Lithium battery health state online detection method and system based on frequency domain analysis

Abstract

The invention relates to the technical field of lithium battery health state detection, in particular to a lithium battery health state online detection method and system based on frequency domain analysis. The method comprises the steps of firstly obtaining lithium battery operation data, aging data, user behavior data and historical data; preprocessing the operation data to obtain lithium battery multi-scale operation frequency domain data, and inputting the aging data into a lithium battery aging frequency domain association model to obtain associated multi-scale operation frequency domain data; then, inputting the user behavior data and the multi-scale operation frequency domain data into a lithium battery multi-scale frequency domain characteristic influence prediction model, and predicting a lithium battery multi-scale frequency domain characteristic influence coefficient; and finally, inputting the operation frequency domain data, the associated multi-scale operation frequency domain data and the lithium battery multi-scale frequency domain characteristic influence coefficient into a lithium battery health state mixed frequency domain prediction model, outputting a lithium battery health state detection value, and performing threshold comparison. The accuracy and reliability of lithium battery health state detection can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery health status detection, and specifically to a method and system for online detection of lithium battery health status based on frequency domain analysis. Background Art

[0002] Lithium batteries age over time, which not only reduces performance indicators such as capacity and power but can also lead to safety risks such as internal short circuits and thermal runaway. Therefore, online monitoring of the health status of lithium batteries is crucial. Traditional battery time-domain detection methods are unable to extract effective features from complex voltage or current signals, thereby reducing the accuracy of lithium battery health status detection. Since signals are more easily distinguished in the frequency domain, using frequency domain analysis to detect battery health status can significantly improve this shortcoming.

[0003] At present, the existing technology still has shortcomings in lithium battery health status detection. On the one hand, the existing technology only uses lithium battery operation data in the frequency domain analysis process, and considers a single factor. On the other hand, the existing technology does not use the relationship between the battery aging stage and the frequency domain characteristics to assist in battery health status detection, which will reduce the accuracy and reliability of online detection of lithium battery health status.

[0004] To this end, a method and system for online detection of lithium battery health status based on frequency domain analysis is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for online detection of lithium battery health status based on frequency domain analysis, which is used for fault prediction and health management of lithium batteries. First, the lithium battery operation data, aging data, user behavior data and historical data are obtained; secondly, the operation data is preprocessed to obtain multi-scale operation frequency domain data of the lithium battery, and the aging data is input into the lithium battery aging frequency domain correlation model to obtain correlated multi-scale operation frequency domain data; then, the user behavior data and operation frequency domain data are input into the lithium battery multi-scale frequency domain characteristic influence prediction model to predict the lithium battery multi-scale frequency domain characteristic influence coefficient; finally, the operation frequency domain data, the correlated multi-scale operation frequency domain data and the lithium battery multi-scale frequency domain characteristic influence coefficient are input into the lithium battery health status hybrid frequency domain prediction model, and the lithium battery health status detection value is output. The value is compared with the set lithium battery health status threshold to realize online detection of the lithium battery health status; the present invention can improve the accuracy and reliability of online detection of lithium battery health status.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for online detection of lithium battery health status based on frequency domain analysis, comprising:

[0008] Obtain lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data;

[0009] Preprocess the lithium battery operation data to obtain multi-scale operation frequency domain data of the lithium battery;

[0010] Construct a lithium battery aging frequency domain correlation model, train it using the historical lithium battery data, and input the lithium battery aging data into the lithium battery aging frequency domain correlation model to obtain correlated multi-scale operation frequency domain data;

[0011] Input the user behavior data and the multi-scale operation frequency domain data of the lithium battery into a pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain a lithium battery multi-scale frequency domain characteristic influence coefficient;

[0012] Input the multi-scale operation frequency domain data of the lithium battery, the correlated multi-scale operation frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient into a lithium battery health state hybrid frequency domain prediction model to obtain a lithium battery health state detection value;

[0013] Online detect the health state of the lithium battery according to the lithium battery health state detection value and the lithium battery health state threshold.

[0014] Further, the lithium battery operation data includes: remaining battery power, battery capacity, current, voltage, and charge-discharge cycles; the user behavior data includes: charge-discharge rate, charge-discharge depth, operating temperature, and idle time; the historical lithium battery data includes: historical lithium battery operation data, historical lithium battery aging data, historical user behavior data, and historical lithium battery health state data.

