Lithium Battery SOH Estimation Method for Intelligent Yellow River Underwater Vehicle

Abstract

A method for estimating the State of Health (SOH) of lithium batteries for intelligent underwater vehicles in the Yellow River includes the following steps: obtaining a lithium battery dataset under four different working conditions; analyzing the lithium battery dataset from the perspective of battery attributes to construct attribute features; analyzing the attribute features from the perspective of data range to construct data range features; analyzing the data range features from the perspective of mathematical statistics to construct mathematical features; combining the attribute features, data range features, and mathematical features to form a feature library; using the highest correlation search method to select the highest correlation features from the feature library to obtain model input features; sending the model input features into a dual-channel interpretable network to obtain the SOH estimation result. The present invention can solve the problem of inputting effective features of lithium batteries, obtain a rich feature library through multi-angle analysis of the battery dataset, realize automatic transparency for feature selection through the highest correlation search method, and cooperate with the dual-channel interpretable network to improve the prediction accuracy and calculation efficiency of the SOH estimation model.

Description

Technical Field

[0001] The present invention relates to the technology for estimating the state of health of lithium-ion batteries in the field of new energy, and particularly to a method for estimating the SOH of lithium batteries for intelligent Yellow River underwater vehicles. Background Art

[0002] The ecological environment of the Yellow River Basin is complex and changeable. Achieving full-time and full-domain ecological dynamic monitoring and governance of the Yellow River is an important task for building the "Intelligent Yellow River" digital platform. For the real-time monitoring of the complex underwater environment of the Yellow River, autonomous underwater vehicles (AUVs) have gradually become the core equipment of the "Intelligent Yellow River" perception platform due to their high flexibility and wide coverage. However, the long-term autonomous underwater operation of AUVs poses higher requirements for battery performance and safety, especially in the fast-flowing and changeable water environment. As the core power source, the state of health (SOH) of lithium batteries directly determines the monitoring efficiency and stability of the autonomous vehicle network system.

[0003] Currently, the methods for estimating the state of health of lithium batteries usually rely on manual experience for feature selection, with low automation and a lack of transparency in the estimation process, making it difficult to reveal the mechanism of battery state changes, thus affecting the reliability of underwater monitoring tasks. At the same time, the complex flow patterns and frequent changes in environmental conditions in the Yellow River waters have a significant impact on battery performance. There is an urgent need for an automatic and transparent feature selection method to accurately extract and track the key features of battery health status, improving the accuracy, generalization ability, and processing efficiency of battery state estimation.

[0004] Therefore, to meet the construction requirements of the "Intelligent Yellow River" project, there is an urgent need to develop a method for estimating the state of health of lithium batteries that is applicable to the SD-AUV network and can achieve automatic and transparent feature selection. Through an automatic and transparent feature selection mechanism, the present invention effectively captures the historical data dependencies and recent trends of the battery, realizes accurate estimation of the SOH of lithium batteries, thereby improving the safety, reliability, and long-term operation performance of the underwater autonomous vehicle network system, and providing important technical support and guarantee for the underwater three-dimensional perception and ecological monitoring of the "Intelligent Yellow River". Summary of the Invention

[0005] In response to the needs in the prior art, the present invention provides a method for estimating the SOH of lithium batteries for intelligent Yellow River underwater vehicles, aiming to solve the problem of automatic and transparent feature selection in the task of estimating the SOH of lithium batteries, dealing with the recent trends and historical dependencies of battery data, selecting favorable features, accelerating the network processing efficiency, and providing new ideas for the field of estimating the SOH of lithium batteries.

[0006] The method for estimating the SOH of lithium batteries for intelligent Yellow River underwater vehicles includes the following steps:

[0007] Step 1: Establish a lithium-ion battery cycle life test platform to obtain a lithium battery dataset under four different working conditions;

[0008] Step 2: Analyze and process the lithium battery dataset to construct a feature library, specifically:

[0009] Step 2.1: Analyze the lithium battery dataset from the perspective of battery attributes to construct attribute features;

[0010] Step 2.2: Analyze the attribute features from the perspective of data range to construct data range features;

[0011] Step 2.3: Analyze the data range features from the perspective of mathematical statistics to construct mathematical features;

[0012] Step 2.4: Combine the attribute features, data range features, and mathematical features to form a feature library;

[0013] Step 3: Use the highest correlation search method to select the highest correlation features from the feature library to obtain model input features;

[0014] Step 4: Feed the model input features into a dual-channel interpretability network to obtain the SOH estimation result.

[0015] Furthermore: The lithium battery dataset is collected from four aging tests, and all four aging tests adopt constant current-constant voltage (CC-CV) and constant current (CC) discharge methods; in the constant current-constant voltage mode, the battery is charged to the upper cut-off voltage using a specified constant current, and then charged using the upper cut-off voltage until the current drops below 0.05C; in the constant current discharge mode, the battery discharges at a specified constant current to a lower cut-off voltage, and the cycle life test will stop until the capacity drops below 80% of the initial capacity; the working condition settings of the lithium-ion battery in the aging test conditions are shown in Table 1:

[0016]

[0017] Among them, L1~L16 are the names of different lithium battery packs, and the lithium battery is a commercial pouch lithium-ion battery with a capacity of 12Ah.

