Battery health prediction method based on local energy characteristics of incomplete charging process

Abstract

The application relates to the technical field of battery management, and discloses a battery health prediction method based on local energy characteristics of an incomplete charging process, which comprises the following steps: collecting charging data of an electric vehicle battery, wherein the charging data comprises constant-voltage charging data and constant-current charging data, and aging characteristics of the constant-voltage charging data and the constant-current charging data are extracted respectively; if the charging data comprises the constant-voltage charging data and condition A is met, then the aging characteristics of the constant-voltage charging data are taken as input, the battery health degree is taken as output, and a model one is established by using a machine learning algorithm; if the charging data only comprises the constant-current charging data and condition B is met, or the charging data comprises the constant-voltage charging data but condition A is not met, then the aging characteristics of the constant-current charging data are taken as input, the battery health degree is taken as output, and a model two is established by using the machine learning algorithm; and the charging data of the electric vehicle is input into the model one or the model two to obtain a health state prediction value.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically to a battery health prediction method based on the local energy characteristics of an incomplete charging process. Background Technology

[0002] Lithium-ion batteries are characterized by long lifespan, low self-discharge rate, and high energy density, making them excellent energy storage devices. This has led to their widespread use in electrical equipment, such as pure electric vehicles, plug-in hybrid electric vehicles, mobile energy storage devices, and power grids.

[0003] Currently, methods for estimating the state of health (SQH) of electric vehicle (EV) batteries can be broadly categorized into two types: model-based methods and data-driven methods. Model-based methods require prior knowledge to construct suitable models that can reflect the battery's electrochemical characteristics to a certain extent, such as electrochemical models and equivalent circuit models. They then utilize recursive algorithms like Kalman filtering and particle filtering, estimating the SQH based on real-time voltage and current data. Data-driven methods, on the other hand, analyze extensive battery aging experimental data, extract features related to battery aging, and then use machine learning tools to describe the correlation between these features and the SQH.

[0004] Currently, there are methods for estimating battery health by extracting features from the capacity increment curve during the charging phase and establishing a mapping relationship between the curve peak features and the battery's health state. However, these methods largely fail to consider that the capacity increment curve may be incomplete under actual charging conditions, and they also largely ignore the impact of temperature on the curve peak. Since battery charge and discharge capacity is easily affected by temperature, the capacity increment curve may also be affected by temperature. Additionally, there are methods that use charge and discharge segment data to extract aging features and estimate health state, but these methods often require extremely large training data segments for application in the uncertain real-world charge and discharge processes.

[0005] Because the charging and discharging behavior of electric vehicles during actual driving is highly uncertain, there are very few instances of complete discharge or full charge, causing many associated aging characteristics to become invalid. Therefore, how to utilize incomplete charging and discharging data for feature extraction and to find aging characteristics with anti-interference capabilities is a common problem in current data-driven methods. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a battery health prediction method based on local energy characteristics of an incomplete charging process.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A battery health prediction method based on local energy characteristics of an incomplete charging process includes the following steps:

[0009] The system collects charging data from electric vehicle batteries, including constant voltage charging data and constant current charging data, and extracts aging characteristics from the constant voltage charging data and constant current charging data respectively. Based on the differences in charging voltage curves under different aging states, the charging voltage in the constant current charging stage can be divided into characteristic voltage ranges.

[0010] If the charging data includes constant voltage charging data and satisfies condition A, then the aging characteristics of the constant voltage charging data are used as input, the battery health is used as output, and a health status prediction model is established using machine learning algorithms.

[0011] If the charging data only contains constant current charging data and meets condition B, or if the charging data contains constant voltage charging data but does not meet condition A, then the aging characteristics of the constant current charging data will be used as input, the battery health status will be used as output, and a health status prediction model II will be established using machine learning algorithms.

[0012] Condition A is: the charging cutoff current during constant voltage charging is less than or equal to x times the charging current during constant current charging, where x is between 0.05 and 1; Condition B is: the constant current charging data contains at least one complete characteristic voltage range.