[0015] Further, the process of preprocessing the lithium battery operation data to obtain multi-scale operation frequency domain data of the lithium battery includes:

[0016] Clean the lithium battery operation data to obtain first lithium battery operation data;

[0017] Segment the first lithium battery operation data to obtain second lithium battery operation data;

[0018] Extract multi-scale frequency domain features from the second lithium battery operation data to obtain third lithium battery operation data;

[0019] Proofread and normalize the third lithium battery operation data to obtain the multi-scale operation frequency domain data of the lithium battery.

[0020] Further, a frequency-domain correlation model for lithium battery aging is constructed and trained using the historical lithium battery data. The process of inputting the lithium battery aging data into the frequency-domain correlation model for lithium battery aging to obtain the correlated multi-scale operating frequency-domain data includes:

[0021] The historical lithium battery operating data and historical lithium battery aging data in the historical lithium battery data are used to train the frequency-domain correlation model for lithium battery aging, obtaining a pre-trained frequency-domain correlation model for lithium battery aging;

[0022] The lithium battery aging data is input into the pre-trained frequency-domain correlation model for lithium battery aging to obtain the importance scores of different frequency-domain features, which are used to analyze the correlation degree of different frequency-domain features with the current aging features;

[0023] The correlated multi-scale operating frequency-domain data sensitive to the lithium battery aging data is extracted according to the importance scores.

[0024] Further, the process of inputting the user behavior data and the lithium battery multi-scale operating frequency-domain data into the pre-trained prediction model for the influence of lithium battery multi-scale frequency-domain characteristics to obtain the influence coefficient of lithium battery multi-scale frequency-domain characteristics includes:

[0025] The historical user behavior data and historical lithium battery operating data in the historical lithium battery data are used to train the prediction model for the influence of lithium battery multi-scale frequency-domain characteristics, obtaining the weights of the historical influence prediction model;

[0026] The user behavior data and the lithium battery multi-scale operating frequency-domain data are input into the pre-trained prediction model for the influence of lithium battery multi-scale frequency-domain characteristics for processing, and the weights of the historical influence prediction model are updated to generate the influence coefficient of lithium battery multi-scale frequency-domain characteristics.

[0027] Further, the process of inputting the lithium battery multi-scale operating frequency-domain data, the correlated multi-scale operating frequency-domain data, and the influence coefficient of lithium battery multi-scale frequency-domain characteristics into the hybrid frequency-domain prediction model for lithium battery health state to obtain the lithium battery health state detection value includes:

[0028] The input layer of the hybrid frequency-domain prediction model for lithium battery health state is used to receive the lithium battery multi-scale operating frequency-domain data, the correlated multi-scale operating frequency-domain data, and the influence coefficient of lithium battery multi-scale frequency-domain characteristics;

[0029] The lithium battery multi-scale operating frequency-domain data and the correlated multi-scale operating frequency-domain data are input into the multi-scale frequency-domain feature extraction layer of the hybrid frequency-domain prediction model for lithium battery health state to obtain the lithium battery multi-scale operating frequency-domain feature representation and the correlated multi-scale operating frequency-domain feature representation;

[0030] Input the multi-scale operating frequency domain features representation of the lithium battery and the associated multi-scale operating frequency domain features representation into the cross-attention layer of the lithium battery health state hybrid frequency domain prediction model to obtain a weighted frequency domain features representation;

[0031] Input the weighted frequency domain features representation and the lithium battery multi-scale frequency domain characteristic influence coefficient into the adaptive adjustment layer of the lithium battery health state hybrid frequency domain prediction model to obtain an adaptive frequency domain adjustment features representation;

[0032] Input the adaptive frequency domain adjustment features representation into the health state prediction layer and the output layer of the lithium battery health state hybrid frequency domain prediction model in sequence to obtain the lithium battery health state detection value.

[0033] An online detection system for the health state of a lithium battery based on frequency domain analysis, comprising: a data acquisition unit, a data preprocessing unit, a lithium battery aging frequency domain association unit, a lithium battery frequency domain characteristic influence prediction unit, a lithium battery health state frequency domain detection unit, and an output unit;

[0034] The data acquisition unit is used to acquire lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data;

[0035] The data preprocessing unit is used to preprocess the lithium battery operation data to obtain multi-scale operating frequency domain data of the lithium battery;

[0036] The lithium battery aging frequency domain association unit is used to input the lithium battery aging data into a lithium battery aging frequency domain association model to obtain associated multi-scale operating frequency domain data;

[0037] The lithium battery frequency domain characteristic influence prediction unit is used to input the user behavior data and the multi-scale operating frequency domain data of the lithium battery into a pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain a lithium battery multi-scale frequency domain characteristic influence coefficient;

