Battery low false alarm rate fault diagnosis method

Abstract

The application relates to a battery low false alarm rate fault diagnosis method, which comprises core optimization of a semi-supervised density clustering method, multi-model training generation and fault battery diagnosis. First, the semi-supervised density clustering method is subjected to core optimization through hyperparameter optimization, kernel domain design and soft boundary definition. Then, normal battery data is divided into multiple categories according to temperature and vehicle state, and the semi-supervised density clustering method is used for training of each category to generate multiple differentiated models. Finally, real-time data is preprocessed, and corresponding pre-trained semi-supervised density clustering models are called according to the current temperature and vehicle state to identify battery faults. The method can realize battery low false alarm rate fault diagnosis, thereby protecting electric vehicle passengers from fire hazards.

Description

Technical Field

[0001] This invention belongs to the field of electrical engineering technology, and specifically relates to a method for diagnosing battery faults with low false alarm rates. Background Technology

[0002] Hundreds of thermal runaway incidents caused by unaddressed defects in electric vehicle batteries have hampered the further development of the electric vehicle industry. Battery defect detection based on abnormal external parameters is an effective way to reduce such thermal runaway incidents and protect electric vehicle consumers from fire hazards. However, the influence of temperature and electric vehicle status (i.e., charging and driving) on ​​battery characteristics complicates the development of such methods. Existing data-driven methods consistently and incorrectly classify normal batteries as faulty, leading to insufficient panic among electric vehicle occupants.

[0003] MinPts and Eps are two important hyperparameters that affect clustering results and need to be set in advance before clustering. However, most existing studies use a trial-and-error approach to continuously try different combinations of MinPts and Eps until the model accuracy is within the expected range. This method is highly dependent on subjective human settings and is difficult to apply to different working conditions in real vehicles. Summary of the Invention

[0004] In response to the aforementioned existing technologies, this invention analyzes electric vehicle status and ambient temperature data using data mining and machine learning methods to perform low-false-report fault diagnosis of the battery, thereby protecting electric vehicle occupants from fire hazards.

[0005] The present invention is based on a data-driven method, which is more in line with the actual driving conditions of electric vehicles, and the model trained with a large amount of data is also more accurate.

[0006] A battery fault diagnosis method with low false alarm rate includes core optimization of a semi-supervised density clustering method, multi-model generation, and fault battery diagnosis. The core optimization of the semi-supervised density clustering method is achieved through hyperparameter optimization, kernel domain design, and soft boundary definition. The multi-model generation involves dividing the system into multiple categories based on ambient temperature and vehicle status, and training each category using the semi-supervised density clustering method to generate multiple differentiated models. The fault battery diagnosis involves preprocessing real-time data and, based on the current temperature and vehicle status, calling the corresponding pre-trained semi-supervised density clustering model to identify battery faults.

[0007] The present invention provides a battery fault diagnosis method with low false alarm rate. Through the core optimization of the above-mentioned semi-supervised density clustering method, multi-model generation, and fault battery diagnosis, the method diagnoses battery faults and achieves battery fault diagnosis with low false alarm rate.

[0008] Preferably, the core optimization of the semi-supervised density clustering method is as follows: Voltage data for different battery cells and time steps are as follows: In the formula, n is the number of batteries, and k is the total time step of the data. Different time steps correspond to different times. Let be the voltage of battery j at time step t.

[0009] Construct a semi-supervised density clustering model, define the objective function, take the minimum value of Eps and the maximum value of MinPts; The objective function is: A multi-level L-shaped optimization method is designed to simultaneously optimize Eps and MinPts. The optimization starts with the minimum Eps and the maximum Minpts in the mesh, and the constraints are defined as follows: In the formula, Let be the voltage of battery j at time step t, k be the total time step of the data, and n be the number of batteries. Representing vectors The transpose of ; A voltage data matrix representing different battery cells and time steps.

[0010] Six polygonal kernel regions were designed: circular, square, equilateral triangle, regular pentagon, regular hexagon, and regular octagon. By covering voltage data with different distribution characteristics through regions of different shapes, the clustering adaptability is improved.