[0018] Furthermore: In Step 2.1, the attribute features include voltage, current, capacity, energy, IC curve, capacity change, energy change, average voltage, ohmic resistance, and polarization resistance, where the ohmic resistance , the polarization resistance ; represents the current change amount, represents the voltage difference within 1s, represents the voltage change from 1s to 20s;

[0019] In step 2.2, according to the charging mode, the attribute features are divided into three parts for feature extraction: the CC stage, the CV stage, and the CC-CV stage. Each charging segment serves as a battery charging data curve;

[0020] In step 2.3, each battery charging data curve is converted into 8 statistical values, namely: quantile, maximum value, minimum value, average value, variance, kurtosis, skewness, and slope, to capture the shape and position evolution of each cycle curve; the quantile reflects the evolution by sampling the values at each quantile position; the average value measures the average level of each curve within a fixed range; the variance is used to evaluate the non-uniformity of the data stream distribution, and the larger the variance, the more non-uniform the distribution; skewness and kurtosis characterize the shape of each curve; the slope represents the transverse polarization of the battery;

[0021] In step 2.4, according to the charging mode, each feature attribute sequence among the 10 attribute features is divided into 3 charging segments, and each charging segment is converted into 8 statistical values. Through the combined statistical method, for the battery under study, a comprehensive feature library with 240 mechanical statistical fusion features is generated.

[0022] Furthermore: Step 3 is specifically the following steps:

[0023] Step 3.1: Evaluate the non-linear correlation between the features in the feature library and the SOH, and select the features with a correlation greater than the first set threshold;

[0024] Step 3.2: Perform a decentralization operation on the selected features and make the mean of the features 0;

[0025] Step 3.3: Calculate the covariance matrix of the decentralized results;

[0026] Step 3.4: Perform eigenvalue decomposition and singular value decomposition on the covariance matrix to obtain eigenvalues and eigenvectors to determine the variance contribution rate and cumulative variance contribution rate;

[0027] Step 3.5: Select the features with a cumulative variance contribution rate greater than the second set threshold, and calculate the split gain of these features through the maximum gradient boosting method;

[0028] Step 3.6: Average the split gains of the features to obtain the importance score of each feature;

[0029] Step 3.7: Select several features with the largest importance scores as the model input features.

[0030] Furthermore: The dual-channel interpretability network includes a global channel, a local channel, an interactive attention mechanism, and a fully connected layer. Step 4 includes the following steps:

[0031] Step 4.1: Perform min-max normalization on the model input features and then send them into the dual-channel interpretability network;

[0032] Step 4.2: The outputs of min-max normalization are respectively sent into the interactive attention mechanism after being processed by the global channel and the local channel;

[0033] Step 4.3: The interactive attention mechanism uses the output of the global channel as the input for its query operation to locate the key time series segments related to the overall health state in the local channel guided by the global degradation trend; and uses the output of the local channel as the input for its key and value to map the local dynamic response into the global degradation framework; generates an attention weight matrix through the Softmax function to quantify the importance of global and local features;

[0034] Step 4.4: After the output of the interactive attention mechanism passes through the fully connected layer, the SOH estimation result is obtained.

[0035] Furthermore: The formula of the interactive attention mechanism is as follows:

[0036] (15)

[0037] Where, 、 and respectively represent the linear transformations of the query, key, and value, represents the transpose of the key, represents the dimension of the key. represents the Softmax activation function.

[0038] Furthermore: The global channel includes a hidden layer, a min-max normalization layer, a B-spline activation function, and a regularization layer. Among them, the input feature is 4-dimensional data, and the hidden layer has 9 nodes; perform min-max normalization processing on the input features of the global channel and scale the input features to a fixed range [-1, 1] proportionally; use the B-spline function as the activation function, and then optimize the model structure through sparse regularization and pruning to obtain the output of the global channel.

[0039] Furthermore: The local channel includes a sliding window, a multi-scale convolutional layer, a ReLU activation function, and a global pooling layer. The sliding window divides the input features into subsequences of a fixed length to capture local time series patterns; uses the multi-scale convolutional layer to extract local patterns of different time scales in parallel and calculate the weighted sum of the local regions; uses the ReLU activation function to set the negative values of the input features to zero and retain the positive values to focus on the positive value region and filter out the local features strongly related to battery aging; compresses the content extracted by the multi-scale convolution into the mean through the global pooling operation to retain the global time series pattern.

