Lithium battery health estimation method and system based on transfer learning and multi-feature fusion
Through transfer learning and multi-feature fusion methods, using a small amount of data to estimate the health status of lithium batteries, solving the problem of real-time estimation that relies on a large amount of data in the prior art, and improving the safety and reliability of lithium batteries.
Patent Information
- Application Number
- CN202411198199.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-06
AI Technical Summary
The existing lithium battery health status estimation methods rely on a large amount of battery data, making it difficult to achieve accurate estimation in a real-time environment, which affects the safety and reliability of lithium battery operation.
Using transfer learning and multi-feature fusion methods, data preprocessing and feature extraction are obtained by obtaining preset charge and discharge cycle data sets, a transfer learning base model based on multi-layer stacked LSTM is constructed, and a small amount of data is used to estimate the health status of lithium batteries. .
It improves the efficiency and accuracy of estimation of the health status of lithium batteries, enhances the safety and reliability of lithium batteries, and solves the problem of scarce battery data caused by data acquisition difficulties.
Smart Images

Figure CN119087264A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of lithium battery technology, and in particular to a lithium battery health estimation method and system based on transfer learning and multi-feature fusion. Background Art
[0002] In recent years, lithium-ion batteries have become the preferred battery technology in many application fields because of their advantages such as high energy density, long life, low self-discharge rate, fast charging capability and environmental protection. They are widely used in mobile electronic devices, electric vehicles, energy storage and renewable energy. However, in practical applications, lithium batteries will have performance degradation problems, especially in high temperature environments. High temperature will accelerate the chemical reaction inside the battery, resulting in battery capacity decay and shortened cycle life. In addition, excessive temperature may also cause thermal runaway of the battery, increasing the risk of thermal fire or explosion of the battery. In order to ensure the safe and reliable operation of energy storage batteries and power batteries, it is necessary to monitor the status of power batteries efficiently, accurately and in real time through battery management systems. By monitoring the status parameters of the battery, such as voltage, current, temperature, etc., abnormal conditions can be discovered in time, measures can be taken to avoid accidents and failures, and improve safety. At the same time, monitoring indicators such as battery capacity decay and internal resistance change can evaluate the health of the battery and extend its service life. In addition, by understanding the battery status, the working condition of the battery can be optimized and controlled to improve performance and efficiency. Monitoring the charge and discharge status and remaining capacity of the battery helps to achieve energy management and maximize the use of battery energy. Finally, through fault diagnosis and prediction, faults can be discovered and handled in a timely manner, improving reliability and availability.
[0003] The health status of a battery refers to the performance, capacity, and life of the battery during use. It reflects the actual working capacity and remaining availability of the battery. The health status of a battery can be evaluated by a variety of indicators, including capacity decay, internal resistance change, charge and discharge efficiency, temperature characteristics, etc. By evaluating the health status of the battery, the reliability, safety, and performance of the system can be ensured while reducing resource consumption and environmental pollution. Therefore, battery health status assessment is an integral part of the battery management system.
[0004] At present, the methods used to estimate the health status of batteries are mainly based on experience, model and data-driven methods. Since the aging mechanism of batteries is very complex, involving multiple factors such as chemical reactions, electrochemical processes, temperature changes, etc., it is very difficult to find an accurate mathematical model to describe the aging process of batteries. Therefore, the empirical and model-based methods face the problems of difficult implementation and inaccurate evaluation results. The data-driven method learns battery aging information from a large amount of battery operation data through machine learning or deep learning models and algorithms, so as to accurately estimate the health status of lithium batteries. However, the existing data-driven methods all rely on a large amount of battery data. In practical applications, it is very difficult to obtain huge battery data in real time and complete model reasoning. Too much collected battery data will affect the efficiency of model reasoning, while too little collected battery data will affect the accuracy of model reasoning, thereby affecting the safety and reliability of lithium battery operation. Summary of the invention
[0005] The purpose of the present invention is to solve one of the technical problems existing in the prior art to at least a certain extent.
[0006] To this end, one purpose of an embodiment of the present invention is to provide a lithium battery health estimation method based on transfer learning and multi-feature fusion, which can complete the estimation of the health status of the lithium battery based on a small amount of battery data, thereby improving the efficiency and accuracy of the lithium battery health status estimation, thereby improving the safety and reliability of the lithium battery operation.
[0007] Another object of an embodiment of the present invention is to provide a lithium battery health estimation system based on transfer learning and multi-feature fusion.
[0008] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present invention include:
[0009] On the one hand, an embodiment of the present invention provides a lithium battery health estimation method based on transfer learning and multi-feature fusion, comprising the following steps:
[0010] Acquire a preset charge-discharge cycle data set, the charge-discharge cycle data set comprising a source data set and a target data set, and perform data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set;
[0011] Extracting features from the preprocessed data set to obtain fused health features, dividing the fused health features into time series to obtain a short-term health feature information set, and then randomly dividing the short-term health feature information set to obtain a training set and a test set;
[0012] Constructing a transfer learning base model based on a multi-layer stacked LSTM, inputting the training set corresponding to the source data set into the transfer learning base model for training, and obtaining the pre-trained transfer learning base model;
[0013] Fine-tune the pre-trained transfer learning base model, and input the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model;
[0014] The lithium battery health state estimation model is tested and verified by using the test set to obtain the trained lithium battery health state estimation model, and then the health state of the lithium battery to be tested is estimated according to the trained lithium battery health state estimation model.
[0015] Further, in one embodiment of the present invention, the step of performing data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set specifically includes:
[0016] Performing data cleaning on the charge-discharge cycle data set to obtain a cleaned charge-discharge cycle data set;
[0017] The cleaned charge-discharge cycle data set is smoothed to obtain the pre-processed data set.
