A method for early warning of battery over-temperature

Abstract

This invention relates to a data-driven battery over-temperature early warning method, which includes two parts: battery temperature prediction and temperature anomaly identification. Battery temperature prediction is achieved by constructing a multi-step, multi-battery temperature synchronous prediction model and defining a loss function to simultaneously optimize prediction accuracy and computation time, thereby balancing the accuracy and timeliness of thermal runaway prediction. Temperature anomaly identification includes identifying abnormal battery temperatures. Based on the predicted battery temperatures, a semi-supervised density space clustering model is constructed to identify abnormal battery temperatures. Through the above two steps of temperature prediction and anomaly identification, abnormal battery temperatures are identified, enabling battery over-temperature early warning and reducing the loss of life and property of vehicle occupants.

Description

Technical Field

[0001] This invention belongs to the field of electrical engineering technology, specifically relating to a battery over-temperature early warning method. Background Technology

[0002] To meet the power and energy demands of vehicles and energy storage power stations, hundreds of individual battery cells are connected in series and parallel to form battery packs. When a single battery overheats and catches fire, the flames can rapidly spread throughout the entire battery pack, threatening the safety of people and property. Utilizing the sparse and limited battery temperature data collected by onboard sensors to achieve timely overheat warnings is a current research focus. Existing methods do not provide early safety warnings before battery overheating, and the warning time is insufficient. Summary of the Invention

[0003] In view of the current state of the prior art, the technical problem to be solved by this invention is to analyze battery data during the charging and discharging process of vehicle batteries through data mining and machine learning methods, to provide early warning of battery overheating, and to reduce the occurrence of safety accidents. The data-driven approach is more consistent with the actual driving and charging conditions of vehicles, and the model trained with a large amount of data also has higher accuracy.

[0004] The invention is summarized as follows:

[0005] The battery over-temperature early warning method includes two steps: battery temperature prediction and temperature anomaly identification. The battery temperature prediction involves using a model to predict the battery temperature. By constructing a multi-step, multi-battery temperature synchronous prediction model, a loss function is defined to synchronously optimize the prediction accuracy and computation time, thereby balancing the accuracy and timeliness of thermal runaway prediction. The temperature anomaly identification involves identifying abnormal battery temperatures. Based on the predicted battery temperatures, a semi-supervised density space clustering model is constructed to identify abnormal battery temperatures. Through the above two steps of battery temperature prediction and temperature anomaly identification, a battery over-temperature early warning is achieved.

[0006] Furthermore, the battery temperature prediction constructs a multi-step, multi-battery temperature synchronous prediction model, specifically as follows:

[0007] In each prediction, a sliding window is used to extract historical temperature time series and temperature-related feature parameters, resulting in matrix A. (k+n)×m As input matrix:

[0008]

[0009] In the formula, n is the number of temperature-related characteristic parameters; m is the number of temperature cells; k is the time step of historical data, with different time steps corresponding to different moments; T t,c (t = 1, 2, ..., k; c = 1, 2, ..., m) represents the temperature of battery c at time step t; F k,j(j=1,2,…,n) represents the j-th feature at time step k.

[0010] The predicted temperature for each battery in the next p steps is:

[0011]

[0012] In the formula, p is the prediction step size.

[0013] When predicting battery temperature synchronously, the temperature prediction error is gradually reduced by iteratively adding a weak predictor, thereby improving the prediction accuracy.

[0014] Input matrix A (k+n)×m With output matrix B p×m The previous relationship can be described as follows:

[0015]

[0016] In the formula, f κ Let K be the κ-th tree, and K be the number of trees.

[0017] The loss function L is:

[0018]

[0019] Where l is a differentiable convex loss function describing the predicted temperature at time t. With actual temperature T t The error between them; M(f) is the penalty term for model complexity; a is the number of trees.

[0020]

[0021] In the formula, Y is the number of leaf nodes; γ is the leaf penalty coefficient; α and λ represent the weights of L1 and L2 regularization; ω q It refers to the number of leaves. The mean square error between the predicted temperature and the actual temperature.

[0022] Furthermore, the temperature anomaly identification specifically involves: labeling all samples with three types of tags: core points, boundary points, and noise points; points with a number of sample points within a neighborhood radius Eps greater than or equal to the minimum neighborhood point MinPts are called core points; points that are not core points but are within the neighborhood of a certain core point are called boundary points; points that are neither core points nor boundary points are noise points; among them, noise points are anomaly points.