[0013] The charging data of the electric vehicle is input into either Health Status Prediction Model 1 or Health Status Prediction Model 2 to obtain the predicted health status value.

[0014] The closer x is to 0.05, the more complete the battery process is. To enhance practicality, x should be a larger value while still meeting the requirement of extracting aging feature data. In this invention, x = 0.5 is chosen.

[0015] Specifically, the aging characteristics of constant voltage charging data include energy increment, charging temperature, and charging rate during the constant voltage charging phase.

[0016] Specifically, based on the differences in charging voltage curves under different aging states, the charging voltage of the constant current charging stage is divided into multiple characteristic voltage intervals. Within each characteristic voltage interval, the area enclosed by the charging voltage curves corresponding to different aging states and the coordinate axes is different. The aging characteristics of the constant current charging data include the energy increment, charging temperature, and charging rate within each characteristic voltage interval of the constant current charging stage as aging characteristics.

[0017] Specifically, if the charging data does not include either constant voltage charging data or constant current charging data, then the previous health status prediction value will be used as the current health status prediction value.

[0018] Compared with the prior art, the beneficial technical effects of the present invention are:

[0019] To reduce the training complexity of the health status prediction model while ensuring the algorithm's practicality and accuracy, this invention considers the statistical characteristics of actual electric vehicle charging behavior. It divides all possible charging processes, extracts aging features for each process, and sets priorities for the health prediction model, allowing for adjustments to aging features and estimation methods for different charging conditions. Furthermore, this invention can predict health status using data from any charging start and end point, an effect not found in most current health prediction methods, which is precisely its key feature. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the battery health prediction method of the present invention. Detailed Implementation

[0021] A preferred embodiment of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] This invention presents a method for extracting battery aging characteristics. According to a survey conducted by FAW Group on the charging and discharging behavior of its electric vehicles, approximately 60% of users charge to a state of charge (SOC) of 95%–100%. The SOC distribution at the start of charging is similar across different regions, primarily ranging from 20% to 80%, with the largest proportion at 35%–45%. This indicates that users typically prefer to fully charge their vehicles in one go to reduce range anxiety. However, the SOC of a battery fluctuates significantly before charging, and the battery's discharge behavior is highly random, making feature extraction difficult.

[0023] Considering that lithium batteries typically require complete constant current charging and partial constant voltage charging to reach a state of charge of over 95%, this invention divides the battery charging process into stages and prioritizes them: when the charging process includes charging data from the constant voltage stage, the aging characteristics of the constant voltage charging data are selected as the input to the prediction model; when only charging data from the constant current stage exists, the aging characteristics of the constant current charging data are selected as the input to the prediction model. The specific scheme is as follows.

[0024] like Figure 1 As shown, a battery health prediction method based on local energy characteristics of an incomplete charging process includes the following steps:

[0025] (1) Collect charging data of electric vehicle batteries. The charging data includes constant voltage charging data and constant current charging data. Extract the aging characteristics of constant voltage charging data and constant current charging data respectively.

[0026] (2) If the charging data includes constant voltage charging data and satisfies condition A, then the aging characteristics of the constant voltage charging data are used as input, the battery health is used as output, and a health status prediction model is established using machine learning algorithms.

[0027] (3) If the charging data only contains constant current charging data and meets condition B, or if the charging data contains constant voltage charging data but does not meet condition A, then the aging characteristics of the constant current charging data are used as input, the battery health is used as output, and a health status prediction model II is established using machine learning algorithms.

[0028] Condition A is: the charging cutoff current during constant voltage charging is less than or equal to x times the charging current during constant current charging, where x is between 0.05 and 1. Condition B is: the constant current charging data contains at least one complete characteristic voltage range. The closer x is to 0.05, the more complete the battery process. To enhance practicality, x should be a larger value while still meeting the requirement for extracting sufficient aging characteristic data. In this invention, x = 0.5. If the charging process includes a constant voltage charging process and satisfies condition A, the constant voltage charging data is sufficient; if the charging process includes a constant voltage charging process but does not satisfy condition A, the constant voltage charging data is insufficient.