[0018] Further, in one embodiment of the present invention, the step of extracting features from the preprocessed data set to obtain fused health features specifically includes:
[0019] Calculating according to the current-time curve in the preprocessed data set to obtain an electric quantity-time curve;
[0020] Calculating based on the charge-time curve and the voltage-time curve in the preprocessed data set to obtain an IC curve;
[0021] Performing interpolation processing and moving average filtering processing on the IC curve to obtain a filtered IC curve;
[0022] The filtered IC curve is sampled at equal distances within the effective voltage range to obtain a plurality of IC characteristic points, and the IC characteristic points are used as the first health characteristic;
[0023] intercepting a plurality of charging voltage ranges, calculating the charging time within each of the charging voltage ranges, obtaining a plurality of equal-pressure difference charging times, and using the equal-pressure difference charging time as a second health feature;
[0024] The first health feature and the second health feature are subjected to feature fusion to obtain the fused health feature.
[0025] Further, in one embodiment of the present invention, the step of performing time series division on the fused health features to obtain a short-term health feature information set specifically includes:
[0026] Acquire the fused health features of each charge and discharge cycle, and generate a health feature vector according to the fused health features;
[0027] The health feature vectors of continuous periods are staggered and divided into time series data in sequence according to the length of the time series to obtain the short-term health feature information set.
[0028] Furthermore, in one embodiment of the present invention, the step of constructing a transfer learning base model based on a multi-layer stacked LSTM specifically includes:
[0029] Construct forget gate based on sigmoid neural network layer and bitwise multiplication structure;
[0030] Construct memory gates based on sigmoid neural network layers, tanh neural network layers, bitwise multiplication structures, and bitwise addition structures;
[0031] Construct output gates based on sigmoid neural network layers, tanh neural network layers, and bitwise addition structures;
[0032] Constructing an LSTM network layer according to the forget gate, the memory gate, and the output gate, and generating a multi-layer stacked LSTM structure according to a plurality of the LSTM network layers;
[0033] The transfer learning base model is constructed according to the multi-layer stacked LSTM structure and the fully connected layer.
[0034] Furthermore, in one embodiment of the present invention, the calculation formula of the transfer learning base model is as follows:
[0035]
[0036] Among them, σ(x) represents the sigmoid function operation on x, tanh(x) represents the tanh function operation on x, and x t represents the tth element of the input sequence, h t-1 represents the hidden state of the previous time step, h t represents the hidden state at the current moment, C t-1 represents the cell state at the previous time step, Indicates the backup information used to update the unit status, C t represents the unit state at the current moment, f t 、i t and tRespectively represent the output of the forget gate, input gate, and output gate at the current moment, W xi , W hi and W ci represents the weight matrix of the input gate, b i represents the bias vector of the input gate, W zf , W hf and W cf represents the weight matrix of the forget gate, b f Represents the bias vector of the forget gate, W xo , W ho and W co represents the weight matrix of the output gate, b o represents the bias vector of the output gate, W zc and W hc represents the weight matrix used to update the unit state, b c Represents the bias vector used to update the cell state.
[0037] Further, in one embodiment of the present invention, the step of inputting the training set corresponding to the source data set into the transfer learning base model for training to obtain the pre-trained transfer learning base model specifically includes:
[0038] Inputting the training set corresponding to the source data set into the transfer learning base model to obtain a first model output value;
[0039] Calculating the first model output value and the true value of the training set using a preset first loss function to obtain a first loss value;
[0040] Based on the first loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the pre-trained transfer learning base model.
[0041] Furthermore, in one embodiment of the present invention, the step of fine-tuning the pre-trained transfer learning base model specifically includes:
[0042] Freeze the weight parameters of some network layers of the transfer learning base model based on a preset fine-tuning strategy, so that the frozen network layers retain the battery aging information learned from the source data set;
[0043] Based on the fine-tuning strategy, the weight parameters of the unfrozen network layers of the transfer learning base model are fine-tuned to obtain the transfer learning base model after model fine-tuning.
[0044] Furthermore, in one embodiment of the present invention, the step of inputting the training set corresponding to the target data set into the transfer learning base model after model fine-tuning for training to obtain the lithium battery health state estimation model specifically includes:
[0045] Inputting the training set corresponding to the target data set into the transfer learning base model after model fine-tuning to obtain a second model output value;
[0046] Calculating the output value of the second model and the true value of the training set using a preset second loss function to obtain a second loss value;
[0047] Based on the second loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the lithium battery health status estimation model.
[0048] On the other hand, an embodiment of the present invention provides a lithium battery health estimation system based on transfer learning and multi-feature fusion, including:
[0049] A data set acquisition module is used to acquire a preset charge-discharge cycle data set, wherein the charge-discharge cycle data set includes a source data set and a target data set, and perform data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set;
[0050] A data set division module, used to extract features from the preprocessed data set to obtain fused health features, divide the fused health features into time series to obtain a short-term health feature information set, and then randomly divide the short-term health feature information set to obtain a training set and a test set;
[0051] A base model pre-training module is used to construct a transfer learning base model based on a multi-layer stacked LSTM, input the training set corresponding to the source data set into the transfer learning base model for training, and obtain the pre-trained transfer learning base model;
[0052] A transfer learning training module, used to fine-tune the pre-trained transfer learning base model, and input the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model;
[0053] The health state estimation module is used to test and verify the lithium battery health state estimation model through the test set to obtain the trained lithium battery health state estimation model, and then estimate the health state of the lithium battery to be tested based on the trained lithium battery health state estimation model.
[0054] On the other hand, an embodiment of the present invention provides a lithium battery health estimation device based on transfer learning and multi-feature fusion, the lithium battery health estimation device including a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein the program, when executed by the processor, realizes the lithium battery health estimation method as described above.
[0055] On the other hand, an embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium used for computer-readable storage, and the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the lithium battery health estimation method as described above.