[0038] The lithium battery health state frequency domain detection unit is used to input the multi-scale operating frequency domain data of the lithium battery, the associated multi-scale operating frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient into a lithium battery health state hybrid frequency domain prediction model to obtain a lithium battery health state detection value;

[0039] The output unit is used to output the online detection result of the lithium battery health state.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] 1. The present invention proposes a lithium battery aging frequency domain correlation model for obtaining correlated multi-scale operating frequency domain data that matches the current aging data. This model is first trained using historical lithium battery operating data and historical lithium battery aging data to learn the relationship between different aging degrees of lithium batteries and frequency domain characteristics. Then, the lithium battery aging data is input into the pre-trained lithium battery aging frequency domain correlation model to obtain the frequency points or frequency band information that is most sensitive to the current aging changes, which is beneficial to improving the frequency domain prediction accuracy of the subsequent battery health state, thereby enhancing the accuracy and reliability of the online detection of the lithium battery health state.

[0042] 2. The present invention proposes a method for predicting the influence of lithium battery frequency domain characteristics for obtaining the influence coefficients of lithium battery multi-scale frequency domain characteristics. This method uses a lithium battery multi-scale frequency domain characteristics influence prediction model to learn the degree of influence of user behavior changes on the lithium battery multi-scale operating frequency domain data. This method inputs the user behavior data and the lithium battery multi-scale operating frequency domain data into the trained lithium battery multi-scale frequency domain characteristics influence prediction model to obtain the influence coefficients of the lithium battery multi-scale frequency domain characteristics. Using this influence factor can effectively adjust the frequency domain characteristics under the operating state of the lithium battery to conform to the actual operating conditions, thereby enhancing the accuracy and reliability of the online detection of the lithium battery health state.

[0043] 3. The present invention proposes a lithium battery health state hybrid frequency domain prediction model for obtaining the lithium battery health state detection value based on multi-source frequency domain data. This model first converts the lithium battery multi-scale operating frequency domain data and the correlated multi-scale operating frequency domain data into the lithium battery multi-scale operating frequency domain feature representation and the correlated multi-scale operating frequency domain feature representation, and then combines the operating frequency domain feature representation and uses the cross-attention layer and frequency domain characteristic adjustment to perform weighting and adjustment respectively. The self-adaptive frequency domain adjustment feature representation is input into the health state prediction layer and the output layer to obtain the lithium battery health state detection value. This model can screen the frequency domain information beneficial to health state prediction through the weighting operation, and can make the operating frequency domain feature representation conform to the actual situation of users through the adjustment operation to ensure its authenticity, thereby enhancing the accuracy and reliability of the online detection of the lithium battery health state. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic flow chart of an online lithium battery health state detection method based on frequency domain analysis according to the present invention;

[0045] Figure 2 It is a schematic structural diagram of the lithium battery health state hybrid frequency domain prediction model of the present invention;

[0046] Figure 3 It is a schematic structural diagram of an online lithium battery health state detection system based on frequency domain analysis according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0048] Please refer to Figures 1 to 3 , the present invention provides an online detection method and system for the health state of lithium batteries based on frequency domain analysis, and the technical solutions are as follows:

[0049] Embodiment 1:

[0050] In order to improve the accuracy and reliability of the online detection of the health state of lithium batteries, a certain lithium battery manufacturer uses an online detection method for the health state of lithium batteries based on frequency domain analysis proposed by the present invention. The process schematic of this method is as shown in Figure 1 shown, and specifically includes:

[0051] Obtain lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data;

[0052] Furthermore, the lithium battery operation data includes: remaining battery power, battery capacity, current, voltage, and charge and discharge cycles; the user behavior data includes: charge and discharge rate, charge and discharge depth, working temperature, and idle time; the historical lithium battery data includes: historical lithium battery operation data, historical lithium battery aging data, historical user behavior data, and historical lithium battery health state data;

[0053] Furthermore, the lithium battery operation data is collected through corresponding sensors; the lithium battery aging data is obtained by inputting the lithium battery operation data into a pre-trained lithium battery aging model. This lithium battery aging model is a machine learning model incorporating knowledge of the battery aging mechanism and is trained using historical lithium battery data; the lithium battery aging data includes aging type, aging index, and aging degree; the user behavior data and historical lithium battery data are obtained by calling the data management background;

[0054] Furthermore, the aging types of lithium batteries include: growth of solid electrolyte interface film (SEI), lithium deposition, cracking, etc.

[0055] By introducing lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data, the present invention can provide a reliable data source for the training and processing requirements of subsequent lithium battery aging frequency domain correlation models, lithium battery multi-scale frequency domain characteristic influence prediction models, and lithium battery health state hybrid frequency domain prediction models, thereby effectively improving the accuracy and reliability of the online detection of the health state of lithium batteries.

