CN-121978544-A - Hybrid deep learning-based battery remaining life prediction method and system

CN121978544ACN 121978544 ACN121978544 ACN 121978544ACN-121978544-A

Abstract

The invention discloses a battery remaining life prediction method and a system based on mixed deep learning, wherein the battery remaining life prediction method is used for extracting an effective charging interval from multidimensional time sequence data of a battery in a charging and discharging process and predicting the battery health state based on a battery health state sample and charging curve characteristics extracted from the effective charging interval; meanwhile, the invention constructs a battery health state prediction model combining a convolutional neural network and a long-short-term memory network, the battery health state prediction model deeply extracts the characteristic relation from the multidimensional original time sequence data through the convolutional neural network, and then takes the extracted characteristic sequence as the input of the long-short-term memory network to learn the descending trend of the battery health state, so that the degradation characteristic in the data can be better captured, and the prediction result of the battery health state and the residual service life becomes more accurate.

Inventors

XIE ZUOQI
FANG YINFENG

Assignees

杭州芯锐特科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A battery remaining life prediction method based on hybrid deep learning is characterized by comprising the following steps: the method comprises the steps of acquiring multi-dimensional time sequence data of a battery, identifying different effective charging intervals from the multi-dimensional time sequence data of the battery, marking cycle times corresponding to each effective charging interval, acquiring a battery health state value corresponding to each effective charging interval, constructing a battery health state sample based on the battery health state value and the cycle times, and extracting charging curve characteristics from the multi-dimensional time sequence data based on the battery health state sample; The method comprises the steps of constructing a battery health state prediction model, wherein the battery health state prediction model comprises a feature extraction module, a trend analysis module and a full connection layer which are sequentially connected, the feature extraction module is used for extracting local features from charging curve features, the trend analysis module is used for analyzing the evolution rule of the local features along with time, and the full connection layer is used for mapping the output result of the trend analysis module to a final battery health state value; The method comprises the steps of training a battery health state prediction model by using a battery health state sample and charging curve characteristics, predicting a battery health state value corresponding to the latest cycle number of a tested battery by using the trained battery health state prediction model, fitting the battery health state value corresponding to the historical cycle number and the latest cycle number, obtaining the life end cycle number by using a fitting curve, and completing the prediction of the remaining life of the battery by taking the difference value between the life end cycle number and the latest cycle number as the remaining service life.
2. The method for predicting the residual life of a battery based on mixed deep learning of claim 1, wherein the method is characterized in that before extracting the characteristics of a charging curve, the battery health state sample is subjected to multistage cleaning, and the specific process is as follows: and respectively carrying out first-order polynomial fitting on the remaining battery health state samples of different batteries, obtaining the slope of a straight line after fitting, and if the number of the remaining battery health state samples of the batteries is smaller than a sample threshold value or the slope is not smaller than zero, removing all the battery health state samples corresponding to the batteries.
3. The method for predicting remaining life of a battery based on hybrid deep learning of claim 1, wherein the effective charging interval satisfies the following condition: the starting point of the effective charging interval is larger than the current threshold value, the end point of the effective charging interval is smaller than the current threshold value, the number of data points in the effective charging interval is not smaller than the data point threshold value, and the residual capacity percentage corresponding to the end point of the effective charging interval is 100%.
4. The battery remaining life prediction method based on hybrid deep learning of claim 1 is characterized in that the battery state of health value corresponding to the effective charging interval is the battery state of health value of the effective charging interval at a stable point, and the stable point obtaining method comprises the step of adding a set index interval on an end index of the effective charging interval to obtain a stable point index.
5. The method for predicting the residual life of a battery based on mixed deep learning of claim 1, wherein the fitted curve is obtained by fitting a second-order polynomial function to a state of health value of the battery.
6. The method for predicting the residual life of a battery based on hybrid deep learning of claim 1, wherein the training process of the battery state of health prediction model is to train the battery state of health prediction model by taking a charging curve characteristic corresponding to historical cycle times as an input of the battery state of health prediction model and a battery state of health value corresponding to current cycle times as a label, and to update model parameters by taking a root mean square error as a loss function.
7. The method for predicting the residual life of a battery based on hybrid deep learning of claim 1 wherein said multi-dimensional time series data comprises a total voltage, a current, a main board temperature, a residual capacity percentage, a residual capacity and a nominal capacity, and said charging curve characteristic is composed of the total voltage, the current and the main board temperature.
8. The method for predicting remaining battery life based on hybrid deep learning of claim 1, wherein the battery state of health value is a ratio of a remaining capacity to a nominal capacity.
9. The method for predicting the residual life of a battery based on hybrid deep learning of claim 1, wherein the feature extraction module adopts a convolutional neural network comprising a plurality of convolutional blocks and a flattening layer which are sequentially connected, and the trend analysis module adopts a long-term and short-term memory network.
10. A battery remaining life prediction system based on hybrid deep learning is characterized by being used for executing the battery remaining life prediction method based on hybrid deep learning, and comprises a data acquisition module, a feature extraction module, a battery health state prediction module and a remaining life prediction module, wherein the data acquisition module is used for acquiring multidimensional time sequence data of a battery in a charging and discharging process, the feature module is used for extracting an effective charging interval from the multidimensional time sequence data and acquiring a battery health state sample and charging curve feature corresponding to the effective charging interval, the battery health state prediction module is used for predicting a battery health state according to the charging curve feature, and the remaining life prediction module is used for fitting the battery health state and acquiring the remaining life according to a fitting curve.