[0040] Advantages of the present invention: By analyzing and processing the lithium battery data sets under four different working conditions, it is possible to solve the problem of effective feature input of lithium batteries under different types, working conditions and materials, obtain a rich feature library by analyzing the battery data sets from multiple angles, and then realize automatic transparency for feature selection through the highest correlation search method, effectively reducing redundant data input, extracting highly correlated features, solving the problems of recent trends and historical dependence relationships in lithium battery data, and providing new insights for the field of lithium-ion battery SOH estimation tasks. Description of the Drawings

[0041] Figure 1 is a flowchart of the present invention;

[0042] Figure 2 is a data processing process diagram of the present invention;

[0043] Figure 3 is a dual-channel interpretability network in the present invention. Detailed Embodiment

[0044] The following will describe the present invention in detail with reference to the drawings. The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention. The orientation terms such as left, middle, right, up, and down in the embodiments of the present invention are only relative concepts to each other or are referenced based on the normal use state of the product, and should not be considered as restrictive.

[0045] A method for estimating the SOH of a lithium battery for an intelligent Yellow River underwater vehicle, in combination with Figure 1 and Figure 2 as shown, includes the following steps:

[0046] Step 1: Establish a lithium-ion battery cycle life test platform to obtain lithium battery data sets under four different working conditions;

[0047] The lithium battery data sets are collected from four aging tests, and all four aging tests use constant current constant voltage (CC-CV) and constant current (CC) discharge methods; in the constant current constant voltage mode, the battery is charged to the upper cut-off voltage using a specified constant current, and then charged using the upper cut-off voltage until the current drops below 0.05C; in the constant current discharge mode, the battery discharges at a specified constant current to a lower cut-off voltage, and the cycle life test will stop until the capacity drops below 80% of the initial capacity; the working condition settings of the lithium-ion battery in the aging test conditions are shown in Table 1:

[0048]

[0049] Among them, L1 to L16 are the names of different lithium - ion battery packs. The lithium - ion battery is a commercial pouch - type lithium - ion battery with a capacity of 12 Ah. From the perspective of degradation paths under different working conditions, the attenuation of the battery under the same working condition generally remains unchanged. The SOH of the battery under working condition 3 shows a linear decline. The characteristics of other working conditions are two - stage degradation, with slow linear attenuation in the first stage and rapid attenuation in the second stage, indicating overall non - linear behavior. There are differences between the usage rate and the aging rate, especially for the four batteries in working condition 1, which decay the fastest. Since the aging rate and the length of the cycle life vary greatly between working states, accurate SOH estimation under different working conditions is obtained; in this work, the SOH of the battery is defined by the capacity, that is, the ratio of the current capacity of aging to the initial capacity, as shown in formula (1):

[0050] (1)

[0051] Where represents the current capacity, represents the initial capacity.

[0052] The following table shows the basic parameters of a 12 - Ah commercial pouch - type lithium - ion battery;

[0053]

[0054] Step 2: Analyze and process the lithium - ion battery dataset to construct a feature library;

[0055] Feature construction is the most crucial part of feature engineering. Traditional feature extraction methods are often designed and optimized for specific applications, so their generalization ability is relatively weak. When the application scenario or the dataset changes, traditional methods need to be further optimized. Good feature construction can enhance the generalization ability of the model for unknown data, enabling the model to maintain good estimation performance under different working conditions and conditions. The present invention creates a method that can, once constructed, comprehensively and exhaustively describe the battery aging behavior from the lithium - ion battery dataset without change; specifically, the feature library constructed by the feature construction method of the present invention is generated by a three - step construction strategy: attribute features, data - range features, and mathematical features. Attribute features are the first step in constructing the feature library, which determines what type of data to extract. Data - range features detail the data ranges of the battery attribute features. As the last step in construction, mathematical features slice the lithium - ion battery dataset from a mathematical perspective into features.

[0056] Step 2.1: Analyze the lithium - ion battery dataset from the perspective of battery attributes and construct attribute features;

[0057] From the perspective of battery attributes, the voltage, current, and time information in the battery aging experiment are monitored, and three types of attribute features are introduced from these three aspects: directly measured information, implicit electrochemical information, and physical property information of the battery;

[0058] The directly measured information consists of four directly monitored data: voltage (V), current (I), capacity (Q), and energy (E). All parameters capture the trend of battery aging at different charging stages. As the polarization effect in thermodynamics decreases, the time in the constant voltage stage increases, resulting in a gradual decrease in the slope of the current curve and the curve expanding to the right during aging. Similarly, the time to reach the upper cut-off voltage also decreases, and the voltage curve in the constant current stage shows an increasing slope and contracting to the left. The above changes are subsequently transformed into eigenvalue in combination with the mathematical perspective to reveal the thermodynamic and kinetic changes in battery life. The temperature rises and then falls in one cycle. Therefore, if the battery temperature is detected during aging, the battery temperature can be used as a directly measured parameter.

