Rapid degradation diagnosis method for lead-carbon batteries based on feature fusion

Abstract

This application relates to a method, system, electronic device, and storage medium for diagnosing rapid degradation of lead-carbon batteries based on feature fusion. By extracting multiple features related to battery aging and adopting an effective feature fusion method, the application comprehensively considers the various changing factors of the battery during the aging process. Compared with the existing technology that relies only on a single parameter or simply independently analyzes multiple parameters, it can more accurately diagnose whether the battery has accelerated aging, greatly improving the accuracy of the diagnosis. In addition, the diagnostic model constructed based on technologies such as deep separable one-dimensional convolution, residual connection, and LSTM can quickly process large amounts of battery data, promptly detect signs of accelerated battery aging, and avoid a sharp decline in the performance of the entire module due to untimely diagnosis, thereby ensuring the stable operation of the lead-carbon energy storage power station and improving the safety and reliability of the system.

Description

Technical Field

[0001] The present application relates to the field of battery diagnosis, and in particular to a method, system, computer device, and computer-readable storage medium for diagnosing rapid battery degradation based on feature fusion. Background Art

[0002] The operating architecture of a lead-carbon energy storage power station usually relies on a large number of single lead-carbon batteries tightly combined in series and parallel. While this combination achieves high energy storage and output, it also places extremely high demands on the performance stability of the single battery.

[0003] Related technologies for diagnosing accelerated aging of single-cell batteries have limitations. Some diagnostic methods rely on a single parameter, such as monitoring battery capacity to determine battery status. However, due to the complex degradation process, a single parameter cannot fully reflect internal changes in the battery. For example, when the battery's internal resistance changes, it is difficult to detect signs of accelerated aging in a timely manner by monitoring capacity alone.

[0004] Furthermore, while some methods consider multiple parameters, they simply analyze each parameter independently, failing to fully explore the inherent connections between them. This results in poor accuracy and timeliness in the diagnosis of complex battery degradation scenarios.

[0005] Currently, no effective solution has been proposed to address the problem of poor accuracy in battery degradation diagnosis in related technologies. Summary of the Invention

[0006] The embodiments of the present application provide a method, system, computer device, and computer-readable storage medium for rapid degradation diagnosis of lead-carbon batteries based on feature fusion, so as to at least solve the problem of poor accuracy in battery degradation diagnosis in related technologies.

[0007] In a first aspect, an embodiment of the present application provides a method for diagnosing rapid degradation of a lead-carbon battery based on feature fusion, which is implemented based on a diagnostic model. The steps of any round of operation of the model include:

[0008] Obtaining battery-related time series vectors and scalar features, where the time series vectors are used to describe the electrical and thermal states of the battery during operation from multiple dimensions, and the scalar features are used to describe the characteristic changes of the battery during aging from multiple dimensions;

[0009] Performing a depth-separable one-dimensional convolution process on the time series vector, and adding the output of the convolution process to the original time series vector through a residual connection method to obtain a deep feature representation;

[0010] Through the LSTM network, the long-term dependency in the time series vector is obtained based on the deep feature representation, and the LSTM output corresponding to each time series vector is average pooled to obtain the optimized time series feature;

[0011] The optimized time series features and the scalar features are fused to obtain combined features, and the combined features are dynamically weighted and fused through a self-attention mechanism to obtain a battery diagnosis result.

[0012] In some embodiments, the method further comprises:

[0013] Collecting raw battery data related to battery aging, and assigning labels to the raw battery data according to whether accelerated aging occurs;

[0014] Extracting multi-dimensional key features reflecting performance changes during battery aging from the raw battery data after label assignment, and constructing time series vectors and scalar features based on the multi-dimensional key features;

[0015] Constructing a data set based on the time series vector and the scalar feature, and dividing the data set into a training set and a test set;

[0016] Based on the training set, model training is performed by iteratively executing the operating steps of the diagnostic model, wherein a cross-entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true label distribution, and based on the difference, a backpropagation algorithm and an optimizer are used to gradually optimize the model parameters, and when the model converges, a trained diagnostic model is obtained;

[0017] The trained diagnostic model is used to perform rapid degradation diagnosis based on target battery data in actual scenarios.

[0018] In some embodiments, the time series vector corresponds to the battery voltage, current and temperature data of a specific time segment; the scalar features correspond to the valley mean of the voltage difference curve and the main peak area and main peak height in the capacity increment curve, which are used to describe the characteristic changes of the battery aging process from multiple dimensions.