[0023] Construct a semi-supervised density clustering model, define an objective function, and use the MinPts and Eps that minimize the objective function as the MinPts and Eps set for the model.

[0024] The objective function OG is:

[0025]

[0026] In the formula, l abm The number of samples for which the model classifies abnormal temperatures as noise points; total The total number of samples in the training dataset; norm It is the number of samples in which the model identifies normal temperature as a core point or a boundary point.

[0027] Compared with existing technologies, the advantages of this invention are as follows: This invention combines big data mining and machine learning methods, using a data-driven approach to determine model parameters, thereby providing early warning of battery overheating. Its conclusions are derived from real data and applied to actual vehicles, resulting in better accuracy and realism, and closer alignment with engineering applications. Compared with traditional methods, it can improve the battery safety warning time window, enabling online early warning and alerting personnel before overheating faults occur, preventing safety accidents. Furthermore, it can continuously optimize and train the model based on newly generated data from the vehicle, improving prediction accuracy. Attached Figure Description

[0028] Figure 1 This invention relates to a temperature prediction method;

[0029] Figure 2 This invention relates to a method for identifying temperature anomalies. Detailed Implementation

[0030] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0031] like Figures 1-2 As shown, the data-driven battery over-temperature early warning method includes three steps: data preprocessing, temperature prediction, and temperature anomaly identification.

[0032] Step 1: Data Preprocessing: Extract parameters related to battery over-temperature measured by sensors, including battery temperature, battery voltage, battery state of charge, and battery current; process missing and outlier values, and smooth noise; integrate data by merging data from multiple data sources.

[0033] Step 2: Battery temperature prediction: Historical temperature time series and temperature-related feature parameters are extracted using a sliding window and input into the temperature prediction model. The temperature prediction model outputs multi-step temperature prediction results.

[0034] A sliding window is used to extract historical temperature time series and temperature-related feature parameters for each prediction:

[0035]

[0036] In the formula, n is the number of temperature-related characteristic parameters; m is the number of temperature cells; k is the time step of historical data, with different time steps corresponding to different moments; T t,c (t = 1, 2, ..., k; c = 1, 2, ..., m) represents the temperature of battery c at time step t; F k,j (j=1,2,…,n) represents the j-th feature at time step k.

[0037] The predicted temperature for each battery in the next p steps is:

[0038]

[0039] In the formula, p is the prediction step size.

[0040] Basic methods for synchronous prediction of battery temperature, such as Figure 1 As shown, by iteratively adding a weak predictor, the temperature prediction error is gradually reduced, thereby improving the prediction accuracy.

[0041] Input matrix A (k+n)×m With output matrix B p×m The relationship between them can be described as follows:

[0042]

[0043] In the formula, f κ Let K be the κ-th tree, and K be the number of trees.

[0044] The loss function is:

[0045]

[0046] Where l is a differentiable convex loss function describing the predicted temperature at time t. With actual temperature T t The error between them; M(f) is the penalty term for model complexity; a is the number of trees.

[0047] The training objective of the model is to reduce the loss function. According to Equation 4, the simpler the model, the less computation time, and the smaller M(f) and the loss function L; the smaller the model error, the smaller l and the loss function L. Therefore, the model balances model accuracy and computation time, making it suitable for the goal of accurate and timely battery temperature prediction in this invention.

[0048]

[0049]

[0050] In the formula, Y is the number of leaf nodes; γ is the leaf penalty coefficient; α and λ represent the weights of L1 and L2 regularization; ω q It refers to the number of leaves. The mean square error between the predicted temperature and the actual temperature.

[0051] Step 3: Temperature Anomaly Identification: Input the predicted temperature into the clustering model and label all samples with three types of tags: core points, boundary points, and noise points. Points with a neighborhood radius Eps equal to or greater than the minimum neighbor point MinPts are called core points; points not belonging to core points but within the neighborhood of a core point are called boundary points; points that are neither core points nor boundary points are noise points; among these, noise points are anomalies. Construct a semi-supervised density clustering model, define an objective function, and use the minimum values ​​of MinPts and Eps for the objective function as the model's set MinPts and Eps.

