Method and device for predicting state of health of battery and computer equipment

Abstract

The invention relates to a battery health state prediction method and device and computer equipment. The method comprises the following steps: acquiring battery data of a target battery under a plurality of working conditions; inputting the battery data into a pre-constructed feature extraction network, and extracting a feature vector representing the operation behavior of the target battery; and inputting the feature vector into a pre-constructed nonlinear function combination network for nonlinear processing to obtain a health state prediction result of the target battery. By adopting the method, the accuracy of a battery health assessment result can be improved.

Description

Technical Field

[0001] The present application relates to the field of energy storage technology, and in particular to a method, device, and computer equipment for predicting the health status of a battery. Background Art

[0002] With the acceleration of industrialization, human production activities have caused significant environmental pollution and resource depletion worldwide. To reduce environmental impact and alleviate the imbalance between energy supply and demand, various green and renewable energy technologies are being widely promoted and applied. Among them, lithium-ion batteries, with their high energy density and long cycle life, have become a key energy storage unit in new energy systems, widely used in a variety of fields, including aerospace and electric vehicles.

[0003] However, with the large-scale deployment of lithium-ion batteries in practical applications, performance degradation issues have gradually emerged during operation. Battery State of Health (SOH) assessment has become an important foundation for ensuring system reliability and safety. Therefore, a battery SOH determination method that can operate stably under various operating conditions is urgently needed to improve the accuracy and consistency of health assessment results and provide a reliable basis for subsequent safety management and performance optimization. Summary of the Invention

[0004] Based on this, it is necessary to provide a battery health status prediction method, device and computer equipment that can improve the accuracy of battery health assessment results to address the above technical problems.

[0005] In a first aspect, the present application provides a method for predicting a battery health state, comprising:

[0006] Obtain battery data of the target battery under multiple operating conditions;

[0007] Input the battery data into a pre-built feature extraction network to extract feature vectors that represent the target battery's operating behavior;

[0008] The feature vector is input into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery.

[0009] In a second aspect, the present application further provides a device for predicting battery health status, comprising:

[0010] An acquisition module, used to obtain battery data of a target battery under multiple operating conditions;

[0011] An extraction module, which is used to input battery data into a pre-built feature extraction network to extract feature vectors that represent the operating behavior of the target battery;

[0012] The processing module is used to input the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery.

[0013] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0014] Obtain battery data of the target battery under multiple operating conditions;

[0015] Input the battery data into a pre-built feature extraction network to extract feature vectors that represent the target battery's operating behavior;

[0016] The feature vector is input into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery.

[0017] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0018] Obtain battery data of the target battery under multiple operating conditions;

[0019] Input the battery data into a pre-built feature extraction network to extract feature vectors that represent the target battery's operating behavior;

[0020] The feature vector is input into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery.

[0021] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0022] Obtain battery data of the target battery under multiple operating conditions;

[0023] Input the battery data into a pre-built feature extraction network to extract feature vectors that represent the target battery's operating behavior;

[0024] The feature vector is input into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery.

[0025] The above-mentioned battery health status prediction method, device and computer equipment first collect the operating data of the target battery under various operating conditions to ensure the comprehensiveness and representativeness of the data and avoid the limitations of prediction in a single environment. Subsequently, a pre-built feature extraction network is used to automatically and efficiently extract features from complex battery data to obtain feature vectors that can accurately reflect the battery's operating behavior. Next, a multi-layer sequentially connected nonlinear function combination network is used to apply a learnable piecewise polynomial activation function to the extracted feature vectors dimension by dimension, fully capturing the complex nonlinear relationship between battery status and health indicators. Through progressive layer-by-layer nonlinear transformation and optimization, the expressive power and robustness of the model are effectively improved. It can be seen that this method realizes systematic processing from data acquisition to nonlinear modeling, significantly enhances the prediction accuracy of battery health status, and helps to improve the accuracy and reliability of lithium-ion battery health status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 is a diagram of the internal structure of a computer device in one embodiment;

[0028] Figure 2 1 is a flow chart of a method for predicting battery health status in one embodiment;

[0029] Figure 3 is a flow chart of a method for predicting battery health status in another embodiment;

[0030] Figure 4 is a flow chart of a method for predicting battery health status in another embodiment;

[0031] Figure 5 is a flow chart of a method for predicting battery health status in another embodiment;

[0032] Figure 6 is a flow chart of a method for predicting battery health status in another embodiment;

[0033] Figure 7 is a flow chart of a method for predicting battery health status in another embodiment;

[0034] Figure 8 is a flow chart of a method for predicting battery health status in another embodiment;

[0035] Figure 9 is a flow chart of a method for predicting battery health status in another embodiment;

[0036] Figure 10 FIG. 4 is a structural block diagram of a device for predicting battery health status in one embodiment. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0038] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 1 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data in the process of predicting the battery health status. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting the battery health status is implemented.

[0039] Those skilled in the art will understand that Figure 1 The 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 computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0040] In an exemplary embodiment, Figure 2 As shown, a method for predicting the health status of a battery is provided. Figure 1 The computer device in the embodiment is used as an example to illustrate the method, including the following steps 201 to 203.

[0041] Step 201: Obtain battery data of a target battery under multiple operating conditions.