[0056] The advantages and beneficial effects of the present invention will be partly given in the following description, partly become apparent from the following description, or be understood through the practice of the present invention:
[0057] The embodiment of the present invention obtains a preset charge and discharge cycle data set, which includes a source data set and a target data set, performs data preprocessing on the charge and discharge cycle data set to obtain a preprocessed data set, then performs feature extraction on the preprocessed data set to obtain a fused health feature, and performs time series division on the fused health feature to obtain a short-term health feature information set, and then randomly divides the short-term health feature information set to obtain a training set and a test set, and then constructs a transfer learning base model based on a multi-layer stacked LSTM, inputs the training set corresponding to the source data set into the transfer learning base model for training, and obtains a pre-trained transfer learning base model, and then fine-tunes the pre-trained transfer learning base model, and inputs the training set corresponding to the target data set into the transfer learning base model after model fine-tuning for training to obtain a lithium battery health state estimation model, and then tests and verifies the lithium battery health state estimation model through a test set to obtain a trained lithium battery health state estimation model, and then estimates the health state of the lithium battery to be tested according to the trained lithium battery health state estimation model. The embodiment of the present invention can migrate the battery aging information learned from the source data set to the target data set to train a lithium battery health state estimation model. Based on the lithium battery health state estimation model, a small amount of battery data can be used to estimate the health state of the lithium battery to be tested, which can effectively solve the problem of scarce battery data caused by data collection difficulties in actual application scenarios, improve the efficiency and accuracy of lithium battery health state estimation, and thus improve the safety and reliability of lithium battery operation. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the technical solution in the embodiments of the present invention, the following introduction is made to the drawings required for use in the embodiments of the present invention. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solution of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0059] Figure 1 A flowchart of the steps of a lithium battery health estimation method based on transfer learning and multi-feature fusion provided in an embodiment of the present invention;
[0060] Figure 2 An IC curve diagram of a source data set provided by an embodiment of the present invention;
[0061] Figure 3 An IC curve diagram of a target data set provided by an embodiment of the present invention;
[0062] Figure 4 A voltage-time curve diagram of a source data set provided by an embodiment of the present invention;
[0063] Figure 5 A voltage-time curve diagram of a target data set provided by an embodiment of the present invention;
[0064] Figure 6 A schematic diagram of the training process of the transfer learning base model provided in an embodiment of the present invention;
[0065] Figure 7 A schematic diagram of test results of the transfer learning base model provided by an embodiment of the present invention on a source data set;
[0066] Figure 8 A schematic diagram of test results of a lithium battery health status estimation model provided by an embodiment of the present invention on a target data set;
[0067] Fig. 9 A schematic diagram of the structure of a lithium battery health estimation system based on transfer learning and multi-feature fusion provided in an embodiment of the present invention;
[0068] Fig.10 A schematic diagram of the hardware structure of a lithium battery health estimation device based on transfer learning and multi-feature fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0069] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limitations on the present application. It should be noted that, although the functional module division is performed in the system schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order from the module division in the system schematic diagram or the flow chart. For the step numbers in the following embodiments, they are only set for the convenience of explanation, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.
[0070] In the description of the present invention, the meaning of "a plurality" is two or more. If there is a description of the first or the second, it is only for the purpose of distinguishing the technical features, and it cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features. In addition, unless otherwise defined, all technical and scientific terms used in this document have the same meaning as those commonly understood by technicians in the technical field of this application. The terms used in this document are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.
[0071] The lithium battery health estimation method based on transfer learning and multi-feature fusion provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a set-top box, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the lithium battery health estimation method based on transfer learning and multi-feature fusion, etc., but is not limited to the above forms.
[0072] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0073] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with the relevant laws, regulations, and standards of the relevant countries and regions. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.
[0074] like Figure 1 FIG. 1 is a flow chart showing a method for estimating the health of a lithium battery based on transfer learning and multi-feature fusion according to an embodiment of the present invention. Figure 1 The embodiment of the present invention provides a lithium battery health estimation method based on transfer learning and multi-feature fusion, which specifically includes the following steps:
[0075] S101 , obtaining a preset charge-discharge cycle data set, the charge-discharge cycle data set including a source data set and a target data set, and performing data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set.
[0076] Specifically, lithium batteries are often in the form of battery modules and battery packs in practical applications, and their internal resistance, remaining capacity, polarization degree and other states and performance indicators are difficult to measure and obtain in real time. However, data such as voltage, current, temperature, and time of lithium batteries are relatively easier to obtain, and these data can be collected and recorded in real time by means of battery management systems, sensors or monitoring equipment. The time variable parameters including the terminal voltage at each moment in the lithium battery charging cycle, the charging current at each moment in the lithium battery charging cycle, and the time corresponding to each moment in the lithium battery charging cycle are obtained as the data of the charging cycle, and the data of multiple consecutive charging cycles are spliced together as the original data to obtain a charge and discharge cycle data set.
[0077] In some optional embodiments, a data set of charge and discharge cycles of lithium batteries of different operating conditions, different models, and different materials is obtained.
[0078] Since lithium batteries come from a wide variety of sources, the performance of lithium batteries from different manufacturers, different materials, different batches, different models, and even different operating conditions will be greatly affected. In a specific embodiment of the present invention, two data sets using batteries of different materials and models are used, and each battery data is classified according to different charge and discharge rates and temperatures.
[0079] In some optional embodiments, the charge-discharge cycle data set is divided into a source data set and a target data set;
[0080] Select one type of battery as the source domain battery, and select another type of battery as the target domain battery. The source domain represents the domain of existing knowledge, and the battery data from the source domain is divided into the source data set; the target domain represents the new knowledge domain that needs to be learned, and the battery data from the target domain is divided into the target data set. Generally speaking, the target data set is usually much smaller than the source data set. By effectively transferring the knowledge of the source domain to the target domain, even with only a small amount of target battery data, an accurate estimation of the health status of the target battery can be achieved.
[0081] Since noise or measurement errors are inevitable in the process of acquiring raw data, the raw data needs to be preprocessed to eliminate meaningless noise in the data, enhance the accuracy and reliability of the data, and better reveal the useful information behind the data, so that the subsequent work of extracting relevant health characteristics can proceed more smoothly.
