the time dimension, and performs fine-grained fault mode classification of abnormal situations based on the output of the CNN-LSTM model.

7. The battery fault monitoring and processing method based on deep learning according to claim 1 is characterized in that: The dimensions of the state vector include multi-source data, internal resistance mapping features, dynamic response features, and fault risk aggregation features. The fault risk aggregation features are calculated by using the cell voltage difference, charging temperature gradient, and polarization impedance index. Among them, Ω F is the failure risk aggregation feature, ΔV cell is the cell voltage difference, G T is the charging temperature gradient, Θ P is the polarization impedance index, and θ1, θ2 and θ3 are power adjustment parameters.

8. The battery fault monitoring and processing method based on deep learning according to claim 7 is characterized in that: A deep Q network is used to construct a state-action value function based on the state vector, an estimated value is initialized for each state-action pair, and historical data is used for iterative updates; After the state-action-value function is updated, the expected value of each action is calculated based on the current state vector, and the strategy distribution is constructed to select the optimal response action using random sampling or greedy strategy; After the optimal response action is selected, the control signal is generated and executed to achieve adaptive processing of the faulty battery.

9. A battery fault monitoring and processing method based on deep learning according to claim 8, characterized in that: The state action value function is: Q(s,a) ′ =Q(s,a)+η[tanh(r+λ·maxa′Qs′,a′-Qs,a, where Qs,a′ is the updated state action value function, Qs,a is the initial state action value function, η is the learning rate, λ is the discount factor, s is the state vector, s ′ is the next state vector, a is the selected action, a ′ is the next selected action, r is the immediate reward, r = -ln(1+Ω F (t)); The strategy distribution is: Among them, π(a|s) is the strategy distribution, κ is the temperature parameter, and a ′′ are all possible actions under the state vector s.

10. A battery fault monitoring and processing system based on deep learning, used to implement a battery fault monitoring and processing method based on deep learning according to any one of claims 1 to 9, characterized in that: include: Data acquisition module: collects multi-source data of batteries in the battery swap cabinet in real time, pre-processes the multi-source data, and constructs battery analysis features based on the pre-processed multi-source data; Time series analysis module: sets the battery analysis features as input, analyzes the historical time series data through the battery health analysis model, and outputs the health status score; Fault determination module: Input multi-source data and battery analysis features of batteries with health status scores less than the preset threshold into the CNN-LSTM hybrid model for real-time fault diagnosis, classify and identify battery anomalies, determine the fault category, and generate fault alarm information; Fault processing module: obtains the state vector of the faulty battery, updates the state-action value function through the reinforcement learning algorithm in the predefined action set, selects the optimal response action according to the strategy distribution, and realizes adaptive processing of the faulty battery.

Images