[0042] Battery data refers to data and information related to the battery's operating status collected during operation, including but not limited to parameters such as voltage, current, temperature, electrochemical impedance, charge and discharge capacity, internal resistance, and SOC (State of Charge). Battery data can be acquired under a variety of typical operating conditions, such as constant current charging, constant voltage discharge, and dynamic power loads, to reflect the battery's performance and behavioral characteristics.

[0043] Multiple operating conditions refer to the various operating environments and modes that a battery may encounter in actual applications, including varying temperature environments, charge and discharge rates, load variations, and initial SOC levels. Collecting data from multiple operating conditions helps improve the generalization and predictive accuracy of subsequent modeling methods.

[0044] In an embodiment of the present application, a computer device acquires battery data from a target battery under multiple operating conditions. Specifically, the battery data can be obtained by collecting performance parameters of the battery under various operating conditions, such as temperature, load, current, voltage, and SOC. The data includes, but is not limited to, the battery's voltage, current, temperature, internal resistance, and electrochemical impedance spectroscopy. By collecting data under multiple operating conditions, the battery's operating behavior in a comprehensive manner can be characterized in its actual operating environment.

[0045] Step 202: Input the battery data into a pre-built feature extraction network to extract a feature vector representing the operating behavior of the target battery.

[0046] The feature extraction network is a neural network structure used to process raw battery data and extract potential temporal or spatial features. It typically uses a convolutional neural network (CNN), a recurrent neural network (RNN), or a modified version of one or more of these structures. This network can model the key behavioral patterns of batteries under multi-dimensional inputs and output a low-dimensional feature representation vector for subsequent analysis.

[0047] A feature vector is a multidimensional numerical combination of values ​​output by the feature extraction network that characterizes the battery's operating behavior. It reflects the key attributes of the battery in its current operating state. This feature vector serves as input to subsequent health prediction models, helping to accurately characterize the evolution of the battery's state.

[0048] In an embodiment of the present application, a computer device inputs the battery data into a pre-built feature extraction network to extract feature vectors that characterize the operating behavior of the target battery. The feature extraction network can use a convolutional neural network, a recursive neural network, or other deep learning structure that can mine the temporal changes in battery status to reduce the dimensionality and abstract the input multi-dimensional battery data to extract key feature information. The output feature vector can effectively reflect the potential health trends and changing status of the battery.

[0049] Step 203 : Input the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain a health status prediction result of the target battery.

[0050] The nonlinear function combination network is a deep neural network structure composed of multiple processing layers, used to model nonlinear mappings of input feature vectors. By introducing learnable nonlinear activation functions in each layer, the network can approximate and model the complex relationship between input and target output.

[0051] The health status prediction result refers to the output result obtained through nonlinear function combination network modeling, which is used to reflect the current or future operating status of the target battery. It is usually given in the form of SOH (State of Health) or related evaluation indicators to guide battery maintenance, replacement or system scheduling.

[0052] In an embodiment of the present application, a computer device inputs the feature vector into a pre-built nonlinear function combination network for nonlinear processing. This network is composed of multiple sequentially connected processing layers, and each layer introduces a learnable nonlinear activation function to model complex nonlinear mapping of the feature vector. Through training, the network parameters are optimized to adaptively fit the nonlinear relationship between the input features and the target health state. Finally, the health state prediction result of the target battery is output.

[0053] In the above-mentioned battery health status prediction method, the operating data of the target battery under various operating conditions is first collected to ensure the comprehensiveness and representativeness of the data, avoiding the limitations of prediction in a single environment; then, a pre-built feature extraction network is used to automatically and efficiently extract features from complex battery data to obtain feature vectors that can accurately reflect the battery's operating behavior; then, a multi-layer sequentially connected nonlinear function combination network is used to apply a learnable piecewise polynomial activation function to the extracted feature vectors dimension by dimension, fully capturing the complex nonlinear relationship between battery status and health indicators; through progressive layer-by-layer nonlinear transformation and optimization, the expressive power and robustness of the model are effectively improved. It can be seen that this method realizes systematic processing from data acquisition to nonlinear modeling, significantly enhances the prediction accuracy of battery health status, and helps to improve the accuracy and reliability of lithium-ion battery health status assessment.

[0054] In an exemplary embodiment, the nonlinear function combination network includes a plurality of sequentially connected processing layers, on this basis, such as Figure 3 As shown, the above-mentioned "inputting the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain the health status prediction result of the target battery" includes steps 301 to 302. Among them:

[0055] Step 301: Input the feature vector into the first processing layer for nonlinear processing to obtain a first output vector.

[0056] A processing layer is the basic structural unit in a nonlinear function combination network. Each processing layer receives the output vector of the previous layer as input and performs nonlinear activation functions, linear transformations (such as weighted summation), or other operations on this input to generate the output vector of this layer. Stacking multiple processing layers sequentially forms a deep neural network architecture, which enhances the model's expressive power.

[0057] A feature vector is a numerical vector generated by a feature extraction network that represents the battery's operating behavior. It typically consists of multiple dimensions, each corresponding to a key operating characteristic, such as temperature variation, impedance characteristics, or voltage fluctuations. This serves as the input for subsequent neural network modeling.

[0058] In an embodiment of the present application, a computer device inputs the target battery feature vector generated by the feature extraction network into the first processing layer of the nonlinear function combination network. The feature vector includes multiple dimensional values, each dimension representing a certain type of timing feature or state quantity during the operation of the battery, such as voltage change trend, impedance response characteristics, current fluctuation pattern, etc. After the first processing layer receives the feature vector, it performs a nonlinear mapping operation on the values ​​of each dimension in the vector, and transforms it through an activation function to enhance the network's ability to express complex nonlinear relationships between features. The values ​​of each dimension after nonlinear mapping are then combined according to the vector structure to form the output vector of the first processing layer.