[0082] As an optional implementation, the step of performing data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set specifically includes:
[0083] S1011 , performing data cleaning on the charge-discharge cycle data set to obtain a cleaned charge-discharge cycle data set.
[0084] Specifically, first import the dataset into an appropriate tool or library, such as Pandas, ensure that the dataset is in the correct format, and set the correct column names and indexes; understand the structure, content, and characteristics of the data by viewing the first few rows of the dataset, statistical summaries, and data visualization. This includes observing the size, data type, missing values, duplicate values, outliers, etc. of the data; identify and process missing values in the dataset, and choose to delete rows or columns containing missing values, or use appropriate methods to fill them, such as mean, median, mode, etc.; identify and process outliers in the dataset, and use statistical methods or visualization methods to detect outliers, and choose to delete, replace, or perform other processing as needed; convert data types, and convert the data types of voltage, temperature, and other values in the dataset to floating point numbers as needed. After completing data cleaning, save the cleaned dataset in an appropriate format (such as CSV, Excel, etc.) for subsequent analysis.
[0085] S1012, smoothing the cleaned charge and discharge cycle data set to obtain a preprocessed data set.
[0086] Specifically, during the collection process of the original battery data, the original data curve may be noisy due to problems such as sensor accuracy or measurement method errors. The curve is not smooth and has many burrs, which is not conducive to feature extraction. Therefore, the original data needs to be smoothed to obtain a smooth data set. The present invention adopts piecewise linear interpolation and moving average methods.
[0087] Piecewise linear interpolation is a commonly used data smoothing method that uses linear interpolation in each segment of the data to fill the gaps between data points, thereby making the data curve smoother. The specific steps are: 1. Divide the data into multiple segments, and the data trend in each segment is relatively smooth; 2. For the data in each segment, use linear interpolation to fill the gaps between data points; 3. Splice the interpolated data in each segment together to obtain a smooth data set.
[0088] In the embodiment of the present invention, the datasets used are the CACLE lithium battery dataset and the NASA lithium battery dataset, both of which have a voltage range of 2.7V to 4.2V, with 4 decimal places reserved. The voltage difference between different adjacent moments is often different. In order for the dQ / dV-V curve to approximate a continuous curve, the dQ / dV-V curve needs to be interpolated. In the interval [V i ,V i+1 ], the formula is as follows:
[0089]
[0090] x k =V i+kΔV (k = 1, 2…, n)
[0091]
[0092] where n represents the number of inserted nodes, ΔV = 0.0001V represents the interpolation gap, represents the inserted node, [V i , V i+1 represents the voltage interval for interpolation.
[0093] The moving average method is a commonly used data smoothing method for reducing noise in data and making the data curve smoother. It achieves the smoothing effect by calculating the average value of data within a certain window size in the data sequence. The specific steps are as follows: 1. Determine the window size of the moving average, that is, the number of data points considered when calculating the average value; 2. Starting from the first data point in the data sequence, calculate the moving average value of each data point in turn according to the window size; 3. Replace the corresponding data point in the original data sequence with the calculated moving average value. Assuming the time series is [x1, x2,..., x N , and taking the number of terms T for the moving average < N, the simple average moving calculation formula is:
[0094]
[0095] where x avg (i) represents the average value of T terms of data, N represents the total number of data terms, and T represents the number of terms selected for the moving average.
[0096] S102. Perform feature extraction on the preprocessed data set to obtain fused health features, perform time series partitioning on the fused health features to obtain a short-term health feature information set, and then randomly partition the short-term health feature information set to obtain a training set and a test set.
[0097] Health features are some meaningful features extracted from the original data, and these features are highly correlated with the degradation process and aging degree of lithium batteries. By extracting these features, the original data can be transformed into a more representative and information-rich form, thereby helping the model better learn and understand the health state of lithium batteries.
[0098] Further, as an optional implementation manner, for the step of performing feature extraction on the preprocessed data set to obtain fused health features, it specifically includes:
[0099] S1021. Calculate based on the current-time curve in the preprocessed data set to obtain the charge-time curve.
[0100] Specifically, calculate the integral of the charging current with respect to time to obtain the charging charge at each time point, and its calculation formula is as follows:
[0101]
[0102] Where Q represents the charge of the battery from time 0 to time T, i represents the total number of charge and discharge cycles, and I i Represents the current of the i-th charge and discharge cycle.
[0103] S1022, performing calculation based on the charge-time curve and the voltage-time curve in the preprocessed data set to obtain an IC curve.
[0104] Specifically, the calculation formula is as follows:
[0105]
[0106] Among them, Q t represents the charge at time t, Q t-1 Represents the charging capacity at time t-1, V t Represents the voltage value at time t, V t-1 Represents the voltage value at time t-1.
[0107] S1023, performing interpolation processing and moving average filtering processing on the IC curve to obtain a filtered IC curve.
[0108] Specifically, the embodiment of the present invention uses the interpolation and moving average processing method consistent with S1012 to smooth the IC curve. Such a processing method can effectively reduce the noise and fluctuation in the curve, making the IC curve smoother.
[0109] S1024, sampling the filtered IC curve at equal intervals within the effective voltage range to obtain multiple IC feature points, and using the IC feature points as the first health feature.
[0110] Specifically, the voltage range selected in this embodiment is 3.8V to 4.1V. Figure 2 FIG. 4 is an IC curve diagram of a source data set provided by an embodiment of the present invention, such as Figure 3 The IC curve diagram of the target data set provided by the embodiment of the present invention is shown. It can be clearly seen that the IC curve moves to the lower right with the increase in the number of charge and discharge cycles, so the change in the health state of the lithium battery can be characterized by the change in the IC curve in this area. By sampling the IC curve in this voltage range at equal distances, the amount of data can be reduced and sufficient information can be retained for subsequent model training.
[0111] S1025. Intercept multiple charging voltage ranges, calculate the charging time within each charging voltage range, obtain multiple equal-pressure difference charging times, and use the equal-pressure difference charging time as the second health feature.