[0056] Preprocess the operation data of the lithium battery to obtain the multi-scale operation frequency-domain data of the lithium battery;

[0057] Furthermore, the process of preprocessing the operation data of the lithium battery to obtain the multi-scale operation frequency-domain data of the lithium battery includes:

[0058] Clean the operation data of the lithium battery to obtain the first operation data of the lithium battery;

[0059] Segment the first operation data of the lithium battery to obtain the second operation data of the lithium battery;

[0060] Extract multi-scale frequency-domain features from the second operation data of the lithium battery to obtain the third operation data of the lithium battery;

[0061] Proofread and normalize the third operation data of the lithium battery to obtain the multi-scale operation frequency-domain data of the lithium battery;

[0062] Furthermore, data cleaning includes: data alignment, missing value processing, outlier processing, noise filtering, and data format conversion;

[0063] Furthermore, data segmentation can adopt fixed window, event-triggered, or adaptive window methods; the time length of the fixed window can be set according to actual needs; the event-triggered method can segment according to the start and end states of lithium battery charging / discharging; the adaptive window method dynamically adjusts the window size according to the local characteristics of the signal;

[0064] Furthermore, the multi-scale frequency-domain feature extraction process includes: using the fast Fourier transform to extract wide-band features, using the discrete wavelet transform to decompose the data signal, and extracting high-frequency band features and low-frequency band features; combining the low-frequency band features, wide-band features, and high-frequency band features to obtain the third operation data of the lithium battery.

[0065] By performing data cleaning, segmentation, multi-scale frequency-domain feature extraction, proofreading, and normalization on the operation data of the lithium battery, it is possible to improve the efficiency of subsequent multi-scale data analysis and processing while ensuring the integrity and accuracy of the data, which is beneficial to improving the accuracy and reliability of online detection of the health state of the lithium battery.

[0066] Construct a lithium battery aging frequency-domain correlation model, train it using historical lithium battery data, input the lithium battery aging data into the lithium battery aging frequency-domain correlation model, and obtain the correlated multi-scale operation frequency-domain data;

[0067] Furthermore, the process of constructing a lithium battery aging frequency-domain correlation model, training it using historical lithium battery data, inputting the lithium battery aging data into the lithium battery aging frequency-domain correlation model, and obtaining the correlated multi-scale operation frequency-domain data includes:

[0068] Training the lithium battery aging frequency domain correlation model using the historical lithium battery operation data and historical lithium battery aging data in the historical lithium battery data to obtain a pre-trained lithium battery aging frequency domain correlation model;

[0069] Inputting the lithium battery aging data into the pre-trained lithium battery aging frequency domain correlation model to obtain the importance scores of different frequency domain features, which are used to analyze the correlation degree of different frequency domain features with the current aging features;

[0070] Extracting the correlated multi-scale operation frequency domain data sensitive to the lithium battery aging data according to the importance scores.

[0071] Furthermore, the lithium battery aging frequency domain correlation model is a CNN-LSTM model based on the attention mechanism. The training process includes: using CNN to extract the spectral features of the historical lithium battery aging data and using the max-pooling operation to reduce the dimension; inputting the features extracted by CNN into the LSTM layer to learn the time series dependence; using the attention layer to learn the importance of different frequency domain features; finally, using the fully connected layer to output the aging prediction data; combining the aging prediction data, historical aging data and the loss function to train and optimize the model;

[0072] Furthermore, the way to obtain the importance scores is to analyze the attention weights of the attention layer in the lithium battery aging frequency domain correlation model. The larger the weight, the more important the corresponding frequency point / band; by setting an importance threshold, the features with scores higher than the threshold are selected as the sensitive correlated multi-scale operation frequency domain data.

[0073] This embodiment can use the lithium battery aging frequency domain correlation model to learn the relationship between different aging degrees of the lithium battery and the frequency domain characteristics. By obtaining the frequency point or band information that is most sensitive to the current aging change, the frequency domain prediction accuracy of the subsequent battery health state can be improved, thereby improving the accuracy and reliability of the online detection of the lithium battery health state.

[0074] Inputting the user behavior data and the lithium battery multi-scale operation frequency domain data into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain the lithium battery multi-scale frequency domain characteristic influence coefficient;

[0075] In this embodiment, considering that the user behavior habits will affect the actual operation status of the lithium battery, thereby changing the frequency domain characteristics under the operation state of the lithium battery, a lithium battery multi-scale frequency domain characteristic influence prediction model is proposed to obtain the frequency domain influence brought by the user behavior change.