[0059] The implicit electrochemical information includes three parameters, namely IC, ΔQ, and ΔE, which are introduced into the attribute features as sufficient data for aging diagnosis. Analyze the voltage-capacity curve at 1C current, where the voltage shows a monotonic increase in the voltage range of 15% (3.5V - 3.75V) and can be charged up to 50%. By processing the voltage-capacity curve into a capacity increment (IC) curve, the aging of the battery can be understood more deeply. IC converts the voltage plateau on the voltage curve into recognizable peaks, which represent the key factors affecting battery aging. The calculation formula of the IC curve is as follows:

[0060] (2)

[0061] where represents the capacity change when the voltage changes from to , represents the number of sampling times, represents the sampling time, is the current, represents the differential.

[0062] The curve is the difference between the capacity of the old battery and the new battery. Analyzing the curve can visually obtain that the capacity dissipation is near 3.5V, 3.65V, and 3.78V. The voltage position of the capacity dissipation platform coincides with the voltage position of the reaction peak.

[0063] The curve reflects the capacity loss in the full voltage range between cycles, while the IC curve finds that the capacity loss mainly occurs during the electrochemical phase change process. The above phenomenon is shown in formula (3):

[0064] (3)

[0065] The curve is also the total change of the full-cycle IC curve, where and are the curves of the new battery and the old battery respectively. The electrochemical meaning of the ΔE(V) curve is similar to that of .

[0066] The physical performance parameters use three variables, namely the average voltage, ohmic resistance, and polarization resistance, as the final part of the aging parameters. The average voltage is determined by the Gibbs free energy exchange reaction of lithium ions and can characterize the thermodynamic decay of the battery. The intensity of the exchange reaction changes during the aging process and is also reflected in the average voltage. The average voltage formula is shown in (4):

[0067] (4)

[0068] where and represent the end capacity and the initial capacity respectively.

[0069] The ohmic resistance and polarization resistance can describe the kinetic performance of the battery. The ohmic resistance is usually considered as the ratio of the instantaneous voltage change to the corresponding current change. In this paper, the voltage difference within 1 s after applying a current of ΔV1s is used for calculation, as shown in Equation (5):

[0070] (5)

[0071] where represents the current change amount.

[0072] Due to the fact that the transfer of lithium ions is less than that of electrons, there is a polarization effect, as shown in Equation (6):

[0073] (6)

[0074] The ratio of the voltage change to the current within 1 s - 20 s is used as the polarization resistance.

[0075] It can be seen from this that the attribute characteristics include voltage (V), current (I), capacity (Q), energy (E), IC curve, capacity change (∆Q), energy change (∆E), average voltage, ohmic internal resistance, and polarization resistance;

[0076] Step 2.2: Analyze the attribute characteristics from the perspective of the data range and construct the data range characteristics;

[0077] The discharge mode depends on the load condition, but the charging mode is relatively regular, generally being a multi-stage constant current or constant current constant voltage mode. Therefore, the data range in this article is divided according to charging. The evolution of measurement parameters such as current, voltage, capacity, and energy at different stages reflects the increase in polarization and thermodynamic losses during battery aging. Therefore, the battery reaches the cut-off voltage earlier, while the battery reaches the cut-off charging current later. Since battery aging occurs at any charging moment, the entire operating data can be divided into multiple segments according to the state to capture the changes in different operating stages. For example, assume that the battery is charged in multiple stages with a constant current. In this case, battery degradation can be tracked through the data evolution of each stage, and each charging segment can be used as a data range. The attribute features are divided into three parts for feature extraction according to the charging mode: the CC stage, the CV stage, and the CC-CV stage, and each charging segment serves as a battery charging data curve;

[0078] Step 2.3: Analyze the data range features from the perspective of mathematical statistics and construct mathematical features; convert each battery charging data curve into 8 statistical values, namely: quantile, maximum value, minimum value, average value, variance, kurtosis, skewness, and slope, to capture the shape and position evolution of each cycle curve; the quantile reflects the evolution by sampling the values at each quantile position; the average value measures the average level of each curve within a fixed range; the variance is used to evaluate the non-uniformity of the data stream distribution, and the larger the variance, the more non-uniform the distribution; skewness and kurtosis characterize the shape of each curve; for the battery voltage curve, due to thermodynamic losses and increasing polarization, a larger skewness indicates a larger slope at the high end of the curve and a smaller percentage of the plateau region; the slope represents the transverse polarization of the battery;

[0079] Step 2.4: Compose the attribute features, data range features, and mathematical features into a feature library; divide each feature attribute sequence of the 10 attribute features into 3 charging segments according to the charging mode, and convert each charging segment into 8 statistical values. Through a combined statistical method, for the battery under study, a comprehensive feature library with 240 mechanical statistical fusion features is generated;

[0080] Step 3: Use the highest correlation search method to select the highest correlation features from the feature library to obtain the model input features;

[0081] The purpose of feature selection is to be able to fully automate the selection of features applicable to any battery. Traditional feature selection methods cannot extract the most representative features. Moreover, too many features may lead to network overfitting and increase the network calculation burden. Therefore, the present invention designs a new feature selection engineering based on the highest correlation search method, selects the most representative 4 feature data from the charge and discharge data of lithium-ion batteries, and realizes the feature engineering innovation of automatic and transparent feature selection.