[0019] In some embodiments, performing separable one-dimensional convolution on the time series vector and adding the output of the convolution operation to the original time series vector through a residual connection to obtain a deep feature representation includes:

[0020] Perform separable one-dimensional convolution processing on each time series vector to extract local features in each time series and reduce computational complexity, and output a set of local features;

[0021] The local feature set is added to the original time series by adopting a residual connection method to obtain the deep feature representation.

[0022] In some embodiments, obtaining long-term dependencies in a time series vector based on the deep feature representation through an LSTM network includes:

[0023] The LSTM network takes the deep feature representation as input and initializes the cell state and hidden state in the network;

[0024] Through the forget gate, determine the long-term features that need to be retained, and through the input gate, evaluate the importance of the current input and generate short-term key information based on the importance;

[0025] Merging the retained long-term features with the short-term key information to update the memory unit, and generating a current hidden state according to the updated memory unit through an output gate;

[0026] By executing the above process time-step by time, the long-term features and short-term anomalies are dynamically balanced to obtain the LSTM output.

[0027] In some embodiments, performing average pooling on the LSTM output corresponding to each time series vector includes:

[0028] The output result of the LSTM network is obtained, and the optimized time series features are obtained by averaging all elements in the LSTM output corresponding to each time series vector and compressing the LSTM output into a vector representation of a fixed length.

[0029] In some embodiments, the combined features are dynamically weighted and fused through a self-attention mechanism to obtain a battery diagnosis result including:

[0030] Obtaining the combined features, and calculating a query vector by combining the combined features with a weight matrix and a bias vector;

[0031] The elements in the query vector are converted into probability values ​​through the Softmax function to obtain attention weights, where the sum of the probabilities corresponding to all elements is 1. The attention weights represent the relative importance of different features under the current attention mechanism;

[0032] The combined feature is combined with the attention weight to obtain an attention output, and the attention output is processed again by the Softnax function to obtain a probability distribution, wherein the probability distribution reflects the possibility that the original battery data belongs to different labels.

[0033] In a third aspect, an embodiment of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect above when executing the computer program.

[0034] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect above.

[0035] Compared to related technologies, the feature fusion-based lead-carbon battery rapid degradation diagnosis method provided in the embodiments of this application extracts multiple features related to battery aging and adopts an effective feature fusion method to comprehensively consider the various changing factors of the battery during the aging process. Compared with existing technologies that rely only on a single parameter or simply analyze multiple parameters independently, it can more accurately diagnose whether the battery is experiencing accelerated aging, greatly improving the accuracy of diagnosis. In addition, the diagnostic model constructed based on technologies such as deep separable one-dimensional convolution, residual connection, and LSTM can quickly process large amounts of battery data and promptly detect signs of accelerated battery aging, avoiding a sharp decline in the performance of the entire module due to untimely diagnosis, ensuring the stable operation of the lead-carbon energy storage power station and improving the safety and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0037] Figure 1 is a flow chart of a lead-carbon battery rapid degradation diagnosis method based on feature fusion according to an embodiment of the present application;

[0038] Figure 2 This is a structural block diagram of a lead-carbon battery rapid degradation diagnosis system based on feature fusion according to an embodiment of the present application;

[0039] Figure 3 Schematic diagram of the internal structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0041] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0042] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0043] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote quantitative limitations and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0044] In the field of lead-carbon energy storage power stations, they are usually composed of a large number of single lead-carbon batteries connected in series and parallel. At present, existing technologies have certain limitations in diagnosing the accelerated aging of single batteries. Some diagnostic methods rely on only a single parameter, such as monitoring only the battery capacity to determine the battery status. This approach is too one-sided because battery degradation is a complex process, and a single parameter cannot fully reflect the various changes within the battery. For example, when the internal resistance of the battery changes, it may not be possible to detect signs of accelerated aging of the battery in a timely manner through capacity monitoring alone. In addition, although some methods take multiple parameters into consideration, they simply analyze each parameter independently and do not fully explore the inherent connections between different parameters. This leads to poor accuracy and timeliness of diagnostic results when faced with complex battery degradation situations.