[0112] Specifically, Figure 4 FIG. 4 is a voltage-time curve diagram of a source data set provided by an embodiment of the present invention, such as Figure 5 The voltage-time curve of the target data set provided by the embodiment of the present invention is shown. As the number of charge and discharge cycles of the lithium battery increases, polarization reaction occurs inside the lithium battery, resulting in an increase in the internal resistance of the lithium battery and a decrease in capacity. Therefore, under the same constant current charging conditions, the greater the number of charge and discharge cycles, the faster the voltage rises at the lithium battery terminal, and the shorter the charging time within the same voltage range. Therefore, the change in the health status of the lithium battery can be characterized by the change in the equal voltage difference charging time. The equal voltage difference charging time calculation formula is:
[0113]
[0114] Among them, V1 is the lower limit of the voltage range, V2 is the upper limit of the voltage range, Indicates the time when the voltage reaches V1, Indicates the time when the voltage reaches V2. The values of V1 and V2 can select multiple sets of valid values within the effective range of the lithium battery starting voltage and cut-off voltage to obtain multiple health characteristics.
[0115] Specifically, the embodiment of the present invention selects 8 voltage ranges: 3.4V-4.2V, 3.5V-4.2V, 3.6V-4.2V, 3.7V-4.2V, 3.8V-4.2V, 3.9V-4.2V, 4.0V-4.2V, and 4.1V-4.2V.
[0116] S1026. Fusing the first health feature and the second health feature to obtain a fused health feature.
[0117] As an optional implementation, the step of dividing the fused health features into time series to obtain a short-term health feature information set specifically includes:
[0118] S1027, obtaining fused health features of each charge and discharge cycle, and generating a health feature vector according to the fused health features;
[0119] S1028. Divide the health feature vectors of continuous periods into time series data in a staggered manner according to the length of the time series to obtain a short-term health feature information set.
[0120] Specifically, a health feature vector X with a length of M is extracted from the i-th charge and discharge cycle of the lithium battery. i =[x i1 ,x i2 ...x iM ], then the characteristic matrix composed of the healthy characteristic vectors of all consecutive cycles is [X1,X2...X N ], divide the time series into [X1:L ,X 2:+1 ,...X N-L+1:N ], where N is the total number of charge and discharge cycles and T is the length of the time series.
[0121] In some optional embodiments, this embodiment adopts a holdout method to randomly divide the data set (including the source data set and the target data set) into a training set and a test set in a ratio of 3:1, and then uses the training set to train the model. By evaluating the performance of the model on the test set, the effectiveness and generalization ability of the model can be judged, and necessary adjustments and improvements can be made.
[0122] S103: construct a transfer learning base model based on a multi-layer stacked LSTM, input the training set corresponding to the source data set into the transfer learning base model for training, and obtain a pre-trained transfer learning base model.
[0123] Specifically, LSTM (Long Short-Term Memory) is a variant of Recurrent Neural Network (RNN) that is specifically designed to process sequence data. Compared with traditional RNN, LSTM effectively solves the long-term dependency problem by introducing a gating mechanism. The core idea of LSTM is to introduce three gating units: input gate, forget gate, and output gate. These gating units control the flow of information by learning parameters, thereby effectively modeling and memorizing sequence data. In LSTM, the input gate determines which information will be updated and stored in the cell state, the forget gate determines which information will be ignored and forgotten, and the output gate determines how the cell state is mapped to the output. The cell state of LSTM is similar to the hidden state in traditional RNN, but it has a stronger memory capacity. Through the gating mechanism, LSTM can selectively update and forget the information in the cell state, thereby better capturing long-term dependencies.
[0124] As an optional implementation, the step of constructing a transfer learning base model based on multi-layer stacked LSTM specifically includes:
[0125] S1031, construct forget gate based on sigmoid neural network layer and bitwise multiplication structure;
[0126] S1032, constructing a memory gate based on a sigmoid neural network layer, a tanh neural network layer, a bitwise multiplication structure, and a bitwise addition structure;
[0127] S1033, constructing an output gate based on a sigmoid neural network layer, a tanh neural network layer, and a bitwise addition structure;
[0128] S1034, constructing an LSTM network layer according to the forget gate, the memory gate, and the output gate, and generating a multi-layer stacked LSTM structure according to the multiple LSTM network layers;
[0129] S1035. Construct a transfer learning base model based on a multi-layer stacked LSTM structure and a fully connected layer.
[0130] Specifically, the core of LSTM is the cell state, denoted by C t , which is used to save the current LSTM state information and pass it to the LSTM at the next moment. The current LSTM receives the cell state C from the previous moment t-1 , the input x received by the current LSTM t Together they generate the current LSTM cell state C t , and C t Passed to the LSTM at the next moment.
[0131] like Figure 6 The figure shows a training process diagram of the transfer learning base model provided by an embodiment of the present invention. The transfer learning base model of the embodiment of the present invention is stacked by two LSTM stacked network layers and one fully connected layer (FC). The first LSTM stacked network layer of the transfer learning base model is a weight freezing layer, and the second LSTM stacked network layer and the fully connected layer are fine-tuning layers. The first LSTM stacked network layer is used to learn the aging information of the battery in the source data set from the source data set, and save the information learned from the source data by freezing the weight parameters of the network layer; the second LSTM stacked network layer and the fully connected layer are fine-tuning layers, which are used to learn the aging information of the target battery from the target data set. The battery aging information learned by the freezing layer and the fine-tuning layer is combined to complete the migration of the battery aging information from the source data set to the target data set.