[0082] Specifically, it includes the following steps:

[0083] Step 3.1: Evaluate the non - linear correlation between the features in the feature library and the SOH (State of Health of the battery), and select the features with a correlation greater than the first set threshold; specifically: The present invention first evaluates the non - linear correlation between the extracted features and the SOH. The non - linear correlation screening results are normalized to be in the range of [0, 1], where 1 represents complete linear correlation and 0 represents no association between the feature and the SOH. The calculation of non - linear correlation is as shown in formula (7):

[0084] (7)

[0085] Where X represents the feature variable, Y represents the target variable (i.e., SOH), p(x, y) is the joint probability distribution, and p(x) and p(y) are the marginal probability distributions. For the convenience of comparison and analysis, the present invention standardizes the features in each feature library so that the results are on the same scale. The present invention selects 24 features with a correlation greater than 0.9, which means these features have a high correlation. In this way, features with low correlation will be discarded, reducing the data input and retaining the high - correlation features that affect the aging factors of lithium - ion batteries.

[0086] Step 3.2: Perform a de - centering operation on the selected features and make the mean of the features equal to 0;

[0087] After the non - linear correlation screening, the high - correlation features are still high - dimensional data. High - dimensional data has the curse of dimensionality and redundancy will also occur. Redundant data will increase the network calculation burden and slow down the calculation speed. To reduce the information coupling between these features, a dimensionality reduction method is then used for dimensionality reduction. The present invention designs a novel contribution - based dimensionality reduction method to identify important features in battery data. This method converts high - dimensional data into low - dimensional data, solves the problem of multicollinearity, and at the same time retains important information to better capture the essential features of battery aging data. Through orthogonal transformation of the high - correlation features after the correlation analysis of lithium - ion batteries, a new set of variables is generated. Subsequently, features with high contribution rates are retained to encapsulate key information, and features with low contribution rates are discarded to achieve data dimensionality reduction and reduce redundancy. The operation process is as follows:

[0088] First, perform a de - centering operation to make the data mean equal to 0, as shown in the formula:

[0089] (8)

[0090] Where is the number of samples with a correlation greater than 0.85, and the value is 14. , and each in it is a feature with a correlation greater than 0.9.

[0091] Step 3.3: Calculate the covariance matrix of the decentralized result; the covariance matrix is:

[0092] (9)

[0093] where is the transpose of, and the operation result after the covariance matrix is .

[0094] Step 3.4: Perform eigenvalue decomposition and singular value decomposition on the covariance matrix to obtain eigenvalues and eigenvectors to determine the variance contribution rate and cumulative variance contribution rate;

[0095] (10)

[0096] (11)

[0097] where represents the variance contribution of the eigenvector , represents the number of healthy features, represents the number of eigenvectors, represents the eigenvalue of the eigenvector , represents the cumulative variance contribution. In this work, the contribution threshold is selected as 1 / 24, which is the average contribution of each feature under ideal conditions. Features below the threshold are defined as invalid features. The core of this process is to ensure that the information contained in the principal components is 90% of the original time series space by setting the cumulative variance rate of 10% - 90% in Step 10. At this time, the features can effectively capture the main features of the data, and the remaining 12 features are selected after the selection. This step eliminates the redundant feature information and reduces the calculation burden of the subsequent SOH estimation model.

[0098] Step 3.5: Select the features with a cumulative variance contribution rate greater than the second set threshold, and calculate the split gain of these features by the maximum gradient boosting method;

[0099] Although highly relevant features can be selected after correlation analysis and contribution degree dimensionality reduction, it is not possible to determine which features are the most influential for battery SOH estimation. The present invention designs a maximum gradient boosting method to select the most relevant features. The maximum gradient boosting method adds a regularization term to the loss function in the traditional gradient boosting decision tree. And since it is difficult to calculate the derivative of some loss functions, the maximum gradient boosting method of the present invention uses the second-order Taylor expansion of the loss function as the fitting of the loss function. The feature set can be further optimized to remove redundant information, making the network training speed more efficient and enabling the SOH estimation model to perform excellently in preventing overfitting and improving generalization ability. The maximum gradient boosting method is an ensemble learning algorithm based on gradient boosting, and the importance of features is measured by the splitting gain of decision trees. The importance score of a feature can represent the frequency of the number of splits of the feature in the tree structure or the gain brought. The more times a feature splits in the tree structure or the greater the gain brought by the split, the higher its importance score in SOC estimation.