[0045] In the actual operation of lead-carbon energy storage power stations, if a single cell exhibits accelerated aging, it can severely impact the performance of the entire module. Due to the shortcomings of existing diagnostic technologies, accelerated aging cells cannot be promptly detected and addressed, leading to a sharp decline in the overall performance of the energy storage station, which in turn creates significant safety hazards such as overheating and fire. The primary reasons for this inaccurate and untimely diagnosis are insufficient analysis of the complex relationships between the various factors influencing battery degradation, as well as a lack of effective multi-parameter fusion analysis methods. Addressing these issues presents the challenges of accurately extracting features associated with accelerated battery aging and effectively fusing different types of features to improve diagnostic accuracy.

[0046] In view of this, the embodiment of the present application provides a method for rapid degradation diagnosis of gold-carbon batteries based on feature fusion. Figure 1 This is a flow chart of a lead-carbon battery rapid degradation diagnosis method based on feature fusion according to an embodiment of the present application, such as Figure 1 As shown in the figure, this method is implemented based on the diagnostic model. The running process of the model in any round includes the following steps:

[0047] S101, collecting raw battery data related to battery aging, and assigning labels to the raw battery data according to whether accelerated aging occurs;

[0048] The raw battery data is data related to battery aging, which can be obtained in a variety of ways. In this embodiment, the specific acquisition path and method are not specifically limited.

[0049] Furthermore, the collected battery aging data is screened and, based on the actual aging state of the battery, clearly divided into stages where accelerated aging occurs and stages where it does not. To facilitate subsequent data analysis and model training, the accelerated aging stage is assigned label 1, and the non-accelerated aging stage is assigned label 0. This approach cleverly transforms the previously complex problem of determining the inflection point of battery aging into a simple binary classification problem. During model training, the labels provide the model with clear learning objectives. The model continuously adjusts its parameters to attempt to make its output as close as possible to the true categories represented by these labels.

[0050] S102, extracting multi-dimensional key features reflecting performance changes during battery aging from the original battery data after label assignment, and constructing time series vectors and scalar features based on the multi-dimensional key features;

[0051] Among them, multi-dimensional key features are extracted from the original battery aging data for subsequent model analysis, which can reflect the performance changes of the battery during the aging process from different angles; among them, multi-dimensional key features include but are not limited to: basic features such as current, voltage, temperature, and derived features under specific conditions, that is, when the battery power is in the range of 20%-70%, the average valley value of the voltage difference curve, and the main peak area and peak height in the capacity increment curve.

[0052] In this embodiment, the extracted multi-dimensional characteristic parameters can reflect changes in battery aging performance from multiple perspectives. Current, voltage, and temperature provide information on the basic operating state of the battery; the average valley value of the voltage differential curve reveals aging characteristics from the perspective of the charge-voltage relationship; and the area and height of the main peak of the capacity increment curve provide aging clues from the perspective of battery capacity changes. Combined, these characteristic parameters provide rich and critical input for subsequent model analysis, helping the model accurately capture the characteristic patterns of accelerated battery aging and improve the accuracy and reliability of accelerated battery aging diagnosis.

[0053] Furthermore, when the lead-carbon energy storage power station is operating, the time series vector composed of the battery voltage (V), current (I) and temperature (T) data within a specific time segment reflects the battery's electrical and thermal status in real time.

[0054] Among them, voltage data can reflect the battery's ability to store and release electrical energy, and its fluctuation over time can reflect the process of chemical reactions inside the battery; the current size is closely related to the battery's charging and discharging efficiency, and changes in the battery's internal resistance during aging will cause the current response to change; the temperature is affected by the thermal effect of the battery's internal reactions, and abnormal temperature changes indicate that there may be problems with the battery.

[0055] Furthermore, variable S1, representing the average value of the trough of the voltage differential curve within the 20%-70% battery charge range, reflects the voltage variation characteristics of the battery within this commonly used charge range, revealing changes in battery structure and reaction. S2 and S3, respectively, represent the main peak area and peak height of the capacity increment curve. The capacity increment curve reflects the changes in battery charge and discharge capacity, and parameters related to its main peak can reflect the battery capacity decay rate and total capacity change. These parameters quantify battery aging characteristics from different dimensions, complementing the vectors V, I, and T. Together, they provide a comprehensive basis for accurately diagnosing accelerated battery aging and provide a deeper understanding of the battery aging process.

[0056] S103, constructing a data set based on the time series vector and scalar features, and dividing the data set into a training set and a test set;

[0057] The training set is used to train the diagnostic model. The model continuously adjusts its parameters by learning the relationship between the features and labels of the data in the training set. For example, the model optimizes the weight matrices and bias vectors of its internal convolutional and fully connected layers based on the characteristic parameters of the aging process of different batteries in the training set, such as voltage, current, and temperature, as well as the corresponding accelerated aging labels (1 for accelerated aging, 0 for non-accelerated aging). Using a backpropagation algorithm and optimizers (such as the Adam optimizer), the model gradually improves its classification accuracy of the training data during training, enabling it to make accurate judgments when presented with battery data with similar feature distributions to the training set.