[0132] The LSTM network layer includes a forget gate, a memory gate, and an output gate; the forget gate is constructed by a sigmoid neural network layer and a bitwise multiplication structure to determine the cell state C t-1 What information in the memory is forgotten; the memory gate is constructed by the input gate, tanh neural network layer, bitwise multiplication structure and bitwise addition structure. The role of the memory gate is to determine the new input information x t and h t-1 Which information in will be retained? The output gate is constructed by the simgoid neural network layer, the tanh neural network layer and the bitwise addition structure. The output gate transmits the cell state C after the forget gate and the memory gate at time t-1. t-1 , and the output signal h at time t-1 t-1 and the input signal x at time tt Integrate them together as the output signal at the current moment. t and h t-1 After a sigmoid neural network layer, a value between 0 and 1 is output. t ,C t After a tanh function, and at o t Multiply to get the output signal h t , while h t It is also passed to the next stage as the input signal at the next moment.
[0133] As an optional implementation, the calculation formula of the transfer learning base model is as follows:
[0134]
[0135] Among them, σ(x) represents the sigmoid function operation on x, tanh(x) represents the tanh function operation on x, and x t represents the tth element of the input sequence, h t-1 represents the hidden state of the previous time step, h t represents the hidden state at the current moment, C t-1 represents the cell state at the previous time step, Indicates the backup information used to update the unit status, C t represents the unit state at the current moment, f t 、i t and t Respectively represent the output of the forget gate, input gate, and output gate at the current moment, W xi , W hi and W ci represents the weight matrix of the input gate, b i represents the bias vector of the input gate, W zf , W hf and W cf represents the weight matrix of the forget gate, b f Represents the bias vector of the forget gate, W xo , W ho and W co represents the weight matrix of the output gate, b o represents the bias vector of the output gate, W zc and W hc represents the weight matrix used to update the unit state, b c Represents the bias vector used to update the cell state.
[0136] As an optional implementation, the training set corresponding to the source data set is input into the transfer learning base model for training to obtain the pre-trained transfer learning base model. The step specifically includes:
[0137] S1036, inputting the training set corresponding to the source data set into the transfer learning base model to obtain a first model output value;
[0138] S1037, using a preset first loss function to calculate the first model output value and the true value of the training set to obtain a first loss value;
[0139] S1038. Based on the first loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the pre-trained transfer learning base model.
[0140] Specifically, the training set of the source data set is input into the transfer learning base model. In the specific embodiment of the present invention, the data dimension of the source data set is cycle_size*seq_len*feature_size, where cycle_size is the number of lithium battery charging cycles, seq_len is the length of the divided short time series, and feature_size is the dimension of the input feature. The dimension of train_x is 2850*10*33, the dimension of train_y is 2850*1, the dimension of test_x is 1045*10*33, and the dimension of test_y is 1045*1.
[0141] The mean square error loss function is used to calculate the model output value and the true value of the training set to obtain the error value. The calculation formula is as follows:
[0142]
[0143] Among them, MSE represents mean square error, Represents the model output value, y i represents the true value of the training set, and n is the number of samples.
[0144] Through the BP algorithm consisting of two processes, information forward propagation and error back propagation, the model parameters are optimized to minimize the error and finally obtain the pre-trained base model;
[0145] The training optimization steps are repeated until a predetermined number of training times is reached, and a pre-trained transfer learning base model is obtained, which has good performance in estimating the battery health state on the source data set.
[0146] S104, fine-tuning the pre-trained transfer learning base model, and inputting the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model.
[0147] As an optional implementation, the step of fine-tuning the pre-trained transfer learning base model specifically includes:
[0148] S1041, freezing weight parameters of some network layers of the transfer learning base model based on a preset fine-tuning strategy, so that the frozen network layers retain the battery aging information learned from the source data set;
[0149] S1042. Fine-tune the weight parameters of the unfrozen network layers of the transfer learning base model based on the fine-tuning strategy to obtain the transfer learning base model after model fine-tuning.
[0150] Specifically, the weight parameters of the first LSTM stacked network layer in the pre-trained transfer learning base model are frozen, that is, the requires_grad attribute of the LSTM network layer is set to False, so that the parameters of this layer will not be updated by gradient in the subsequent training process, and only the weights of the unfrozen layers are updated, thereby retaining the battery aging information learned by the model from the source data.
[0151] As an optional implementation, the training set corresponding to the target data set is input into the transfer learning base model after model fine-tuning for training to obtain the lithium battery health state estimation model, which specifically includes:
[0152] S1043, inputting the training set corresponding to the target data set into the transfer learning base model after model fine-tuning to obtain a second model output value;
[0153] S1044, using a preset second loss function to calculate the second model output value and the true value of the training set to obtain a second loss value;
[0154] S1045. Based on the second loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the lithium battery health status estimation model.
[0155] Specifically, the training set of the target data set is input into the pre-trained transfer learning base model for retraining. In an embodiment of the present invention, the data dimension of the source data set is cycle_size*seq_len*feature_size, where cycle_size is the number of lithium battery charging cycles, seq_len is the length of the divided short time series, and feature_size is the dimension of the input feature. The dimension of train_x is 18*10*33, the dimension of train_y is 18*1, the dimension of test_x is 157*10*33, and the dimension of test_y is 157*1. Using the model training process consistent with that in S103, a lithium battery health status estimation model is finally obtained, which has good performance in estimating the battery health status on the target data set.
[0156] S105, testing and verifying the lithium battery health state estimation model through a test set to obtain a trained lithium battery health state estimation model, and then estimating the health state of the lithium battery to be tested according to the trained lithium battery health state estimation model.
[0157] In some optional embodiments, after 5 training cycles, the test data set is input into the trained lithium battery health state estimation model for testing to obtain test results.
[0158] Specifically, after 5 training cycles, the test data set is input into the trained lithium battery health state estimation model for testing to obtain the estimation result. The error between the estimation result and the true value is calculated as the performance index of the current model. The smaller the error, the better the model performance. The calculation formula of the performance index is as follows:
[0159]
[0160] Where MAPE represents the performance evaluation index, Represents the model output prediction value, y i represents the true value of the test set, and n represents the number of samples.