[0059] In step 302, the i-th output vector is input to the next level processing layer for nonlinear processing to obtain the i+1-th output vector; wherein i is a positive integer, and the output vector of the last level processing layer is the health status prediction result of the target battery.

[0060] The i-th output vector represents the processing result of the i-th processing layer. In a nonlinear function combination network, each layer transforms the input vector and outputs an intermediate result, which is the i-th output vector. i is a positive integer index representing the number of processing layers.

[0061] Health status prediction results are indicators of the target battery's current health level, typically expressed as numerical values, derived from feature extraction and nonlinear modeling. These indicators reflect key performance parameters such as capacity retention and internal resistance change. These predictions can be used to guide battery maintenance, lifespan assessment, or fault warning.

[0062] In this embodiment of the present application, the output vector of the first processing layer is passed as input to the next processing layer. The second processing layer performs nonlinear transformations and combination processing on each dimension of the input vector in the same manner to obtain a new intermediate output vector. The entire nonlinear function combination network is composed of multiple processing layers connected in series. The structure and function of each processing layer are consistent, and all include two sub-steps: nonlinear mapping and vector construction, which are used to extract deeper and more abstract high-order nonlinear features.

[0063] During the processing of the nonlinear function combination network, the computer equipment continuously iterates the "input vector, nonlinear mapping, and output vector" operations. The output of each layer incorporates the transformation results of the previous layer's features and introduces more complex feature associations on this basis. After multi-layer processing, the network can fully learn the nonlinear pattern characteristics contained in the target battery's operating data under various operating conditions. This multi-layer nonlinear modeling structure can effectively capture the implicit interactive relationships between features, thereby improving prediction accuracy.

[0064] Ultimately, after layer-by-layer nonlinear processing across all processing layers, the final output vector represents the target battery's state of health prediction. This output can be used to characterize the battery's current health, reflecting its capacity decay, performance degradation, and other indicators related to its service life, demonstrating strong application feasibility and potential for widespread adoption. The entire prediction process requires no manual feature or rule definition, relying entirely on the network's ability to model complex relationships, enabling automated, highly accurate prediction of battery health.

[0065] In an exemplary embodiment, Figure 4 As shown, the above-mentioned “inputting the i-th output vector into the next level processing layer for nonlinear processing to obtain the i+1-th output vector” includes steps 401 to 402.

[0066] in:

[0067] In step 401, each dimension value in the i-th output vector is input into the next level processing layer for activation processing to obtain the dimension transformation result corresponding to each dimension value.

[0068] The dimensionality transformation result refers to the new mapping of the values ​​of each dimension after the activation process, that is, the output after the nonlinear function transformation. Each dimensionality transformation result corresponds to the nonlinear response value of a specific dimension in the input vector, which is used to reflect the processing effect of that feature dimension at the current processing layer.

[0069] In an embodiment of the present application, the computer device inputs the feature vector into a first processing layer for nonlinear processing. In the first processing layer, a learnable piecewise polynomial activation function is applied to each dimension value in the input vector, performing dimension-level nonlinear mapping to obtain a dimensional transformation result corresponding to each dimension value. Here, each dimension value refers to each scalar feature value in the input vector, representing the state information of the target battery in a specific operating dimension; the dimensional transformation result is the nonlinear response value generated by the nonlinear mapping of the value.

[0070] During the activation process, the activation function used is a learnable B-spline function, which is composed of multiple basis functions, and multiple control points control the local curve shape of the function in different intervals. Through the training process, the control point positions and basis function weight parameters are jointly optimized, allowing the activation function to adaptively adjust its output response during training, effectively fitting the complex nonlinear relationship between features and target variables.

[0071] Step 402 : Perform vector-level combination processing on the dimensional transformation results corresponding to the multiple dimensional values ​​to obtain the (i+1)th output vector.

[0072] In this embodiment of the present application, the computer device combines all dimensional transformation results in the order in which they appear in the original input vector, i.e., performs vector-level combination processing to construct a new output vector. This vector retains the dimensional structure of the input vector while incorporating the nonlinear features extracted by the current processing layer, and serves as the output result of that processing layer.

[0073] The output vector of the first processing layer is fed into the next processing layer, repeating the nonlinear mapping and combination steps. In the i-th processing layer, the computer takes the i-1-th output vector as input and applies a nonlinear activation function to each dimension of the vector to extract deeper feature representations. This process continues sequentially through each processing layer, forming a multi-layered, nested nonlinear modeling structure.

[0074] Ultimately, after processing through all layers, the output vector of the final processing layer is the health status prediction result of the target battery. This prediction result is used to quantify the current health level of the target battery and reflect its performance degradation trend, providing decision support for battery system maintenance, scheduling, and life management.

[0075] In an exemplary embodiment, Figure 5 As shown, the above-mentioned "performing vector-level combination processing on the dimensional transformation results corresponding to the multiple dimensional values ​​to obtain the (i+1)th output vector" includes steps 501 to 502. Among them:

[0076] Step 501 : Arrange and sort the dimension transformation results according to the position order of the values ​​of each dimension in the i-th output vector to obtain an arranged result.