[0058] The test set is used to evaluate model performance. Since the test set data is not used during model training, when the model demonstrates good diagnostic performance on the test set, it indicates good generalization. Proper data segmentation can effectively avoid this situation, ensuring that the model can accurately diagnose whether the battery is experiencing accelerated aging based on different battery data samples, whether from different time periods within the same energy storage power station or from different energy storage power stations. This ensures the stable operation of the lead-carbon energy storage power station and improves the reliability and safety of the entire system in practical applications.

[0059] S104, based on the training set, the model is trained by iteratively executing the operation steps of the diagnostic model, wherein the cross entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true label distribution. Based on the difference, the model parameters are gradually optimized using the back propagation algorithm and the optimizer. When the model converges, a trained diagnostic model is obtained;

[0060] Specifically, an exemplary operation process of the diagnostic model includes the following steps:

[0061] Step 1: Obtain battery-related time series vectors and scalar features. The time series vectors are used to describe the electrical and thermal states of the battery during operation from multiple dimensions, and the scalar features are used to describe the characteristic changes during battery aging from multiple dimensions.

[0062] Through the above steps, time series vectors and scalar features are combined as inputs for subsequent model training. These input features cover both immediate and long-term, macroscopic and microscopic changes in the battery's state, comprehensively capturing the characteristics of battery aging. Existing techniques that rely solely on a single parameter to diagnose battery aging are flawed. For example, considering only capacity fails to promptly detect aging caused by changes in internal resistance. These two inputs integrate multiple factors, avoiding one-sidedness, providing a sufficient basis for accurate diagnosis, and improving diagnostic reliability and accuracy.

[0063] Step 2: Perform depth-wise separable one-dimensional convolution on the time series vector and add the output of the convolution operation to the original time series vector through residual connection to obtain deep feature representation.

[0064] Specifically, for the time series vectors (V, I, T), the operations Yv = 1D_CNN(V), YI = 1D_CNN(I), and YT = 1D_CNN(T) are performed, respectively. During this process, a specific convolution kernel slides along the time dimension, performing weighted summation and other operations on each time point and its adjacent data to extract local features. For example, for the voltage time series vector V, the convolution kernel focuses on the voltage value changes within a certain time period, capturing characteristics such as the amplitude and frequency of voltage fluctuations. These characteristics reflect the changes in the battery's electrical performance during this period. The same principle applies to the current vector I and the temperature vector T.

[0065] Compared to traditional convolutional methods, depthwise separable one-dimensional convolution decomposes the convolution operation into depthwise convolution and pointwise convolution, significantly reducing the amount of computation. This reduction in computational complexity is particularly important when processing large-scale battery data, improving model training and inference efficiency while preserving key feature information, ensuring the model can quickly and accurately process battery aging-related data.

[0066] Furthermore, a residual connection method is used to add the output of the depthwise separable one-dimensional convolution to the original input to obtain a deep feature representation that contains more information and enhances learning ability. This is achieved through the following formula:

[0067] Yresl_V = Yv + V, Yresl_I = Yi + I, Yresl_T = YT + T;

[0068] As you can understand, the added features fuse the original input features with the local features extracted by convolution. The original input contains the most basic information about the data, while the depthwise separable one-dimensional convolution extracts local characteristic patterns within the data. Adding these two together creates a new deep feature representation that retains the original information of the data while incorporating the more representative local features extracted by the convolution operation, enabling the model to understand and learn from data from multiple perspectives.

[0069] Furthermore, in deep neural networks, as the number of network layers increases, the gradient can become very small during backpropagation, making it difficult for the model to learn deeper features, a phenomenon known as the vanishing gradient problem. In this embodiment, residual connections (adding the convolutional output to the original input) provide a direct path for gradient propagation. During backpropagation, the gradient can be directly propagated from the added result back to the original input, preventing the gradient from decaying in the multi-layer network. This allows the model to learn deeper features more efficiently, improving training effectiveness and performance.