[0100] During the training process of the maximum gradient boosting method, the algorithm automatically calculates the importance score of each feature. Specifically, the feature importance of the maximum gradient boosting method is calculated based on the information gain in each tree:

[0101] (12)

[0102] where is feature k, denotes feature the splitting gain of in decision tree j;

[0103] Step 3.6: By averaging the splitting gains of all features, the importance score of each feature is obtained;

[0104] Step 3.7: The higher the score, the higher the representativeness of the feature in SOH estimation. The present invention selects the 4 features with the largest scores as the model input features;

[0105] Step 4: Feed the model input features into the dual-channel interpretability network. As shown in combination with Figure 3 , the dual-channel interpretability network includes a global channel (KAN), a local channel (1D-CNN), an interactive attention mechanism, and a fully connected layer, and specifically includes the following steps:

[0106] Step 4.1: Perform maximum-minimum normalization on the model input features to eliminate the dimensional difference, and then feed them into the dual-channel interpretability network;

[0107] Step 4.2: The outputs of the min-max normalization are respectively sent into the interactive attention mechanism after being processed by the global channel and the local channel; the global channel includes a hidden layer, a min-max normalization layer, a B-spline activation function, and a regularization layer. Among them, the input features of the model are 4-dimensional data, and the hidden layer has 9 nodes, effectively reducing the dimension of the model input features while retaining important information structures; the input features of the global channel are subjected to min-max normalization processing, and the input features are scaled to a fixed range [-1, 1] to improve the computational stability of the model; the B-spline function is used as the activation function, and then the model structure is optimized by sparse regularization and pruning to reduce input redundancy and obtain the output of the global channel. The local channel includes a sliding window, a multi-scale convolutional layer, a ReLU activation function, and a global pooling layer. The sliding window divides the input features into subsequences of a fixed length, the window length is set to 32, and the step size is 10, which is used to capture local temporal patterns; the multi-scale convolutional layer is used to parallelly extract local patterns of different time scales and calculate the weighted sum of the local regions; the ReLU activation function is used to set the negative values of the input features to zero and retain the positive values to focus on the positive value region, screening out local features strongly related to battery aging and introducing non-linear expression ability into the model; the content extracted by the multi-scale convolution is compressed into the mean value through the global pooling operation to retain the global temporal pattern;

[0108] Step 4.3: The interactive attention mechanism uses the output of the global channel as the input of its own query operation to locate the key temporal segments related to the overall health state in the local channel guided by the global degradation trend; and uses the output of the local channel as the input of its own key and value to map the local dynamic response to the global degradation framework, enhancing the sensitivity of the model to abnormal events; generates an attention weight matrix through the Softmax function to quantify the importance of global and local features. Among them, the formula of the interactive attention mechanism is as follows:

[0109] (15)

[0110] Where 、 and respectively represent the linear transformations of the query, key, and value, represents the transpose of the key, represents the dimension of the key. represents the Softmax activation function;

[0111] Step 4.4: After the output of the interactive attention mechanism passes through the fully connected layer, the SOH estimation result is obtained.

[0112] In view of the fact that traditional weight parameters are replaced by univariate function parameters at the edge of the network. The present invention proposes a new type of SOH estimation network: a dual-channel interpretability network (KAN-1DCNN). The design idea of the dual-channel structure is to divide the network into a local channel and a global channel. After the model input features pass through the dual-channel structure, they pass through a fully connected layer to automatically learn and fuse the features from the two channels to obtain the estimated SOH estimation result. The local channel uses one-dimensional convolution to extract local features, focusing on capturing local details. The global channel uses an interpretability layer to process global context information. The dual-channel interpretability network can help the model maintain sensitivity to local features and the ability to grasp global information at the same time. The dual-channel interpretability network does not rely on a fixed activation function. Instead, it uses a learnable univariate function, thus enhancing its ability to adapt to complex data patterns. In the dual-channel interpretability network, each node aggregates the outputs of these functions without any non-linear transformation. This feature enables the SOH estimation model to improve the prediction accuracy and computational efficiency while maintaining a high level of model interpretability.

[0113] The present invention uses four evaluation metrics to comprehensively evaluate the performance of the proposed KAN-1DCNN. These metrics help objectively measure the prediction ability of the network and reveal its accuracy and robustness from different perspectives. The definitions of the four evaluation metrics are as follows:

[0114] Root Mean Square Error (RMSE): The root mean square error is the square root of the mean square error, which is used to measure the average difference between the network prediction value and the true value, that is, the square root of the prediction error. RMSE is more sensitive to outliers and can be used to measure the precision of the network. Mean Absolute Error (MAE): The mean absolute error is the average of the absolute errors calculated between the prediction value and the true value, measuring the average error of the prediction. Different from MSE, MAE does not amplify the influence of large errors, so it can better reflect the overall accuracy of the prediction. Mean Absolute Percentage Error (MAPE): The mean absolute percentage error is the average of the relative differences between the prediction value and the true value, expressed as a percentage. It measures the relative error of the network within different data ranges and can reflect the relative accuracy of the prediction. Maximum Absolute Error (MAXE): The maximum absolute error is the maximum of the absolute differences between the prediction value and the true value, which identifies the prediction error of the network in the worst case. MAXE is particularly sensitive to outliers and helps to understand the maximum risk in the network prediction.