[0077] In an embodiment of the present application, the computer device sorts and arranges the dimensional transformation results obtained after the nonlinear activation function transformation according to the original position order of the dimensional values ​​in the input vector. Specifically, each dimensional transformation result corresponds to a specific feature dimension in the input vector, so its order consistency in the vector structure must be maintained to ensure data alignment between the upper and lower layers. This operation results in an ordered arrangement, which is called a sorted result.

[0078] Step 502 : Perform vector construction processing on the sorting result to obtain the (i+1)th output vector; the (i+1)th output vector has a fixed-dimensional structure.

[0079] In this embodiment of the present application, the computer device performs vector construction on the above-mentioned sorting results, that is, sequentially concatenating the sorted dimensional transformation results to construct a new vector with a unified structure. This output vector maintains the same dimensionality as the input vector of the current processing layer and has a fixed dimensional structure, making it easier to continue serving as input for deep feature modeling in subsequent processing layers.

[0080] Finally, the output vector is passed to the next processing layer of the nonlinear function combination network as the output result of the current processing layer, that is, the i+1th output vector. Through these steps, while maintaining input and output consistency, nonlinear features are extracted layer by layer to complete the health status prediction model of the target battery's operating behavior.

[0081] In an exemplary embodiment, Figure 6 As shown, the above-mentioned "inputting each dimension value in the i-th output vector into the next level processing layer for activation processing to obtain the dimension transformation result corresponding to each dimension value" includes steps 601 to 602. Among them:

[0082] Step 601: For each dimension value in the i-th output vector, the dimension value is input as an independent variable into the corresponding piecewise polynomial function, and based on the position of the control point in the piecewise polynomial function and the weight of the basis function, an initial nonlinear output result corresponding to the dimension value is generated; the piecewise polynomial activation function is a learnable B-spline function, which is composed of multiple basis functions, and each basis function corresponds to a control point, and the control point is used to fit the nonlinear change of the dimension value in the corresponding area.

[0083] The dimension value refers to the value corresponding to each dimension in the input vector or intermediate output vector. In this method, the feature vector consists of multiple dimensions, each of which represents a quantitative expression of a specific characteristic of the battery, such as voltage, current, impedance, temperature, and other parameters.

[0084] A piecewise polynomial function is a function composed of multiple polynomial segments, each defined within a specific interval, and is used to perform a nonlinear mapping of input data. In this method, this function is used as an activation function, applying a separate nonlinear transformation to each input dimension to enhance the model's ability to fit complex relationships.

[0085] B-spline functions are constructed from a set of basis functions. Each basis function operates within a specific interval, and its shape and position are determined by a corresponding control point. By adjusting the positions of the control points and the weights of the basis functions, precise modeling of nonlinear mapping relationships can be achieved.

[0086] Basis functions are the basic building blocks of B-spline functions. Each basis function has a non-zero value in a small part of the domain and is zero elsewhere. Therefore, B-spline functions have locality and can independently control different segments of the input.

[0087] Control points are key parameters that define the shape of a B-spline function. Each control point influences the direction or curvature of the function within a specific region. By adjusting the position of the control points, you can flexibly control the local fit of the nonlinear function.

[0088] The initial nonlinear output result refers to the function output value obtained after the value of each dimension is nonlinearly mapped by its corresponding B-spline function, which reflects the nonlinear change trend of the input feature in the current dimension.

[0089] In an embodiment of the present application, the computer device processes each dimension value in the i-th output vector separately. Specifically, the dimension value is input as an independent variable into the corresponding piecewise polynomial function, and the piecewise polynomial function is a learnable B-spline function. The B-spline function is composed of multiple basis functions, each basis function corresponds to a control point, and the position of the control point and the weight parameter of the basis function jointly determine the shape of the function in the corresponding interval, so that the nonlinear change of the dimension value within its defined interval can be accurately fitted. In this way, the computer device calculates the initial nonlinear output result corresponding to the dimension value based on the structure and parameters of the B-spline function.

[0090] Step 602 : performing scale normalization and vector construction processing on the multiple initial nonlinear output results to obtain a dimensionality transformation result corresponding to the i-th output vector.

[0091] Among them, scale normalization refers to the unified processing of the numerical range of multiple initial nonlinear output results, usually mapping them to [0, 1] or other unified intervals to eliminate the differences in numerical sizes between data of different dimensions, facilitating subsequent vector combination and model training.

[0092] Vector construction involves recombining the nonlinear outputs of the normalized dimensions into a new vector in the order of the original dimensions. This process ensures the consistency of feature order, allowing the output of each processing layer to serve as the standardized input for the next layer.

[0093] The dimensionality transformation result is the new output vector formed after all the dimensional values ​​in the input vector are nonlinearly mapped using the B-spline function, normalized, and constructed. This vector will serve as the input to the next processing layer to complete deep feature extraction.

[0094] In an embodiment of the present application, the computer device performs a uniform scale normalization process on the initial nonlinear output results of all dimensions. The purpose of this step is to eliminate the differences in the numerical ranges between different dimensions, making subsequent data processing more stable and improving the robustness of model training and prediction. Normalization typically involves mapping the values ​​to a standard interval, such as [0, 1] or a distribution with a mean of 0 and a variance of 1.