[0070] Step 3: Through the LSTM network, based on deep feature representation, the long-term dependencies in the time series vector are obtained, and the LSTM output corresponding to each time series vector is average pooled to obtain the optimized time series features;

[0071] It should be noted that the above steps have performed operations such as deep separable one-dimensional convolution and residual network on the battery's voltage, current, temperature and other time series vectors, and obtained outputs that contain local features and have been fused. However, these operations have limited ability to mine long-term dependencies in the data. The battery aging process is a process that gradually changes over time and has obvious long-term dependency characteristics, such as the impact of the battery's charge and discharge cycles over a long period of time on its aging degree.

[0072] Long Short-Term Memory (LSTM) is a special type of recurrent neural network (RNN) that effectively captures long-term dependencies in time series. LSTM uses its unique gating mechanism (input gate, forget gate, and output gate) to control the flow and memory of information, selectively remembering or forgetting information in a time series. When processing battery aging data in this application scenario, LSTM can learn the correlation between battery state changes at different time points, such as the impact of battery temperature changes over a period of time on the current battery aging level, thereby extracting more valuable time series features.

[0073] Specifically, the operation process of an LSTM network includes the following steps:

[0074] S1, the LSTM network takes the deep feature representation as input and initializes the cell state and hidden state in the network;

[0075] S2, through the forget gate, determines the long-term features that need to be retained, and through the input gate, evaluates the importance of the current input and generates short-term key information based on the importance;

[0076] S3, merges the retained long-term features with short-term key information to update the memory unit, and generates the current hidden state based on the updated memory unit through the output gate;

[0077] S4, by executing the above process time-step by time, dynamically balances long-term features and short-term anomalies to obtain LSTM output.

[0078] In the above steps, the forget gate allows the LSTM to selectively forget irrelevant past information, avoiding the vanishing or exploding gradient problems of traditional RNNs and enabling the model to process longer time series. Furthermore, the presence of the input and output gates enables the LSTM to selectively update the cell state and output hidden state based on the current input and past hidden states, effectively capturing long-term dependencies in the time series.

[0079] It's understandable that after LSTM processing, each time series vector has a corresponding LSTM output. Furthermore, a global average pooling operation processes these LSTM outputs. Optionally, all elements in the LSTM output corresponding to each time series vector are averaged, compressing the features of the entire time series into a fixed-length vector representation to obtain optimized time series features. This can be achieved using the following formula:

[0080]

[0081] It can be understood that through the global average pooling operation, the features of different time series can be processed on a unified dimension during subsequent feature fusion, which facilitates effective fusion with other scalar features (such as the average value of the trough of the voltage difference curve and the area and peak height of the main peak in the capacity increment curve).

[0082] In step 4, the optimized time series features and scalar features are fused to obtain combined features. The converted scalar features are dynamically weighted and fused with the time series deep feature representation through the self-attention mechanism to obtain the battery diagnosis results.

[0083] Specifically, the scalar features S1, S2, and S3 are transformed through a fully connected layer (FCL) to enable effective fusion with the previously processed time series features. The FCL learns appropriate weights and biases to map S1, S2, and S3 to a feature space that matches the time series features, resulting in the fused feature S.

[0084] Furthermore, the optimized time series features and scalar features are fused to obtain combined features, where the combined features can be expressed by the following expression:

[0085]

[0086] In the feature fusion stage, these features from different sources and types need to be integrated and further processed, and then query vector calculation, attention weight calculation and other operations are performed based on them to achieve more effective feature fusion and accelerated battery aging diagnosis.

[0087] Furthermore, the combined features are dynamically weighted and fused through the self-attention mechanism to obtain the battery diagnosis results, which specifically includes the following steps:

[0088] S1, obtains the combined features and calculates the query vector by combining the combined features with the weight matrix and bias vector;

[0089] The query vector is expressed as follows:

[0090]

[0091] in is the weight matrix, which determines the The weighting degree of different features in The calculated query vector is used to adjust the calculation results. The calculated query vector is used in the subsequent calculation of attention weights, which measure the importance of different locations in the data. This helps the model focus on key information and improves data processing and analysis. For example, it can better extract and utilize effective features in tasks such as classification.

[0092] S2, through the Softmax function, converts the elements in the query vector into probability values ​​to obtain attention weights, where the sum of the probabilities corresponding to all elements is 1. The attention weights represent the relative importance of different features under the current attention mechanism;

[0093] The specific flow can be expressed by the following formula:

[0094]

[0095] Among them, A is the attention weight.

[0096] The Softmax function is essentially a normalized exponential function. For each element in the query vector, the attention weights are converted by the Softmax function. The sum of these attention weights is 1. In the context of battery aging diagnosis, these attention weights represent the relative importance of features at different positions within the current attention mechanism.