[0076] Furthermore, the process of inputting the user behavior data and the lithium battery multi-scale operation frequency domain data into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain the lithium battery multi-scale frequency domain characteristic influence coefficient includes:

[0077] Train a lithium battery multi-scale frequency domain characteristic impact prediction model using historical user behavior data and historical lithium battery operation data in historical lithium battery data to obtain historical impact prediction model weights;

[0078] Input user behavior data and lithium battery multi-scale operation frequency domain data into the pre-trained lithium battery multi-scale frequency domain characteristic impact prediction model for processing, and update the historical impact prediction model weights to generate lithium battery multi-scale frequency domain characteristic impact coefficients;

[0079] Furthermore, the lithium battery multi-scale frequency domain characteristic impact prediction model includes: a data embedding layer, a fusion layer, a Transformer encoder, and a prediction layer;

[0080] Furthermore, the data embedding layer is used to map user behavior data and lithium battery multi-scale operation frequency domain data into user behavior embedding vectors and operation frequency domain embedding vectors; the fusion layer is used to perform channel splicing and fusion on the user behavior embedding vectors and operation frequency domain embedding vectors; the Transformer encoder is used to capture the correlation between user behavior and operation frequency domain characteristics using the multi-head self-attention mechanism to obtain encoded feature vectors; the prediction layer is used to map the encoded feature vectors to lithium battery multi-scale frequency domain characteristic impact coefficients using a fully connected layer.

[0081] This embodiment uses a lithium battery multi-scale frequency domain characteristic impact prediction model to learn the impact degree of user behavior changes on lithium battery multi-scale operation frequency domain data; by dynamically adjusting the frequency domain features in the lithium battery multi-scale frequency domain characteristic impact prediction model using lithium battery multi-scale frequency domain characteristic impact coefficients, it can effectively adjust the frequency domain characteristics of the lithium battery during operation to conform to the actual operation situation, thereby improving the accuracy and reliability of online detection of the lithium battery health state.

[0082] Input lithium battery multi-scale operation frequency domain data, associated multi-scale operation frequency domain data, and lithium battery multi-scale frequency domain characteristic impact coefficients into the lithium battery health state hybrid frequency domain prediction model to obtain a lithium battery health state detection value;

[0083] Furthermore, the structure of the lithium battery health state hybrid frequency domain prediction model is as Figure 2 shown, including: an input layer, a multi-scale frequency domain feature extraction layer, a cross-attention layer, an adaptive adjustment layer, a health state prediction layer, and an output layer;

[0084] Furthermore, the process of inputting lithium battery multi-scale operation frequency domain data, associated multi-scale operation frequency domain data, and lithium battery multi-scale frequency domain characteristic impact coefficients into the lithium battery health state hybrid frequency domain prediction model to obtain a lithium battery health state detection value includes:

[0085] The input layer of the hybrid frequency domain prediction model for the health state of a lithium battery receives multi-scale operating frequency domain data of the lithium battery, correlates the multi-scale operating frequency domain data, and the influence coefficient of the multi-scale frequency domain characteristics of the lithium battery;

[0086] Input the multi-scale operating frequency domain data of the lithium battery and the correlated multi-scale operating frequency domain data into the multi-scale frequency domain feature extraction layer of the hybrid frequency domain prediction model for the health state of the lithium battery to obtain the multi-scale operating frequency domain feature representation and the correlated multi-scale operating frequency domain feature representation of the lithium battery;

[0087] Input the multi-scale operating frequency domain feature representation and the correlated multi-scale operating frequency domain feature representation of the lithium battery into the cross-attention layer of the hybrid frequency domain prediction model for the health state of the lithium battery to obtain the weighted frequency domain feature representation;

[0088] Input the weighted frequency domain feature representation and the influence coefficient of the multi-scale frequency domain characteristics of the lithium battery into the adaptive adjustment layer of the hybrid frequency domain prediction model for the health state of the lithium battery to obtain the adaptive frequency domain adjustment feature representation;

[0089] Input the adaptive frequency domain adjustment feature representation into the health state prediction layer and the output layer of the hybrid frequency domain prediction model for the health state of the lithium battery in sequence to obtain the health state detection value of the lithium battery;

[0090] Furthermore, the multi-scale frequency domain feature extraction layer uses two independent gated recurrent units (GRUs) to extract features from the multi-scale operating frequency domain data and the correlated multi-scale operating frequency domain data of the lithium battery respectively;