[0095] After normalization, the computer arranges the normalized transformation results in the order of the dimensions in the original input vector, forming a vector with a fixed dimensional structure. This vector undergoes vector construction and becomes the output vector of the i-th processing layer, representing the dimensional transformation result of that layer. This entire process ensures that the input features undergo a meticulous nonlinear mapping and normalization transformation, effectively capturing complex nonlinear relationships and providing high-quality feature representation for deep learning in subsequent layers.

[0096] In an exemplary embodiment, Figure 7 As shown, the above method also includes:

[0097] In step 701, each reference dimension value in the reference input vector is input as an independent variable into the piecewise polynomial function corresponding to the reference dimension value, and a reference nonlinear output result corresponding to the reference dimension value is generated based on the position of the control point in the piecewise polynomial function and the weight of the basis function.

[0098] In this embodiment of the present application, the computer device first inputs the numerical value of each dimension of the reference input vector as an independent variable into the piecewise polynomial function corresponding to that dimension. During this process, the piecewise polynomial function maps the input numerical value based on the control point positions and weight parameters of each basis function set within it, generating a corresponding reference nonlinear output result.

[0099] Step 702: Compare the reference nonlinear output result with the expected output result to calculate the target error.

[0100] In an embodiment of the present application, the computer device compares the reference nonlinear output result with the predefined expected output result item by item to calculate the error between the two. The error is the target error, which reflects the deviation between the current function mapping and the ideal result.

[0101] Step 703 : Based on the target error, gradient update or optimization processing is performed on the control point positions and corresponding basis function weight parameters in the piecewise polynomial function to obtain an optimized piecewise polynomial function.

[0102] In an embodiment of the present application, based on the calculated target error, the computer device can use the gradient descent method or other appropriate optimization algorithm to adjust and optimize the position of each control point in the piecewise polynomial function and the weight parameters of the corresponding basis function. Through this parameter update mechanism based on error feedback, the shape of the piecewise polynomial function is gradually improved, so that it can more accurately capture the complex nonlinear relationship between the input data and the target output. This optimization process is repeated through multiple iterations until the target error reaches the preset convergence condition, and finally an optimized piecewise polynomial function with excellent performance is obtained, thereby improving the prediction accuracy and generalization ability of the overall system.

[0103] In an exemplary embodiment, the feature extraction network includes at least two one-dimensional convolutional layers and at least one one-dimensional pooling layer. On this basis, Figure 8 As shown, the above-mentioned "inputting battery data into a pre-built feature extraction network to extract a feature vector representing the target battery operating behavior" includes steps 801 to 802. Among them:

[0104] Step 801 : Perform multiple convolution operations on the battery data using at least two one-dimensional convolution layers to extract local time series features of the target battery's operating state.

[0105] In an embodiment of the present application, the computer device first obtains battery data of the target battery under multiple operating conditions. These data generally include multi-dimensional information such as the battery's voltage, current, temperature, and electrochemical impedance. Subsequently, the computer device inputs the collected battery data into a pre-built feature extraction network, which is mainly composed of at least two one-dimensional convolutional layers and at least one one-dimensional pooling layer. In the one-dimensional convolutional layer, the computer device performs multiple convolution operations on the input data through a sliding convolution kernel to mine the local timing characteristics of the battery's operating status, thereby capturing the dynamic change pattern of the battery under different operating conditions.

[0106] Step 802 : Using a one-dimensional pooling layer to perform dimensionality reduction processing on the local time series features, and using a maximum pooling operation to generate a feature vector representing the operating behavior of the target battery.

[0107] In an embodiment of the present application, the computer device uses a one-dimensional pooling layer to perform dimensionality reduction processing on the extracted time series features, adopts a maximum pooling operation, selects the maximum value of the local area, and reduces the data dimension. After processing by the convolution layer and the pooling layer, a feature vector representing the operating behavior of the target battery is generated. The computer device inputs the feature vector into a pre-built nonlinear function combination network, and uses multiple sequentially connected processing layers to perform a nonlinear transformation on the feature vector. Specifically, the feature vector is input into the first processing layer, and a learnable piecewise polynomial activation function is applied to obtain the first output vector. Subsequently, the i-th output vector is input into the i+1-th processing layer, and the nonlinear transformation is repeatedly applied until the last layer generates a health status prediction result for the target battery.

[0108] In the above embodiment, the computer device uses a one-dimensional pooling layer to perform dimensionality reduction processing on the local time series features extracted by the convolution layer. The maximum pooling method is usually adopted, that is, the maximum value of the feature is selected within a certain window to reduce the data dimension and computational complexity, while maintaining the significance and robustness of the feature. The pooling process helps to reduce the impact of noise and improve the expressiveness of the feature. After the cascade processing of multiple layers of convolution and pooling, the system finally generates a low-dimensional feature vector that can fully characterize the operating behavior of the target battery. This feature vector not only contains the timing information of the battery operation, but also integrates the multi-dimensional state change characteristics, providing a rich and effective input data basis for the subsequent nonlinear function combination network to achieve accurate prediction of the battery health status.

[0109] In an exemplary embodiment, the battery data at least includes electrochemical impedance data. Figure 9 As shown, the above method further includes steps 901 to 902. Among them:

[0110] Step 901 : For each measurement data in the electrochemical impedance data, perform an integral transformation on the real part and the imaginary part of the measurement data to obtain a transformation result.

[0111] Among them, electrochemical impedance data refers to the battery electrochemical characteristic data obtained by measuring electrochemical impedance spectroscopy technology, which usually includes the real and imaginary parts of the impedance at different frequencies and is used to reflect the electrochemical process and state inside the battery.