[0097] If the attention weight corresponding to a certain feature is large, it means that the feature is more critical in the process of the current model judging whether the battery is aging rapidly. The model will pay more attention to these important features in subsequent processing, thereby improving the accuracy and reliability of the diagnosis, helping to more accurately identify the key information of the battery aging status and avoid misjudgment due to interference from secondary information.

[0098] S3 combines the combined features with the attention weights to obtain the attention output, and processes the attention output again through the Softmax function to obtain the probability distribution, where the probability distribution reflects the possibility that the original battery data belongs to different labels.

[0099] Among them, for the combined features , calculate the attention output and re-weight the combined features.

[0100] For example, if the attention weight corresponding to the battery voltage feature at a certain point in time is high, then when calculating the attention output When , the voltage feature at that time point will be strengthened in the new deep feature representation, and its value will be relatively increased; while for features with lower weights, The influence will be weakened.

[0101] The advantage of doing this is that it can highlight the feature information that is more critical to the diagnosis of accelerated battery aging, so that the model can focus more on these important features when performing subsequent operations such as classification output, reducing the interference of irrelevant or minor features, thereby improving the accuracy and reliability of the entire diagnostic model.

[0102] Furthermore, after calculating the attention output, the classification output is calculated, where the output is converted into a probability distribution through the softmax() function and the final result is output;

[0103] This process can be expressed by the following formula:

[0104]

[0105] After the preceding series of feature extraction, fusion, and attention mechanisms, the results need to be further transformed into an intuitive form with clear judgment criteria. The softmax function maps output values ​​between 0 and 1, with the sum of all output values ​​being 1, forming a probability distribution.

[0106] When determining whether a battery has accelerated aging, different probability values ​​in this probability distribution correspond to the likelihood of the battery being in different states (accelerated aging or non-accelerated aging). For example, if the probability value corresponding to accelerated aging in the output probability distribution is close to 1, and the probability value corresponding to non-accelerated aging is close to 0, then it can be determined with a high degree of certainty that the battery has experienced accelerated aging.

[0107] During the operation of the above model, LSTM is used to capture the long-term dependencies of battery aging data and explore the hidden information of data changing over time. It is further combined with global average pooling to compress features into fixed-length vectors and unify the feature dimensions. The feature fusion step uses a fully connected layer to fuse the converted scalar features with the processed time series features, integrating multiple types of features and enriching the diagnostic information dimension. Finally, the self-attention mechanism further optimizes the fusion effect, calculates the query vector, attention weight and output, re-weights the features to highlight key information, and finally converts the results into probability distributions through the softmax function. This provides an intuitive and accurate basis for judging whether the battery is aging rapidly, enhances the model's ability to diagnose complex battery aging conditions, and ensures the stable operation of lead-carbon energy storage power stations.

[0108] In combination with the above embodiments, it can be understood that the overall model training process in the present application scheme is based on the training set and is carried out by iteratively executing the operating steps of the diagnostic model. Specifically, the model training process uses the cross-entropy loss function to accurately measure the degree of difference between the model prediction results and the true labels.

[0109] The loss function can be expressed by the following formula:

[0110]

[0111] Furthermore, during training, the model parameters are updated using a combination of the backpropagation algorithm and the Adam optimizer. The backpropagation algorithm uses the gradient information calculated from the loss function to propagate the error from the output layer back to each hidden layer and the input layer, thereby determining the contribution of each parameter to the error. The Adam optimizer utilizes an adaptive learning rate adjustment strategy to dynamically adjust the learning step size for different parameters, enabling faster and more stable model convergence during training.

[0112] Through continuous iterative training and repeated adjustment of the weight matrices and bias vectors involved in each convolutional layer and fully connected layer, the model continuously learns the characteristic patterns and rules in battery aging data, gradually reduces the loss value, improves classification performance, and ultimately achieves accurate classification of the accelerated aging status of the battery, effectively solving the problems of inaccurate and untimely diagnosis in existing technologies, and ensuring the safe and stable operation of lead-carbon energy storage power stations.

[0113] S105, using the trained diagnostic model, perform rapid degradation diagnosis based on the target battery in the actual scenario.

[0114] As you can understand, after completing the diagnostic model training, the practical application phase begins. In real-world scenarios, the target battery's relevant time series data, such as voltage, current, and temperature, as well as characteristic data such as voltage differential curves and capacity increment curves, are collected and processed to meet the model input requirements.