[0115] The present invention uses four different working condition datasets obtained from two test platforms, and uses L1, L2, L3, L5, L6, L7, L9, L10, L11, L13, L14 and L15 as the training set, and uses L4, L8, L12 and L16 as the test set. To train the network, the present invention selects the mean absolute error function as the loss function of the network. The hyperparameters in this experiment are determined as follows: batchsize = 32, epoch = 160, learning rate = 0.00225. The present invention uses the gradient-based AdamW optimization algorithm. AdamW is a variant of the Adam optimization algorithm, which improves the original Adam algorithm by adding a weight decay mechanism, making the optimization algorithm more flexible and helping to better select hyperparameters.

[0116] Through comparative experiments, the accuracy of the proposed KAN-1DCNN is verified, and comparative experiments are carried out with multiple existing networks (TCN, LSTM and BiGRU). In this series of comparative experiments, the present invention maintains unified experimental parameter settings and the same environmental conditions to ensure the comparability of experimental results. The experimental results are shown in Table 3.

[0117] Table 3 Comparative experimental results

[0118]

[0119] Observing the experimental results, it can be seen that KAN-1DCNN shows significant advantages under all evaluation indexes. Taking the L4 dataset as an example, compared with TCN and BiGRU, KAN-1DCNN improves by 82%, 77%, 81% and 63% respectively in the four evaluation indexes of RMSE, MAE, MAPE and MAXE. For LSTM, KAN-1DCNN improves by 72%, 81%, 78% and 62% respectively in the four evaluation indexes of RMSE, MAE, MAPE and MAXE. For BiGRU, KAN-1DCNN improves by 79%, 80%, 82% and 52% respectively in the four evaluation indexes of RMSE, MAE, MAPE and MAXE.

[0120] The KAN-1DCNN network performs excellently in terms of prediction accuracy, providing a reliable basis for the assessment and management of the battery health state. The comprehensive comparative test results show that the KAN-1DCNN network is accurate and stable on the test set. Compared with existing networks such as TCN, LSTM and BiGRU, KAN-1DCNN shows significant advantages in multiple evaluation indexes. This strengthens the excellent performance of KAN-1DCNN in the estimation of the lithium-ion battery health state, providing strong support and guidance for the optimization of battery management and maintenance strategies.

[0121] In the robustness experiment, the present invention evaluated the impact of noise on the performance of the KAN-1DCNN network, aiming to explore its performance under different noise levels. During the experiment, the present invention introduced noises with different amplitudes (50 mV, 100 mV, and 150 mV) to simulate the uncertainty of real battery data. Through the analysis of the experimental results, a deep understanding of the stability and robustness of the KAN-1DCNN network when facing noise can be obtained. To ensure the reliability of the experiment, the above parameter settings were adopted, and the experiment was carried out under the same environmental conditions. The results are shown in Table 4.

[0122] Table 4 Results of the Robustness Experiment

[0123]

[0124] It is clearly visible from the result table that as the noise level increases, the performance of each network decreases. However, KAN-1DCNN still maintains a high prediction accuracy at each noise level. Taking the test set L8 as an example, as the noise level increases from 50 mV to 150 mV, the MAE of KAN-1DCNN only increases from 0.55% to 1.64%. This indicates that KAN-1DCNN has strong robustness to noise and can resist the influence of data uncertainty to a certain extent. In addition, the team noticed that KAN-1DCNN performed excellently on the L4 dataset. Although the performance of all networks decreased at high noise levels, KAN-1DCNN still maintained relatively low evaluation index values, confirming its stability under different data qualities. The results of the robustness experiment further verified the excellence of KAN-1DCNN when dealing with noise. Compared with other networks, KAN-1DCNN showed stronger stability and robustness, providing strong support for its reliability in actual battery data scenarios. This is of great significance for the practical application of battery data management, especially in cases involving the uncertainty of external factors.