[0112] Electrochemical impedance is a complex number consisting of a real part (corresponding to the resistance component) and an imaginary part (corresponding to the capacitance and inductive reactance components). The real and imaginary parts together describe the impedance characteristics of the battery at different frequencies.

[0113] Integral transformation processing usually refers to the use of an integral transformation method based on the Cauchy-Kronig (KK) relationship to perform mathematical integration operations on the real and imaginary parts of the electrochemical impedance to verify the causality and consistency of the data and ensure the physical rationality of the measured data.

[0114] The transformation result refers to the impedance value calculated by the integral transformation, which should ideally be highly consistent with the original measurement data.

[0115] In the present embodiment, a computer device first receives and reads measurement data obtained from a battery electrochemical impedance test. Each piece of data includes the real and imaginary parts of the impedance. For each piece of measurement data, the computer device uses an integral transformation algorithm based on the KK relationship to perform mathematical integration calculations on the real and imaginary parts of the impedance to obtain the corresponding transformation results.

[0116] Step 902 : When the difference between the transformation result and the measurement data is outside a preset tolerance range, the measurement data is eliminated or corrected to obtain corrected measurement data.

[0117] The preset tolerance range is a set numerical range used to determine whether the difference between the transformation result and the original measurement data is within a reasonable range. If it exceeds this range, it indicates that the data may be abnormal.

[0118] Elimination or correction processing is to screen out the measurement data that does not meet the tolerance standards or correct it through interpolation, fitting and other methods to ensure the accuracy of the data and the reliability of subsequent analysis.

[0119] The corrected measurement data is electrochemical impedance spectroscopy data that conforms to physical laws and can be used for subsequent analysis after elimination or correction.

[0120] In an embodiment of the present application, a computer device compares the real and imaginary parts of the transformation result with the original measurement data and calculates the error between the two. If the error is within a pre-set tolerance range, the data is determined to conform to physical laws and can be used as valid data; if the error exceeds the tolerance range, the measurement data is considered to be abnormal. For abnormal data, the computer device can perform a elimination operation to directly exclude it from the data set, or use numerical correction methods such as interpolation, fitting or filtering to correct the abnormal data to make it conform to physical rationality. Through this process, the computer device outputs the electrochemical impedance data after elimination and correction, ensuring that the data used for subsequent battery state analysis and model training is accurate and reliable.

[0121] In an exemplary embodiment, the method further includes:

[0122] 1. Build the test platform and collect data.

[0123] First, a battery charge and discharge test platform is built to collect battery data under different working conditions, including time data, electrochemical impedance data, battery capacity decay data, voltage and current data, and battery health status data. Among them, the electrochemical impedance data needs to be verified by KK transformation to ensure the causal and linear response characteristics of the data. The principle formula of KK transformation is shown in formula (1):

[0124]

[0125] In formula (1), z represents the electrochemical impedance and w represents the angular frequency. This transformation screens valid data by verifying the mathematical relationship between the real and imaginary parts of the impedance.

[0126] 2. Data Cleaning Processing

[0127] The collected initial sample data is cleaned, and the operations include filling missing values, deleting outliers, removing duplicate values, eliminating data noise, and normalizing the data to obtain a high-quality battery data set, providing reliable input for subsequent modeling.

[0128] 3. Feature Extraction of Convolutional Neural Networks (CNN)

[0129] A CNN neural network consisting of two one-dimensional convolutional layers and two one-dimensional pooling layers is constructed. The cleaned battery data is subjected to convolution operations through the one-dimensional convolutional layers to extract the local time series features of the target battery's operating status (such as voltage fluctuation trends and capacity attenuation patterns). The calculation formula is shown in formula (2):

[0130]

[0131] where y l is the output of the lth convolutional layer, ReLU is the linear rectification activation function, b l is the bias term, is the convolution kernel weight, x l-1 is the input data of the previous layer, and n is the number of convolution kernels. The local features obtained by convolution are then reduced in dimension through a one-dimensional pooling layer, and a maximum pooling operation is used to generate a feature vector representing the battery operation behavior, thereby reducing the computational complexity of the model. The calculation formula is shown in formula (3):

[0132] p l =MAX Pool(y l ) (3)

[0133] Among them, p l is the output of the pooling layer, y l is the output of the previous convolution layer.

[0134] 4. KAN neural network health status prediction

[0135] The KAN (Kolmogorov–Arnold Networks) neural network is stacked after the CNN neural network. The network is designed based on the Kolmogorov–Arnold representation theorem. The theorem states that any multivariable continuous function can be decomposed into a finite combination of single-variable functions, as shown in formula (4):

[0136]

[0137] Where f(x) is a multivariable function, x1,…,x n is the input variable, φ q and φ q,p (x p ) is a single variable activation function, and m is the number of function combinations. The KAN network parameterizes the activation function through a learnable B-spline function, decomposing the high-dimensional function into a one-dimensional function combination. The B-spline function expression is shown in formula (5):

[0138] spline(x)=∑ i c i B i (x) (5)

[0139] Among them, spline(x) is the output of the spline function, c i is the optimization coefficient, B i The network consists of three KANLinear layers, with spline interpolation used between layers for feature transformation. The Adam optimizer is used to optimize parameters, ultimately completing the battery health status prediction based on the extracted features.

[0140] In an exemplary embodiment, the method further includes:

[0141] Step 1: Obtain battery data of a target battery under multiple operating conditions.