[0115] This data is then fed into the trained diagnostic model, which quickly determines whether the target battery is experiencing accelerated aging and decline based on the characteristic patterns and patterns of battery aging learned during training. This step leverages the value of the previous work, including data preprocessing, feature extraction, model building, and training, to promptly and accurately identify aging issues in the target battery. This provides strong support for the safe and stable operation of lead-carbon energy storage power plants, effectively avoiding module performance degradation and safety hazards caused by undiagnosed accelerated battery aging.

[0116] Through steps S101 to S105 above, multiple features related to battery aging are extracted. An effective feature fusion method is employed to comprehensively account for the various factors affecting battery aging. Compared to existing techniques that rely solely on a single parameter or simply analyze multiple parameters independently, this method can more accurately diagnose whether a battery is experiencing accelerated aging, significantly improving diagnostic accuracy. Furthermore, the diagnostic model, built using technologies such as deep separable one-dimensional convolution, residual connections, and LSTM, can rapidly process large amounts of battery data and promptly detect signs of accelerated battery aging, preventing a sharp decline in overall module performance due to untimely diagnosis. This ensures the stable operation of the lead-carbon energy storage power station and improves the safety and reliability of the system.

[0117] In a second aspect, the present application also provides a lead-carbon battery rapid degradation diagnosis system based on feature fusion. Figure 2 This is a structural block diagram of a lead-carbon battery rapid degradation diagnosis system based on feature fusion according to an embodiment of the present application. Figure 2 As shown, the system includes: a pre-processing module 20, a feature optimization module 21 and an output module 22, wherein:

[0118] The preprocessing module 20 is used to obtain battery-related time series vectors and scalar features, where the time series vectors are used to describe the electrical and thermal states of the battery during operation from multiple dimensions, and the scalar features are used to describe the characteristic changes of the battery during aging from multiple dimensions; and perform depthwise separable one-dimensional convolution on the time series vectors, and add the output of the convolution process to the original time series vectors through a residual connection to obtain a deep feature representation;

[0119] The feature optimization module 21 is used to obtain the long-term dependency in the time series vector based on the deep feature representation through the LSTM network, and average pool the LSTM output corresponding to each time series vector to obtain the optimized time series feature;

[0120] The output module 22 is used to fuse the optimized time series features and scalar features to obtain combined features, and dynamically weighted fuse the combined features through the self-attention mechanism to obtain battery diagnosis results.

[0121] This system extracts multiple features related to battery aging and employs an effective feature fusion method to comprehensively account for the various factors affecting battery aging. Compared to existing technologies that rely solely on a single parameter or simply analyze multiple parameters independently, this system can more accurately diagnose whether a battery is experiencing accelerated aging, significantly improving diagnostic accuracy. Furthermore, a diagnostic model built using technologies such as deep separable one-dimensional convolution, residual connections, and LSTM can rapidly process large amounts of battery data, promptly identifying signs of accelerated battery aging and preventing a sharp decline in overall module performance due to untimely diagnosis. This ensures the stable operation of lead-carbon energy storage power plants and improves system safety and reliability.

[0122] In one embodiment, Figure 3 is a schematic diagram of the internal structure of an electronic device according to an embodiment of the present application, such as Figure 3 As shown, an electronic device is provided, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 3 As shown. The electronic device includes a processor, a network interface, an internal memory, and a non-volatile memory connected via an internal bus, wherein the non-volatile memory stores an operating system, a computer program, and a database. The processor is used to provide computing and control capabilities, the network interface is used to communicate with external terminals via a network connection, the internal memory is used to provide an environment for the operation of the operating system, the computer program, when executed by the processor, implements a lead-carbon battery rapid degradation diagnosis method based on feature fusion, and the database is used to store data.

[0123] Those skilled in the art will understand that Figure 3The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device to which the solution of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0124] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0125] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A lead-carbon battery rapid degradation diagnosis method based on feature fusion, characterized in that: Based on the diagnostic model implementation, the steps of any round of operation of the model include: Obtaining time series vectors and scalar features related to lead-carbon batteries, where the time series vectors correspond to battery voltage, current, and temperature data for a specific time segment, used to describe the electrical and thermal states of the battery during operation from multiple dimensions. The scalar features correspond to the valley mean of the voltage difference curve and the main peak area and main peak height in the capacity increment curve, used to describe the characteristic changes during battery aging from multiple dimensions; Performing depth-separable one-dimensional convolution processing on the time series vector, and adding the output of the convolution processing to the original time series vector through a residual connection method, fusing the original input features with the local features extracted by convolution to obtain a deep feature representation; Through the LSTM network, the long-term dependency in the time series vector is obtained based on the deep feature representation, and the LSTM output corresponding to each time series vector is average pooled to obtain the optimized time series feature; The optimized time series features and the scalar features are fused to obtain combined features, and the combined features are dynamically weighted and fused through a self-attention mechanism to obtain a battery diagnosis result.