[0125] The above has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A lithium battery SOH estimation method for a smart Yellow River underwater vehicle, characterized by: The following steps are involved: Step 1: Establish a lithium-ion battery cycle life test platform and obtain lithium battery data sets under four different working conditions; Step 2: Analyze and process the lithium battery data set to build a feature library, specifically: Step 2.1: Analyze the lithium battery dataset from the perspective of battery attributes and construct attribute features; Step 2.2: Analyze attribute characteristics from the perspective of data range and construct data range characteristics; Step 2.3: Analyze the data range characteristics from the perspective of mathematical statistics and construct mathematical features; Step 2.4: compose a feature library with attribute features, data range features, and mathematical features; Step 3: Use the highest correlation search method to select the feature with the highest correlation with SOH from the feature library to obtain the model input feature; Step 4: Send the model input features into the dual-channel interpretable network to obtain the SOH estimation result; wherein the dual-channel interpretable network includes a global channel, a local channel, an interactive attention mechanism and a fully connected layer. Step 4 specifically includes the following steps: Step 4.1: Perform maximum and minimum normalization on the model input features and then send them to the dual-channel interpretability network; Step 4.2: The output of the maximum and minimum normalization is processed by the global channel and the local channel respectively and then sent to the interactive attention mechanism; the global channel includes a hidden layer, a maximum and minimum normalization layer, a B-spline activation function and a regularization layer; the B-spline function is used as the activation function, and the model structure is optimized by sparse regularization and pruning to obtain the output of the global channel; the local channel includes a sliding window, a multi-scale convolution layer, a ReLU activation function and a global pooling layer. The sliding window divides the input features into subsequences of fixed length to capture local temporal patterns; a multi-scale convolution layer is used to extract local patterns of different time scales in parallel and calculate the weighted sum of the local area; the ReLU activation function is used to set the negative values ​​of the input features to zero and retain the positive values ​​to focus on the positive area and screen out local features that are strongly related to battery aging; the content extracted by the multi-scale convolution is compressed to the mean through the global pooling operation to retain the global temporal pattern; Step 4.3: The interactive attention mechanism uses the output of the global channel as the input of its own query operation to locate the key time sequence segments related to the overall health status in the local channel guided by the global degradation trend; and uses the output of the local channel as the input of its own key and value to map the local dynamic response to the global degradation framework; generates the attention weight matrix through the Softmax function to quantify the importance of global and local features; Step 4.4: The output of the interactive attention mechanism passes through the fully connected layer to obtain the SOH estimation result.

2. The lithium battery SOH estimation method for the smart Yellow River underwater vehicle according to claim 1 is characterized by: The lithium battery data set is collected from four aging tests, all of which adopt constant current constant voltage (CC-CV) and constant current (CC) discharge methods; In constant current and constant voltage mode, the battery is charged to the upper cut-off voltage using a specified constant current, and then charged using the upper cut-off voltage until the current drops below 0.05C; in constant current discharge mode, the battery is discharged to a lower cut-off voltage at a specified constant current, and the cycle life test will stop until the capacity drops below 80% of the initial capacity; The aging test includes the operating conditions of the lithium-ion battery.

3. The lithium battery SOH estimation method for the smart Yellow River underwater vehicle according to claim 1 is characterized by: In step 2.1, the attribute characteristics include voltage, current, capacity, energy, IC curve, capacity change, energy change, average voltage, ohmic internal resistance and polarization resistance, among which ohmic resistance , polarization resistance ; Indicates the current change, Indicates the voltage difference within 1s, Indicates voltage changes within 1s-20s; In step 2.2, the attribute features are divided into three parts for feature extraction according to the charging mode: CC stage, CV stage and CC-CV stage, and each charging stage is taken as a battery charging data curve; In step 2.3, each battery charging data curve is converted into 8 statistical values, namely: quantile, maximum value, minimum value, mean value, variance, kurtosis, skewness and slope, to capture the shape and position evolution of each cycle curve; quantile reflects the evolution by sampling the value of each quantile position; mean value measures the average level of each curve within a fixed range; variance is used to evaluate the unevenness of data stream distribution, and the larger the variance, the more uneven the distribution; skewness and kurtosis characterize the shape of each curve; slope represents the transverse polarization of the battery; In step 2.4, each characteristic attribute sequence of the 10 attribute features is divided into 3 charging segments according to the charging mode, and each charging segment is converted into 8 statistical values. Through the combined statistical method, a comprehensive feature library with 240 mechanical statistical fusion features is generated for the battery under study.

4. The lithium battery SOH estimation method for the smart Yellow River underwater vehicle according to claim 1 is characterized by: Step 3 is specifically as follows: Step 3.1: Evaluate the nonlinear correlation between the features in the feature library and the SOH, and select the features whose correlation is greater than the first set threshold; Step 3.2: Decentralize the selected features and make the mean of the features 0; Step 3.3: Calculate the covariance matrix of the decentralized results; Step 3.4: Perform eigenvalue decomposition and singular value decomposition of the covariance matrix to obtain eigenvalues ​​and eigenvectors to determine the variance contribution rate and cumulative variance contribution rate; Step 3.5: Select features whose cumulative variance contribution rate is greater than the second set threshold, and calculate the split gains of these features by the maximum gradient boosting method; Step 3.6: Average the split gains of the features to get the importance score of each feature; Step 3.7: Select several features with the largest importance scores as model input features.

5. The lithium battery SOH estimation method for the smart Yellow River underwater vehicle according to claim 1 is characterized by: The formula of the interactive attention mechanism is as follows: （15） in, , and represent linear transformations of queries, keys, and values, respectively, represents the transposition of the key, represents the dimension of the key, Represents the Softmax activation function.

Images