[0142] Step 2: For each measurement data in the electrochemical impedance data, perform an integral transformation on the real part and the imaginary part of the measurement data to obtain a transformation result.

[0143] Step 3: When the difference between the transformation result and the measurement data is outside the preset tolerance range, the measurement data is eliminated or corrected to obtain corrected measurement data.

[0144] Step 4: Use at least two one-dimensional convolutional layers to perform multiple convolution operations on the battery data to extract local time series features of the target battery operating status.

[0145] In step 5, a one-dimensional pooling layer is used to reduce the dimensionality of the local time series features, and a maximum pooling operation is used to generate a feature vector that characterizes the operating behavior of the target battery.

[0146] Step 6: Input the feature vector into the first processing layer for nonlinear processing to obtain the first output vector.

[0147] Step 7: For each dimension value in the i-th output vector, the dimension value is input as an independent variable into the corresponding piecewise polynomial function, and based on the position of the control point in the piecewise polynomial function and the weight of the basis function, the initial nonlinear output result corresponding to the dimension value is generated; the piecewise polynomial activation function is a learnable B-spline function, which is composed of multiple basis functions, and each basis function corresponds to a control point, which is used to fit the nonlinear change of the dimension value in the corresponding area.

[0148] Step 8: Perform scale normalization and vector construction processing on the multiple initial nonlinear output results to obtain the dimension transformation result corresponding to the i-th output vector.

[0149] Step 9: Arrange the dimension transformation results according to the position order of the values ​​of each dimension in the i-th output vector to obtain an arranged result.

[0150] Step 10: Perform vector construction processing on the sorted results to obtain the i+1th output vector; the i+1th output vector has a fixed-dimensional structure; wherein i is a positive integer, and the output vector of the last level of processing layer is the health status prediction result of the target battery.

[0151] Step 11: Input each reference dimension value in the reference input vector as an independent variable into the piecewise polynomial function corresponding to the reference dimension value, and generate a reference nonlinear output result corresponding to the reference dimension value based on the position of the control point in the piecewise polynomial function and the weight of the basis function.

[0152] Step 12: Compare the reference nonlinear output result with the expected output result and calculate the target error;

[0153] Step 13: Based on the target error, gradient update or optimization processing is performed on the control point positions and corresponding basis function weight parameters in the piecewise polynomial function to obtain an optimized piecewise polynomial function.

[0154] It should be understood that, although the steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0155] Based on the same inventive concept, embodiments of the present application also provide a device for predicting the battery health state for implementing the aforementioned method for predicting the battery health state. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the device for predicting the battery health state provided below can be found in the aforementioned limitations of the method for predicting the battery health state, and will not be repeated here.

[0156] In an exemplary embodiment, Figure 10 As shown, a device for predicting the health status of a battery is provided, comprising: an acquisition module 1001, an extraction module 1002, and a processing module 1003, wherein:

[0157] An acquisition module 1001 is configured to acquire battery data of a target battery under multiple operating conditions;

[0158] Extraction module 1002, used to input battery data into a pre-built feature extraction network to extract feature vectors representing the operating behavior of the target battery;

[0159] The processing module 1003 is used to input the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain a health status prediction result of the target battery.

[0160] In an exemplary embodiment, the processing module 1003 is specifically configured to input the feature vector into a first processing layer for nonlinear processing to obtain a first output vector;

[0161] The i-th output vector is input into the next level processing layer for nonlinear processing to obtain the i+1-th output vector; where i is a positive integer, and the output vector of the last level processing layer is the health status prediction result of the target battery.

[0162] In an exemplary embodiment, the processing module 1003 is specifically configured to input each dimension value in the i-th output vector into the next level processing layer for activation processing, thereby obtaining a dimension transformation result corresponding to each dimension value;

[0163] Perform vector-level combination processing on the dimensional transformation results corresponding to multiple dimensional values ​​to obtain the i+1th output vector.

[0164] In an exemplary embodiment, the processing module 1003 is specifically configured to sort and arrange the dimensionality transformation results according to the position order of the dimensional values ​​in the i-th output vector to obtain a sorted result;

[0165] The sorting result is vectorized to obtain an i+1th output vector; the i+1th output vector has a fixed-dimensional structure.

[0166] In an exemplary embodiment, the processing module 1003 is specifically configured to input, for each dimension value in the i-th output vector, the dimension value as an independent variable into a corresponding piecewise polynomial function, and generate an initial nonlinear output result corresponding to the dimension value based on the position of the control point and the weight of the basis function in the piecewise polynomial function; the piecewise polynomial activation function is a learnable B-spline function, which is composed of multiple basis functions, and each basis function corresponds to a control point, and the control point is used to fit the nonlinear change of the dimension value in the corresponding region;

[0167] The multiple initial nonlinear output results are scaled and normalized and vectorized to obtain the dimension transformation result corresponding to the i-th output vector.

[0168] In an exemplary embodiment, the battery health status prediction device is specifically configured to input each reference dimension value in a reference input vector as an independent variable into a piecewise polynomial function corresponding to the reference dimension value, and generate a reference nonlinear output result corresponding to the reference dimension value based on the position of the control point and the weight of the basis function in the piecewise polynomial function;

[0169] Compare the reference nonlinear output result with the expected output result and calculate the target error;

[0170] Based on the target error, the control point positions and the corresponding basis function weight parameters in the piecewise polynomial function are gradient updated or optimized to obtain the optimized piecewise polynomial function.