[0011] Define the soft boundary of Eps: Among them, and These are the minimum and maximum values ​​of the statistical noise in the training dataset, respectively.

[0012] Expand the Eps domain into three intervals: Within the range, Minpts are counted as 1 when calculating; Outside the range, it is counted as 0; within In this context, the equivalent sample size ES is calculated using the Gaussian cumulative function, with a value ranging from 0 to 1. The equivalent sample size ES is described as follows:

[0013] in and The mean and standard deviation of the statistical noise are given.

[0014] Preferably, the multi-model generation specifically includes: The ambient temperature was divided into 0~10℃, 10~20℃, 20~30℃, and 30~40℃. The vehicle status was divided into charging status and driving status. A large number of training samples with different MinPts and Eps were used. By using an optimized semi-supervised density clustering method, eight models with different temperatures and statuses were trained to generate multiple differentiated models.

[0015] Preferably, the faulty battery diagnosis specifically includes: During each real-time battery fault diagnosis, a sliding window is used to extract voltage data for different battery cells and time steps. : In the formula, Let be the voltage of battery j at time step t, w be the length of the sliding window, and n be the number of batteries. Principal component analysis of high-dimensional voltage matrices The dimensionality reduction process is as follows: Further zero averaging yields the matrix : in, Let be the zero-average voltage of battery j over time step t, and its calculation formula is: covariance matrix The calculation formula is: In the formula, Let represent the transpose of matrix D.

[0016] calculate eigenvalues and the corresponding feature vector , (t=1, 2, ..., w) and arrange them as follows: In the formula, Represents the eigenvectors The feature matrix obtained by arranging.

[0017] Select The first column a is , Representation matrix The transpose of the matrix. Calculate the dimensionality-reduced semi-supervised density clustering input matrix. :

[0018] The number of principal components, 'a', is determined using the cumulative percentage variance, and is described as follows: Where CPV is the cumulative percentage variance; It is the covariance matrix The i-th diagonal element. Select the number of principal components, 'a', to achieve a predetermined cumulative percentage variance.

[0019] All battery samples are labeled with three types: core points, boundary points, and noise points. Points with a neighborhood radius of Eps that have a number of sample points 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. Noise points are far away from all point clusters and are considered outliers that do not belong to any cluster category, i.e., batteries with abnormal voltage.

[0020] Based on the current vehicle status and temperature, the corresponding pre-trained semi-supervised density model is invoked, and the dimensionality-reduced voltage data is input into the model to identify faulty batteries through density clustering.

[0021] This invention proposes a data-driven method combining semi-supervised density clustering and multi-level optimization to achieve early and accurate diagnosis of battery faults. Its core advantage lies in significantly reducing the false alarm rate. Through innovative multi-core domain and soft-boundary design, it effectively resists noise interference in real-world vehicle environments, resulting in an extremely low false alarm rate for normal batteries and avoiding unnecessary maintenance interventions.

[0022] This invention possesses powerful adaptive capabilities for various operating conditions. By training specialized models for different temperatures and vehicle states, it overcomes the limitations of traditional fixed-threshold methods in complex application scenarios. The early warning conclusions are derived from real-world data and are more closely aligned with engineering applications. Compared to traditional methods, this invention can improve the safety warning time window, enabling real-time online warnings and promptly alerting personnel before a fault occurs, effectively preventing safety accidents. Furthermore, this method can continuously optimize the training model based on newly generated data, constantly improving prediction accuracy and reliability. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the multi-level L-shaped optimization method of the present invention; Figure 2 This is a schematic diagram illustrating different kernel domains as defined in this invention; Figure 3 This is a schematic diagram of the soft boundary of the present invention; Figure 4 This invention provides a battery low false alarm rate fault diagnosis algorithm. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0025] like Figures 1-4 As shown, the battery fault diagnosis method with low false alarm rate based on data-driven and machine learning includes three steps: core optimization of semi-supervised density clustering method, multi-model generation, and battery fault diagnosis. Step 1: Core optimization of semi-supervised density clustering method: The core optimization of semi-supervised density clustering method is carried out through hyperparameter optimization, kernel domain design and soft boundary definition.