[0091] Furthermore, the process of obtaining the weighted frequency domain feature representation includes: first, linearly map the multi-scale operating frequency domain feature representation and the correlated multi-scale operating frequency domain feature representation of the lithium battery to obtain the key vector and value vector of the multi-scale operating frequency domain feature representation of the lithium battery and the query vector of the correlated operating frequency domain feature; calculate the similarity between the query vector and the key vector using the dot product, and perform scaling and Softmax normalization operations on it to obtain the cross-attention weight; finally, weight the value vector using the cross-attention weight to obtain the weighted frequency domain feature representation;

[0092] Furthermore, the adaptive adjustment layer uses conditional attention to adjust the weighted frequency domain feature representation using the influence coefficient of the multi-scale frequency domain characteristics of the lithium battery, that is, after calculating the attention weight of the weighted frequency domain feature representation, use the influence coefficient of the multi-scale frequency domain characteristics of the lithium battery to adjust the attention weight;

[0093] Furthermore, the health state prediction layer uses a fully connected layer to predict the health state of the lithium battery.

[0094] In this embodiment, the hybrid frequency-domain prediction model for the health state of a lithium battery combines the multi-scale operating frequency-domain feature representation and the associated multi-scale operating frequency-domain feature representation of the lithium battery, and uses the cross-attention layer and frequency-domain characteristic adjustment to perform weighting and adjustment respectively. Through the weighting operation, the frequency-domain information beneficial to predicting the health state of the battery can be screened, and through the adjustment operation, the operating frequency-domain feature representation can be made to conform to the actual situation of the user to ensure its authenticity, thereby improving the accuracy and reliability of the online detection of the health state of the lithium battery.

[0095] Online detection of the health state of a lithium battery is performed according to the lithium battery health state detection value and the lithium battery health state threshold.

[0096] Further, the lithium battery health state threshold is set to 0.73; the setting of this threshold is related to the material, workmanship and model of the lithium battery, and can be adjusted according to the actual situation.

[0097] To illustrate the lithium battery health state detection value proposed by the present invention, three groups of test data from different users are randomly selected; the test data includes: lithium battery operation data, lithium battery aging data, and user behavior data; the lithium batteries are of the same model, and the data are respectively recorded as Data One, Data Two and Data Three; each group of data is successively subjected to preprocessing, aging frequency-domain correlation processing, frequency-domain characteristic influence prediction and battery health state frequency-domain detection to obtain the lithium battery health state detection value of each group and compare it according to the threshold to obtain the lithium battery health state detection result, and this detection result is shown in Table 1.

[0098] Table 1 Lithium Battery Health State Detection Results

[0099]

[0100] This embodiment provides an online detection method for the health state of a lithium battery based on frequency-domain analysis. This method first obtains lithium battery operation data, aging data, user behavior data and historical data; secondly, preprocesses the operation data to obtain the multi-scale operating frequency-domain data of the lithium battery, and inputs the aging data into the lithium battery aging frequency-domain correlation model to obtain the associated multi-scale operating frequency-domain data; then, inputs the user behavior data and the operating frequency-domain data into the lithium battery multi-scale frequency-domain characteristic influence prediction model to predict the lithium battery multi-scale frequency-domain characteristic influence coefficient; finally, inputs the operating frequency-domain data, the associated multi-scale operating frequency-domain data and the lithium battery multi-scale frequency-domain characteristic influence coefficient into the lithium battery health state hybrid frequency-domain prediction model to output the lithium battery health state detection value, and compares it with the set lithium battery health state threshold to realize the online detection of the health state of the lithium battery; the present invention can effectively improve the accuracy and reliability of the online detection of the health state of the lithium battery.

[0101] Embodiment Two:

[0102] The present invention also proposes an online detection system for the health state of lithium batteries based on frequency domain analysis. The structure of this system can be referred to Figure 3 , and it includes: a data acquisition unit, a data preprocessing unit, a lithium battery aging frequency domain correlation unit, a lithium battery frequency domain characteristic influence prediction unit, a lithium battery health state frequency domain detection unit, and an output unit;

[0103] The data acquisition unit is used to obtain lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data;

[0104] The data preprocessing unit is used to preprocess the lithium battery operation data to obtain multi-scale operation frequency domain data of the lithium battery; <S

[0105] The lithium battery aging frequency domain correlation unit is used to input the lithium battery aging data into the lithium battery aging frequency domain correlation model to obtain correlated multi-scale operation frequency domain data;

[0106] The lithium battery frequency domain characteristic influence prediction unit is used to input the user behavior data and the multi-scale operation frequency domain data of the lithium battery into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain the lithium battery multi-scale frequency domain characteristic influence coefficient;