[0171] In an exemplary embodiment, the extraction module 1002 is specifically configured to perform multiple convolution operations on the battery data using at least two one-dimensional convolution layers to extract local time series features of the target battery operating state;

[0172] A one-dimensional pooling layer is used to reduce the dimensionality of local time series features, and a maximum pooling operation is used to generate a feature vector that characterizes the operating behavior of the target battery.

[0173] In an exemplary embodiment, the above-mentioned battery health status prediction device is specifically used to perform integral transformation processing on the real part and imaginary part of each measurement data in the electrochemical impedance data to obtain a transformation result; when the difference between the transformation result and the measurement data is outside a preset tolerance range, the measurement data is eliminated or corrected to obtain corrected measurement data.

[0174] Each module in the above-mentioned battery health status prediction device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0175] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0176] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0177] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0178] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and 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, database or other media used in the embodiments provided in this application may include at least one of non-volatile memory and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a programmable logic unit (PLC), a data processing logic unit based on quantum computing, an artificial intelligence (AI) processor, and the like.

[0179] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0180] 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 application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting battery health status, characterized in that: The method comprises: Obtain battery data of the target battery under multiple operating conditions; Inputting the battery data into a pre-built feature extraction network to extract a feature vector representing the operating behavior of the target battery; The characteristic vector is input into a pre-built nonlinear function combination network for nonlinear processing to obtain a health status prediction result of the target battery.

2. The method according to claim 1, characterized in that The nonlinear function combination network includes a plurality of sequentially connected processing layers. Inputting the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain a health status prediction result of the target battery includes: Inputting the feature vector into a first processing layer for nonlinear processing to obtain a first output vector; The i-th output vector is input into the next level processing layer for nonlinear processing to obtain the i+1-th output vector; where i is a positive integer, and the output vector of the last level processing layer is the health status prediction result of the target battery.

3. The method according to claim 2, characterized in that The i-th output vector is input to the next level processing layer for nonlinear processing to obtain the i+1-th output vector, including: Input each dimension value in the i-th output vector into the next level processing layer for activation processing, and obtain the dimension transformation result corresponding to each dimension value; Vector-level combination processing is performed on the dimensional transformation results corresponding to the multiple dimensional values ​​to obtain an (i+1)th output vector.

4. The method according to claim 3, characterized in that The performing vector-level combination processing on the dimensional transformation results corresponding to the plurality of dimensional values ​​to obtain the (i+1)th output vector includes: Arrange the dimensionality transformation results according to the position order of the dimensional values ​​in the i-th output vector to obtain an arranged result; Vector construction processing is performed on the sorting result to obtain the (i+1)th output vector; the (i+1)th output vector has a fixed-dimensional structure.

5. The method according to claim 3, characterized in that The step of inputting each dimension value in the i-th output vector into the next level processing layer for activation processing to obtain a dimension transformation result corresponding to each dimension value includes: For each dimension value in the i-th output vector, the dimension value is input as an independent variable into the corresponding piecewise polynomial function, and based on the position of the control point and the weight of the basis function in the piecewise polynomial function, an initial nonlinear output result corresponding to the dimension value is generated; the piecewise polynomial activation function is a learnable B-spline function, the B-spline function is composed of multiple basis functions, and each basis function corresponds to a control point, and the control point is used to fit the nonlinear change of the dimension value in the corresponding area; Scale normalization and vector construction processing are performed on the multiple initial nonlinear output results to obtain a dimensionality transformation result corresponding to the i-th output vector.

6. The method according to claim 5, characterized in that The method further comprises: Inputting each reference dimension value in the reference input vector as an independent variable into a piecewise polynomial function corresponding to the reference dimension value, and generating a reference nonlinear output result corresponding to the reference dimension value based on the position of the control point in the piecewise polynomial function and the weight of the basis function; Compare the reference nonlinear output result with the expected output result and calculate the target error; Based on the target error, the control point positions and corresponding basis function weight parameters in the piecewise polynomial function are gradient updated or optimized to obtain an optimized piecewise polynomial function.

7. The method according to claim 1, characterized in that The feature extraction network includes at least two one-dimensional convolutional layers and at least one one-dimensional pooling layer. The battery data is input into the pre-built feature extraction network to extract a feature vector representing the operating behavior of the target battery, including: Performing multiple convolution operations on the battery data using at least two of the one-dimensional convolution layers to extract local time series features of the target battery operating state; The one-dimensional pooling layer is used to perform dimensionality reduction processing on the local time series features, and a maximum pooling operation is used to generate a feature vector representing the operating behavior of the target battery.

8. The method according to any one of claims 1 to 7, characterized in that The battery data includes at least electrochemical impedance data, and the method further includes: For each measurement data in the electrochemical impedance data, performing an integral transformation process on the real part and the imaginary part of the measurement data to obtain a transformation result; When the difference between the transformation result and the measurement data is outside a preset tolerance range, the measurement data is eliminated or corrected to obtain corrected measurement data.

9. A device for predicting battery health status, characterized in that: The device comprises: An acquisition module, used to obtain battery data of a target battery under multiple operating conditions; An extraction module, configured to input the battery data into a pre-built feature extraction network to extract a feature vector representing the operating behavior of the target battery; The processing module is used to input the feature vector into a pre-built nonlinear function combination network for nonlinear processing to obtain a health status prediction result of the target battery.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images