[0161] The specific test of the embodiment of the present invention is to select the optimizer Adam for model training, the learning rate lr = 0.01, and the number of iterations epoch = 3000. The test results of the pre-trained transfer learning base model on the test set of the source data set are as follows: Figure 7 As shown, the MAPE error is 0.46%; the test results of the finally trained lithium battery health status estimation model on the test set of the target data set are as follows Figure 8 As shown, the MAPE error is 1.01%.
[0162] In some optional embodiments, the step of estimating the health state of the lithium battery to be tested according to the trained lithium battery health state estimation model specifically includes:
[0163] S1051, obtaining real-time charge and discharge information of the lithium battery to be tested;
[0164] S1052: input the real-time charge and discharge information into the trained lithium battery health state estimation model to obtain the health state estimation result of the lithium battery to be tested.
[0165] The above is an explanation of the method steps of the embodiment of the present invention. It can be understood that the embodiment of the present invention can migrate the battery aging information learned from the source data set to the target data set to train a lithium battery health state estimation model, based on which a small amount of battery data can be used to estimate the health state of the lithium battery to be tested, which can effectively solve the problem of scarce battery data caused by data collection difficulties in actual application scenarios, improve the efficiency and accuracy of lithium battery health state estimation, and thus improve the safety and reliability of lithium battery operation.
[0166] like Fig. 9 FIG. 1 is a schematic diagram of a lithium battery health estimation system based on transfer learning and multi-feature fusion according to an embodiment of the present invention. Fig. 9 , an embodiment of the present invention provides a lithium battery health estimation system based on transfer learning and multi-feature fusion, including:
[0167] A data set acquisition module is used to acquire a preset charge-discharge cycle data set, the charge-discharge cycle data set includes a source data set and a target data set, and perform data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set;
[0168] A data set partitioning module is used to extract features from the preprocessed data set to obtain fused health features, and to perform time series partitioning on the fused health features to obtain a short-term health feature information set, and then to perform random partitioning on the short-term health feature information set to obtain a training set and a test set;
[0169] The base model pre-training module is used to build a transfer learning base model based on multi-layer stacked LSTM, input the training set corresponding to the source data set into the transfer learning base model for training, and obtain the pre-trained transfer learning base model;
[0170] A transfer learning training module is used to fine-tune the pre-trained transfer learning base model, and input the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model;
[0171] The health state estimation module is used to test and verify the lithium battery health state estimation model through a test set to obtain a trained lithium battery health state estimation model, and then estimate the health state of the lithium battery to be tested based on the trained lithium battery health state estimation model.
[0172] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0173] The embodiment of the present invention also provides a lithium battery health estimation device based on transfer learning and multi-feature fusion, the lithium battery health estimation device includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, and when the program is executed by the processor, the lithium battery health estimation method based on transfer learning and multi-feature fusion is realized. The lithium battery health estimation device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.
[0174] like Fig.10 FIG. 1 is a schematic diagram of the hardware structure of a lithium battery health estimation device based on transfer learning and multi-feature fusion provided by an embodiment of the present invention, referring to FIG. Fig.10 , an embodiment of the present invention provides a lithium battery health estimation device based on transfer learning and multi-feature fusion, comprising:
[0175] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention;
[0176] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the lithium battery health estimation method of the embodiment of the present invention;
[0177] Input / output interface 1003, used to implement information input and output;
[0178] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);
[0179] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );
[0180] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .
[0181] An embodiment of the present invention also provides a storage medium, which is a computer-readable storage medium used for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the above-mentioned lithium battery health estimation method based on transfer learning and multi-feature fusion.
[0182] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0183] The embodiment of the present invention also discloses a computer program product or a computer program, wherein the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes Figure 1 The method shown.
[0184] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the above-mentioned boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.
[0185] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified to the contrary, one or more of the above-mentioned functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.
[0186] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the above methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0187] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.
[0188] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the above-mentioned program is printed, since the above-mentioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or processing in other suitable ways as necessary, and then stored in a computer memory.
[0189] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0190] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0191] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.
[0192] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A lithium battery health estimation method based on transfer learning and multi-feature fusion, characterized in that: The following steps are involved: Acquire a preset charge-discharge cycle data set, the charge-discharge cycle data set comprising a source data set and a target data set, and perform data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set; Extracting features from the preprocessed data set to obtain fused health features, dividing the fused health features into time series to obtain a short-term health feature information set, and then randomly dividing the short-term health feature information set to obtain a training set and a test set; Constructing a transfer learning base model based on a multi-layer stacked LSTM, inputting the training set corresponding to the source data set into the transfer learning base model for training, and obtaining the pre-trained transfer learning base model; Fine-tune the pre-trained transfer learning base model, and input the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model; The lithium battery health state estimation model is tested and verified by using the test set to obtain the trained lithium battery health state estimation model, and then the health state of the lithium battery to be tested is estimated according to the trained lithium battery health state estimation model.
2. According to claim 1, a lithium battery health estimation method based on transfer learning and multi-feature fusion is characterized in that: The step of performing data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set specifically includes: Performing data cleaning on the charge-discharge cycle data set to obtain a cleaned charge-discharge cycle data set; The cleaned charge-discharge cycle data set is smoothed to obtain the pre-processed data set.
3. According to claim 1, a lithium battery health estimation method based on transfer learning and multi-feature fusion is characterized in that: The step of extracting features from the preprocessed data set to obtain fused health features specifically includes: Calculating according to the current-time curve in the preprocessed data set to obtain an electric quantity-time curve; Calculating based on the charge-time curve and the voltage-time curve in the preprocessed data set to obtain an IC curve; Performing interpolation processing and moving average filtering processing on the IC curve to obtain a filtered IC curve; The filtered IC curve is sampled at equal distances within the effective voltage range to obtain a plurality of IC characteristic points, and the IC characteristic points are used as the first health characteristic; intercepting a plurality of charging voltage ranges, calculating the charging time within each of the charging voltage ranges, obtaining a plurality of equal-pressure difference charging times, and using the equal-pressure difference charging time as a second health feature; The first health feature and the second health feature are subjected to feature fusion to obtain the fused health feature.