[0107] The lithium battery health state frequency domain detection unit is used to input the multi-scale operation frequency domain data of the lithium battery, the correlated multi-scale operation frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient into the lithium battery health state hybrid frequency domain prediction model to obtain the lithium battery health state detection value;

[0108] Furthermore, the process by which the lithium battery health state frequency domain detection unit uses the lithium battery health state hybrid frequency domain prediction model to obtain the lithium battery health state detection value includes:

[0109] Using the input layer of the lithium battery health state hybrid frequency domain prediction model to receive the multi-scale operation frequency domain data of the lithium battery, the correlated multi-scale operation frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient; <S

[0110] Inputting the multi-scale operation frequency domain data of the lithium battery and the correlated multi-scale operation frequency domain data into the multi-scale frequency domain feature extraction layer of the lithium battery health state hybrid frequency domain prediction model to obtain the multi-scale operation frequency domain feature representation of the lithium battery and the correlated multi-scale operation frequency domain feature representation;

[0111] Inputting the multi-scale operation frequency domain feature representation of the lithium battery and the correlated multi-scale operation frequency domain feature representation into the cross-attention layer of the lithium battery health state hybrid frequency domain prediction model to obtain the weighted frequency domain feature representation;

[0112] The weighted frequency-domain feature representation and the lithium battery multi-scale frequency-domain characteristic influence coefficient are input into the adaptive adjustment layer of the lithium battery state of health hybrid frequency-domain prediction model to obtain an adaptive frequency-domain adjustment feature representation;

[0113] The adaptive frequency-domain adjustment feature representation is successively input into the state of health prediction layer and the output layer of the lithium battery state of health hybrid frequency-domain prediction model to obtain a lithium battery state of health detection value.

[0114] The output unit is used to output the online detection result of the lithium battery state of health.

[0115] In order to verify the effectiveness of the lithium battery state of health detection process proposed by the present invention, 500 groups of test data are randomly selected for the effectiveness test of the battery state of health detection scheme; three different detection schemes are used to process the test data to obtain the lithium battery state of health detection values of each scheme; then, according to manual inspection, the proportion of the detection results of each scheme within a reasonable range is obtained;

[0116] The detection schemes are respectively: the lithium battery state of health detection scheme proposed by the present invention, that is, combining aging frequency-domain correlation, frequency-domain characteristic influence prediction and state of health frequency-domain detection, denoted as Scheme 1; removing the aging frequency-domain correlation operation, denoted as Scheme 2; removing the frequency-domain characteristic influence prediction operation, denoted as Scheme 3;

[0117] The effectiveness test results of the lithium battery state of health detection scheme are shown in Table 2.

[0118] Table 2 Effectiveness test results of the lithium battery state of health detection scheme

[0119]

[0120] As can be seen from Table 2, the detection scheme of the present invention, that is, Scheme 1, has better detection results in terms of the effectiveness of the lithium battery state of health detection scheme than those using other schemes. Therefore, it is necessary to combine aging frequency-domain correlation, frequency-domain characteristic influence prediction and state of health frequency-domain detection, which is beneficial to improving the accuracy and reliability of the online detection of the lithium battery state of health.

[0121] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirits of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An online detection method for the state of health of a lithium battery based on frequency domain analysis, characterized in that, Including: Obtain the operating data of the lithium battery, the aging data of the lithium battery, the user behavior data, and the historical lithium battery data; Preprocess the operating data of the lithium battery and extract multi-scale frequency domain features to obtain the multi-scale operating frequency domain data of the lithium battery; Construct a lithium battery aging frequency domain correlation model, train it using the historical lithium battery data, and input the lithium battery aging data into the lithium battery aging frequency domain correlation model to obtain the correlated multi-scale operating frequency domain data; Input the user behavior data and the multi-scale operating frequency domain data of the lithium battery into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain the lithium battery multi-scale frequency domain characteristic influence coefficient; Input the multi-scale operating frequency domain data of the lithium battery, the correlated multi-scale operating frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient into the lithium battery health state hybrid frequency domain prediction model for adaptive adjustment and health state prediction to obtain the lithium battery health state detection value; Online detect the health state of the lithium battery according to the lithium battery health state detection value and the lithium battery health state threshold.

2. The online detection method for the state of health of a lithium battery based on frequency domain analysis according to claim 1, characterized in that The operating data of the lithium battery includes: remaining battery power, battery capacity, current, voltage, and charge-discharge cycle; the user behavior data includes: charge-discharge rate, charge-discharge depth, operating temperature, and idle time; the historical lithium battery data includes: historical lithium battery operating data, historical lithium battery aging data, historical user behavior data, and historical lithium battery health state data.