[0052] Based on predicted multi-step battery temperatures, abnormal battery temperatures are accurately identified, thus enabling battery over-temperature early warning. The process is as follows: Figure 2 As shown, all samples are labeled with three types: core points, boundary points, and noise points. Points with a neighborhood radius Eps greater than or equal to the minimum neighbor MinPts are called core points; points not belonging to core points but within the neighborhood of a core point are called boundary points; and points that are neither core points nor boundary points are noise points. Noise points are far from any cluster and are considered outliers that do not belong to any cluster category. In summary, this method can label outliers during the clustering process, thereby identifying abnormal battery temperatures. For B... p×m When performing clustering, each battery corresponds to one point. Core points and boundary points have many adjacent points within Eps, indicating good battery temperature consistency, and are judged as normal. Noise points have no adjacent points within Eps, indicating poor battery temperature consistency, and are judged as abnormal.

[0053] Depend on Figure 2 As can be seen from the process, MinPts and Eps are two important hyperparameters affecting the clustering results and need to be set in advance before clustering. However, most existing studies use a trial-and-error method to continuously try different combinations of MinPts and Eps until the model accuracy is within the expected range. Its accuracy is heavily dependent on subjective human settings and is difficult to apply to different working conditions of real vehicles. To address the above problems, this invention constructs a semi-supervised density clustering model, defines an objective function, and uses actual operating data to determine MinPts and Eps.

[0054] According to the requirements for abnormal temperature identification of the present invention, the objective function should consider the following aspects: (1) The model should be able to accurately identify abnormal battery temperatures and thus predict battery thermal runaway in a timely manner. (2) The model should avoid generating false alarms for normal batteries to prevent unnecessary panic among occupants. Therefore, the objective function is defined as:

[0055]

[0056] In the formula, l abmThe number of samples for which the model classifies abnormal temperatures as noise points; total The total number of samples in the training dataset; norm It is the number of samples in which the model identifies normal temperature as a core point or a boundary point.

[0057] The MinPts and Eps at which the objective function is minimized are used as the MinPts and Eps of the model.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A battery over-temperature early warning method, comprising two steps: battery temperature prediction and temperature anomaly identification, characterized in that, The battery temperature prediction includes using a model to predict the battery temperature. By constructing a multi-step, multi-battery temperature synchronous prediction model, a loss function is defined to synchronously optimize the prediction accuracy and computation time, thereby balancing the accuracy and timeliness of thermal runaway prediction. The temperature anomaly identification includes identifying abnormal battery temperatures. Based on the predicted battery temperatures, a semi-supervised density space clustering model is constructed to identify abnormal battery temperatures. Through the above two steps of battery temperature prediction and temperature anomaly identification, a battery over-temperature early warning is achieved. The battery temperature prediction constructs a multi-step, multi-battery temperature synchronous prediction model, specifically as follows: In each prediction, a sliding window is used to extract historical temperature time series and temperature-related feature parameters, resulting in matrix A. (k+n)×m As input matrix: In the formula, n is the number of temperature-related characteristic parameters; m is the number of temperature cells; k is the time step of historical data, with different time steps corresponding to different moments; T t,c (t = 1, 2, ..., k; c = 1, 2, ..., m) represents the temperature of battery c at time step t; F k,j (j = 1, 2, ..., n) represents the j-th feature at time step k; The predicted temperature for each battery in the next p steps is: In the formula, p is the prediction step size; When predicting battery temperature synchronously, the temperature prediction error is gradually reduced by iteratively adding a weak predictor, thereby improving the prediction accuracy. Input matrix A (k+n)×m With output matrix B p×m The relationship between them is: In the formula, f κ Let K be the κ-th tree, and K be the number of trees; The loss function L is: Where l is a differentiable convex loss function describing the predicted temperature at time t. With actual temperature T t The error between them; M(f) is the penalty term for model complexity; a is the number of trees; In the formula, Y is the number of leaf nodes; γ is the leaf penalty coefficient; α and λ represent the weights of L1 and L2 regularization; ω q It is the number of leaves; The mean square error between the predicted temperature and the actual temperature; The temperature anomaly identification specifically involves: All samples are labeled with three types: core points, boundary points, and noise points. Points whose number of sample points within a neighborhood radius Eps is greater than or equal to the minimum neighborhood point MinPts are called core points. Points that are not core points but are in the neighborhood of a core point are called boundary points. Points that are neither core points nor boundary points are noise points. Among them, noise points are outliers. Construct a semi-supervised density clustering model, define an objective function, and use the MinPts and Eps that minimize the objective function as the MinPts and Eps set for the model. The objective function OG is: In the formula, l abm The number of samples for which the model classifies abnormal temperatures as noise points; total The total number of samples in the training dataset; norm It is the number of samples in which the model identifies normal temperature as a core point or a boundary point.

Images