[0029] (4) Input the charging data of the electric vehicle into either Health Status Prediction Model 1 or Health Status Prediction Model 2 to obtain the predicted health status value.

[0030] (5) If the charging data does not include constant voltage charging data or constant current charging data, the previous health status prediction value shall be used as the current health status prediction value.

[0031] Figure 1 In dashed box I, the battery health prediction method selects the charging energy, charging temperature, and charging rate during the constant voltage charging stage as the input features of health status prediction model one.

[0032] Based on the differences in charging voltage curves under different aging states, the charging voltage during the constant current charging stage is divided into multiple characteristic voltage ranges. Within each characteristic voltage range, the area enclosed by the charging voltage curves corresponding to different aging states and the coordinate axes is different.

[0033] Figure 1 In dashed box II, the battery health prediction method selects the energy increment, charging temperature and charging rate of all characteristic voltage ranges included in the constant current charging stage as the input features of health state prediction model II.

[0034] Figure 1In dashed box III, the battery health prediction method directly reads the previous health status prediction value as the current health status prediction value, indicating that the charging process is too short and the aging characteristics contained in the charging data are insufficient for health status prediction.

[0035] In the health prediction method, the priority of the three prediction schemes mentioned above decreases sequentially, and all possible charging conditions are considered. Then, based on the conditions met by the current charging process, the prediction is made according to... Figure 1 The illustrated process utilizes the highest-level method. This approach enables the extraction of aging features from incomplete charging voltage, current, and temperature data of the power battery in a more realistic charging and discharging scenario. It analyzes the correlation between local charging energy increments and battery aging, thereby predicting the battery's current health status. Clearly, this method is suitable for online estimation of battery health status under arbitrary charging start and end points, and can adjust aging features and estimation methods for different charging conditions.

[0036] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0037] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A battery health prediction method based on partial energy characteristics of incomplete charging process, comprising the following steps: Collecting charging data of an electric vehicle battery, the charging data including constant-voltage charging data and constant-current charging data, and extracting aging characteristics of the constant-voltage charging data and the constant-current charging data respectively; According to the difference of charging voltage curves under different aging states, the charging voltage of the constant-current charging stage can be divided into characteristic voltage intervals; If the charging data contains constant-voltage charging data and meets condition A, the aging characteristics of the constant-voltage charging data are taken as input, the battery health degree is taken as output, and a health state prediction model one is established by using a machine learning algorithm; If the charging data only contains constant-current charging data and meets condition B, or the charging data contains constant-voltage charging data but does not meet condition A, the aging characteristics of the constant-current charging data are taken as input, the battery health degree is taken as output, and a health state prediction model two is established by using a machine learning algorithm; Wherein condition A is that the charging cutoff current in the constant-voltage charging process is less than or equal to x times of the charging current in the constant-current charging process, and x is between 0.05 and 1; Condition B is that the constant-current charging data contains at least one complete characteristic voltage interval; The charging data of the electric vehicle is input into the health state prediction model one or the health state prediction model two to obtain a health state prediction value; The aging characteristics of the constant-voltage charging data include energy increment, charging temperature and charging rate in the constant-voltage charging stage; The aging characteristics of the constant-current charging data include energy increment, charging temperature and charging rate in each characteristic voltage interval in the constant-current charging stage.

2. The battery health prediction method based on partial energy signature of incomplete charging process according to claim 1, characterized in that: According to the difference of charging voltage curves under different aging states, the charging voltage of the constant-current charging stage can be divided into multiple characteristic voltage intervals, and in each characteristic voltage interval, the charging voltage curves corresponding to different aging states are different from the area enclosed by the coordinate axes.

3. The battery health prediction method based on partial energy signature of incomplete charging process according to claim 1, characterized in that: If the charging data neither contains constant-voltage charging data nor contains constant-current charging data, the last health state prediction value is taken as the health state prediction value this time.

Images