3. An online detection method for the health state of a lithium battery based on frequency domain analysis according to claim 1, characterized in that, The process of preprocessing the operating data of the lithium battery to obtain the multi-scale operating frequency domain data of the lithium battery includes: Clean the operating data of the lithium battery to obtain the first lithium battery operating data; Segment the first lithium battery operating data to obtain the second lithium battery operating data; Extract multi-scale frequency domain features from the second lithium battery operating data to obtain the third lithium battery operating data; Proofread and normalize the third lithium battery operating data to obtain the multi-scale operating frequency domain data of the lithium battery.

4. An online detection method for the health state of a lithium battery based on frequency domain analysis according to claim 1, characterized in that, The process of obtaining the correlated multi-scale operating frequency domain data includes: Train the lithium battery aging frequency domain correlation model using the historical lithium battery operating data and historical lithium battery aging data in the historical lithium battery data to obtain the pre-trained lithium battery aging frequency domain correlation model; Input the lithium battery aging data into the pre-trained lithium battery aging frequency domain correlation model to obtain the importance scores of different frequency domain features for analyzing the correlation degree of different frequency domain features with the current aging features; Extract the correlated multi-scale operating frequency domain data sensitive to the lithium battery aging data according to the importance scores.

5. The on-line detection method for the health state of a lithium battery based on frequency domain analysis according to claim 1, characterized in that, The process of obtaining the lithium battery multi-scale frequency domain characteristic influence coefficient includes: Train the lithium battery multi-scale frequency domain characteristic influence prediction model using the historical user behavior data and historical lithium battery operating data in the historical lithium battery data to obtain the historical influence prediction model weights; Input the user behavior data and the lithium battery multi-scale operation frequency domain data into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing, and update the weights of the historical influence prediction model to generate the lithium battery multi-scale frequency domain characteristic influence coefficient.

6. The online detection method for the health state of a lithium battery based on frequency domain analysis according to claim 1, characterized in that The process of obtaining the lithium battery health state detection value includes: Use the input layer of the lithium battery health state hybrid frequency domain prediction model to receive the lithium battery multi-scale operation frequency domain data, the associated multi-scale operation frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient; Input the lithium battery multi-scale operation frequency domain data and the associated multi-scale operation frequency domain data into the multi-scale frequency domain feature extraction layer of the lithium battery health state hybrid frequency domain prediction model to obtain the lithium battery multi-scale operation frequency domain feature representation and the associated multi-scale operation frequency domain feature representation; Input the lithium battery multi-scale operation frequency domain feature representation and the associated multi-scale operation frequency domain feature representation into the cross-attention layer of the lithium battery health state hybrid frequency domain prediction model to obtain the weighted frequency domain feature representation; Input the weighted frequency domain feature representation and the lithium battery multi-scale frequency domain characteristic influence coefficient into the adaptive adjustment layer of the lithium battery health state hybrid frequency domain prediction model to obtain the adaptive frequency domain adjustment feature representation; Input the adaptive frequency domain adjustment feature representation into the health state prediction layer and the output layer of the lithium battery health state hybrid frequency domain prediction model in sequence to obtain the lithium battery health state detection value.

7. An online detection system for the health state of a lithium battery based on frequency domain analysis, characterized in that, It includes: A data acquisition unit, a data preprocessing unit, a lithium battery aging frequency domain association unit, a lithium battery frequency domain characteristic influence prediction unit, a lithium battery health state frequency domain detection unit, and an output unit; the data acquisition unit is used to obtain lithium battery operation data, lithium battery aging data, user behavior data, and historical lithium battery data; the data preprocessing unit is used to preprocess the lithium battery operation data to obtain lithium battery multi-scale operation frequency domain data; the lithium battery aging frequency domain association unit is used to input the lithium battery aging data into the lithium battery aging frequency domain association model to obtain associated multi-scale operation frequency domain data; the lithium battery frequency domain characteristic influence prediction unit is used to input the user behavior data and the lithium battery multi-scale operation frequency domain data into the pre-trained lithium battery multi-scale frequency domain characteristic influence prediction model for processing to obtain the lithium battery multi-scale frequency domain characteristic influence coefficient; the lithium battery health state frequency domain detection unit is used to input the lithium battery multi-scale operation frequency domain data, the associated multi-scale operation frequency domain data, and the lithium battery multi-scale frequency domain characteristic influence coefficient into the lithium battery health state hybrid frequency domain prediction model to obtain the lithium battery health state detection value; the output unit is used to output the online detection result of the lithium battery health state.

Images