Description

Hybrid deep learning-based battery remaining life prediction method and system Technical Field The invention belongs to the technical field of battery health management, and particularly relates to a battery residual life prediction method and system based on hybrid deep learning. Background In recent years, with the vigorous development of new energy industry, vehicles such as battery cars have been popularized on a large scale, and particularly under the high-frequency use situations such as takeaway, express delivery and the like, the performance, the service life and the safety of the battery become important as the core power source. In order to improve the working efficiency and the user experience, the battery replacement by the charging cabinet becomes a novel power conversion mode, and the state of health and the residual life of each battery in the charging cabinet are accurately estimated and predicted, so that the battery replacement mode becomes a core for realizing fine asset management. The prediction of the residual service life (RUL) of the battery is a key technology in Battery Health Management (BHM), and can early warn potential safety risks of the battery in advance, so that a key decision basis is provided for maintenance, replacement and elimination of the battery, and the operation cost is remarkably reduced, and the safety is improved. Traditional battery life prediction methods rely primarily on simplified physical models, empirical models, or simple threshold alarms. For example, by recording the number of cycles or the time of use of the battery, or by triggering an alarm when a single indicator of voltage, temperature, etc. exceeds a preset threshold. The method has obvious limitations that firstly, the description capability of the method on the complex and nonlinear internal electrochemical decay process of the battery is insufficient, the method is difficult to adapt to changeable actual working environments (such as different charge-discharge multiplying powers, environment temperatures and the like), secondly, the threshold value alarm is a passive response mechanism, the alarm can only be carried out when faults are about to occur or occur, and the true prediction and early warning cannot be realized, and furthermore, the method usually ignores rich state information contained in complete time sequence curves of voltage, current, temperature and the like in the charge-discharge process of the battery. Therefore, how to automatically extract deep features of battery aging degree from massive time series data provided by a Battery Management System (BMS) and accurately predict future life of the battery by using an advanced machine learning algorithm has become a technical difficulty and a research hotspot to be solved in the current battery management field. Disclosure of Invention In order to solve the technical problems that the traditional battery life prediction method mentioned in the background art is low in precision, massive time sequence data cannot be effectively utilized, early warning capability is poor and the like, the invention aims to provide a battery residual life prediction system and method based on hybrid deep learning, which are used for accurately predicting the state of health and the residual life of a battery. In a first aspect, the present invention provides a method for predicting remaining life of a battery based on hybrid deep learning, the method comprising: the method comprises the steps of acquiring multi-dimensional time sequence data of a battery, identifying different effective charging intervals from the multi-dimensional time sequence data of the battery, marking cycle times corresponding to each effective charging interval, acquiring a battery health state value corresponding to each effective charging interval, constructing a battery health state sample based on the battery health state value and the cycle times, and extracting charging curve characteristics from the multi-dimensional time sequence data based on the battery health state sample; The method comprises the steps of constructing a battery health state prediction model, wherein the battery health state prediction model comprises a feature extraction module, a trend analysis module and a full connection layer which are sequentially connected, the feature extraction module is used for extracting local features from charging curve features, the trend analysis module is used for analyzing the evolution rule of the local features along with time, and the full connection layer is used for mapping the output result of the trend analysis module to a final battery health state value; The method comprises the steps of training a battery health state prediction model by using a battery health state sample and charging curve characteristics, predicting a battery health state value corresponding to the latest cycle number of a tested battery by using the trained battery health state pre