4. According to claim 1, a lithium battery health estimation method based on transfer learning and multi-feature fusion is characterized in that: The step of dividing the fused health features into time series to obtain a short-term health feature information set specifically includes: Acquire the fused health features of each charge and discharge cycle, and generate a health feature vector according to the fused health features; The health feature vectors of continuous periods are staggered and divided into time series data in sequence according to the length of the time series to obtain the short-term health feature information set.
5. According to claim 1, a lithium battery health estimation method based on transfer learning and multi-feature fusion is characterized in that: The step of constructing a transfer learning base model based on multi-layer stacked LSTM specifically includes: Construct forget gate based on sigmoid neural network layer and bitwise multiplication structure; Construct memory gates based on sigmoid neural network layers, tanh neural network layers, bitwise multiplication structures, and bitwise addition structures; Construct output gates based on sigmoid neural network layers, tanh neural network layers, and bitwise addition structures; Constructing an LSTM network layer according to the forget gate, the memory gate, and the output gate, and generating a multi-layer stacked LSTM structure according to a plurality of the LSTM network layers; The transfer learning base model is constructed according to the multi-layer stacked LSTM structure and the fully connected layer.
6. A lithium battery health estimation method based on transfer learning and multi-feature fusion according to claim 5, characterized in that: The calculation formula of the transfer learning base model is as follows: Among them, σ(x) represents the sigmoid function operation on x, tanh(x) represents the tanh function operation on x, and x t represents the tth element of the input sequence, h t-1 represents the hidden state of the previous time step, h t represents the hidden state at the current moment, C t-1 represents the cell state at the previous time step, Indicates the backup information used to update the unit status, C t represents the unit state at the current moment, f t 、i t and t Respectively represent the output of the forget gate, input gate, and output gate at the current moment, W xi , W hi and W ci represents the weight matrix of the input gate, b i represents the bias vector of the input gate, W zf , W hf and W cf represents the weight matrix of the forget gate, b f Represents the bias vector of the forget gate, W xo , W ho and W co represents the weight matrix of the output gate, b o Represents the bias vector of the output gate, W zc and W hc represents the weight matrix used to update the unit state, b c Represents the bias vector used to update the cell state.
7. The lithium battery health estimation method based on transfer learning and multi-feature fusion according to claim 1 is characterized in that: The step of inputting the training set corresponding to the source data set into the transfer learning base model for training to obtain the pre-trained transfer learning base model specifically includes: Inputting the training set corresponding to the source data set into the transfer learning base model to obtain a first model output value; Calculating the first model output value and the true value of the training set using a preset first loss function to obtain a first loss value; Based on the first loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the pre-trained transfer learning base model.
8. The lithium battery health estimation method based on transfer learning and multi-feature fusion according to claim 1 is characterized in that: The step of fine-tuning the pre-trained transfer learning base model specifically includes: Freeze the weight parameters of some network layers of the transfer learning base model based on a preset fine-tuning strategy, so that the frozen network layers retain the battery aging information learned from the source data set; Based on the fine-tuning strategy, the weight parameters of the unfrozen network layers of the transfer learning base model are fine-tuned to obtain the transfer learning base model after model fine-tuning.
9. A lithium battery health estimation method based on transfer learning and multi-feature fusion according to any one of claims 1 to 8, characterized in that: The step of inputting the training set corresponding to the target data set into the transfer learning base model after model fine-tuning for training to obtain the lithium battery health state estimation model specifically includes: Inputting the training set corresponding to the target data set into the transfer learning base model after model fine-tuning to obtain a second model output value; Calculating the output value of the second model and the true value of the training set using a preset second loss function to obtain a second loss value; Based on the second loss value, the model parameters of the transfer learning base model are optimized by the BP algorithm and the gradient descent method, and the step of inputting the training set corresponding to the source data set into the transfer learning base model is returned until the preset number of training times is reached, and the training is stopped to obtain the lithium battery health status estimation model.
10. A lithium battery health estimation system based on transfer learning and multi-feature fusion, characterized in that: include: A data set acquisition module is used to acquire a preset charge-discharge cycle data set, wherein the charge-discharge cycle data set includes a source data set and a target data set, and perform data preprocessing on the charge-discharge cycle data set to obtain a preprocessed data set; A data set division module, used to extract features from the preprocessed data set to obtain fused health features, divide the fused health features into time series to obtain a short-term health feature information set, and then randomly divide the short-term health feature information set to obtain a training set and a test set; A base model pre-training module is used to construct a transfer learning base model based on a multi-layer stacked LSTM, input the training set corresponding to the source data set into the transfer learning base model for training, and obtain the pre-trained transfer learning base model; A transfer learning training module, used to fine-tune the pre-trained transfer learning base model, and input the training set corresponding to the target data set into the fine-tuned transfer learning base model for training to obtain a lithium battery health status estimation model; The health state estimation module is used to test and verify the lithium battery health state estimation model through the test set to obtain the trained lithium battery health state estimation model, and then estimate the health state of the lithium battery to be tested based on the trained lithium battery health state estimation model.
Citation Information
Patent Citations
Rotary machinery fault diagnosis method based on transfer learning
CN113392881A
Multi-type energy storage battery capacity estimation method based on transfer learning fine tuning strategy
CN115221710A
Lithium battery health state estimation method based on hybrid neural network and transfer learning
CN116679232A
Lithium battery health prediction method and system based on LSTM and multiple features
CN116908699A
Cited By
Hybrid energy storage battery state monitoring method and system based on big data
CN119780745A
A big data-based hybrid energy storage battery state monitoring method and system
CN119780745B
Lithium battery charging strategy sample efficiency enhancement method and system based on model short branch deduction
CN120142941A
A method and system for estimating the state of health of a lithium battery in a wide temperature range based on MGPR and improved timing RBF
CN120214615B
Electric vehicle battery full life cycle life evaluation method based on transfer learning
CN121188489A