2. The method according to claim 1, characterized in that The method further comprises: Collecting raw battery data related to battery aging, and assigning labels to the raw battery data according to whether accelerated aging occurs; Extracting multi-dimensional key features reflecting performance changes during battery aging from the raw battery data after label assignment, and constructing time series vectors and scalar features based on the multi-dimensional key features; Constructing a data set based on the time series vector and the scalar feature, and dividing the data set into a training set and a test set; Based on the training set, model training is performed by iteratively executing the operating steps of the diagnostic model, wherein a cross-entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true label distribution, and based on the difference, a backpropagation algorithm and an optimizer are used to gradually optimize the model parameters, and when the model converges, a trained diagnostic model is obtained; The trained diagnostic model is used to perform rapid degradation diagnosis based on target battery data in actual scenarios.

3. The method according to claim 1, characterized in that Performing separable one-dimensional convolution on the time series vector and adding the output of the convolution operation to the original time series vector through a residual connection to obtain a deep feature representation includes: Perform separable one-dimensional convolution processing on each time series vector to extract local features in each time series and reduce computational complexity, and output a set of local features; The local feature set is added to the original time series by adopting a residual connection method to obtain the deep feature representation.

4. The method according to claim 3, characterized in that The long-term dependencies in the time series vector are obtained by using the LSTM network based on the deep feature representation, including: The LSTM network takes the deep feature representation as input and initializes the cell state and hidden state in the network; Through the forget gate, determine the long-term features that need to be retained, and through the input gate, evaluate the importance of the current input and generate short-term key information based on the importance; Merging the retained long-term features with the short-term key information to update the memory unit, and generating a current hidden state according to the updated memory unit through an output gate; By executing the above process time-step by time, the long-term features and short-term anomalies are dynamically balanced to obtain the LSTM output.

5. The method according to claim 4, characterized in that Average pooling of the LSTM output corresponding to each time series vector includes: The output result of the LSTM network is obtained, and the optimized time series features are obtained by averaging all elements in the LSTM output corresponding to each time series vector and compressing the LSTM output into a vector representation of a fixed length.

6. The method according to claim 4, characterized in that The combined features are dynamically weighted and fused through the self-attention mechanism to obtain the battery diagnosis results including: Obtaining the combined features, and calculating a query vector by combining the combined features with a weight matrix and a bias vector; The elements in the query vector are converted into probability values ​​through the Softmax function to obtain attention weights, where the sum of the probabilities corresponding to all elements is 1. The attention weights represent the relative importance of different features under the current attention mechanism; The combined feature is combined with the attention weight to obtain an attention output, and the attention output is processed again by the Softnax function to obtain a probability distribution, wherein the probability distribution reflects the possibility that the original battery data belongs to different labels.

7. A lead-carbon battery rapid degradation diagnosis system based on feature fusion, characterized in that: The system includes: a preprocessing module, a feature optimization module and an output module, wherein: The preprocessing module is used to obtain time series vectors and scalar features related to lead-carbon batteries, wherein the time series vectors correspond to battery voltage, current, and temperature data of a specific time segment, and are used to describe the electrical and thermal states of the battery during operation from multiple dimensions; the scalar features correspond to the valley mean of the voltage difference curve and the main peak area and main peak height in the capacity increment curve, and are used to describe the characteristic changes during the battery aging process from multiple dimensions; Furthermore, a depth-separable one-dimensional convolution process is performed on the time series vector, and the output of the convolution process is added to the original time series vector through a residual connection method, and the original input features are fused with the local features extracted by the convolution to obtain a deep feature representation; The feature optimization module is used to obtain the long-term dependency in the time series vector based on the deep feature representation through the LSTM network, and perform average pooling on the LSTM output corresponding to each time series vector to obtain optimized time series features; The output module is used to fuse the optimized time series features and the scalar features to obtain combined features, and dynamically weighted fuse the combined features through a self-attention mechanism to obtain a battery diagnosis result.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images