[0026] Voltage data for different battery cells and time steps are as follows: In the formula, n is the number of batteries, and k is the total time step of the data. Different time steps correspond to different times. Let be the voltage of battery j at time step t.

[0027] Construct a semi-supervised density clustering model, define the objective function, take the minimum value of Eps and the maximum value of MinPts; The objective function is: like Figure 1 As shown, a multi-level L-shaped optimization method is designed to simultaneously optimize Eps and MinPts. The optimization starts with the minimum Eps and the maximum Minpts in the mesh, and the constraints are defined as follows: In the formula, Let be the voltage of battery j at time step t, k be the total time step of the data, and n be the number of batteries. Representing vectors The transpose of ; A voltage data matrix representing different battery cells and time steps.

[0028] The above-mentioned optimized semi-supervised density clustering model defines the objective function, designs constraints, and uses actual running data to determine MinPts and Eps. While ensuring accuracy, it improves the effectiveness of battery defect diagnosis. Through design constraints and training on normal battery data, it enhances the distinction between voltage inconsistencies caused by normal factors or defects.

[0029] like Figure 2 As shown, six polygonal kernel regions are designed: circular, square, equilateral triangle, regular pentagon, regular hexagon, and regular octagon. By covering voltage data with different distribution characteristics through regions of different shapes, the clustering adaptability is improved.

[0030] like Figure 3As shown, the Eps soft boundary is defined: Among them, and These are the minimum and maximum values ​​of the statistical noise in the training dataset, respectively.

[0031] Expand the Eps domain into three intervals: Within the range, Minpts are counted as 1 when calculating; Outside the range, it is counted as 0; within In this context, the equivalent sample size ES is calculated using the Gaussian cumulative function, with a value ranging from 0 to 1. The equivalent sample size ES is described as follows:

[0032] in and The mean and standard deviation of the statistical noise are given.

[0033] Step 2: Multi-model generation: Divide the vehicles into multiple categories based on ambient temperature and vehicle status, and train each category using a semi-supervised density clustering method to generate multiple differentiated models.

[0034] The ambient temperature was divided into 0~10℃, 10~20℃, 20~30℃, and 30~40℃. The vehicle status was divided into charging status and driving status. A large number of training samples with different MinPts and Eps were used. By using an optimized semi-supervised density clustering method, eight models with different temperatures and statuses were trained to generate multiple differentiated models.

[0035] Step 3: Battery fault diagnosis: Preprocess the real-time data, and based on the current temperature and vehicle status, call the corresponding pre-trained semi-supervised density clustering model to identify battery faults.

[0036] Principal component analysis was used to reduce the dimensionality of voltage data extracted from different battery cells and time steps using a sliding window. The dimensionality-reduced voltages were then input into a clustering model, and all samples were labeled with three types of tags: core points, boundary points, and noise points. Noise points represent batteries with abnormal temperatures.

[0037] During each real-time battery fault diagnosis, a sliding window is used to extract voltage data for different battery cells and time steps. : In the formula, Let be the voltage of battery j at time step t, w be the length of the sliding window, and n be the number of batteries. Principal component analysis is used to analyze the high-dimensional voltage matrix. The dimensionality reduction process is as follows:

[0038] Further zero averaging yields the matrix : in, Let be the zero-average voltage of battery j over time step t, and its calculation formula is: covariance matrix The calculation formula is: In the formula, Let represent the transpose of matrix D.

[0039] calculate eigenvalues and the corresponding feature vector And arrange them as follows: In the formula, Represents the eigenvectors The feature matrix obtained by arranging.

[0040] Select The first column a is , Representation matrix The transpose of the matrix. Calculate the dimensionality-reduced semi-supervised density clustering input matrix. :

[0041] The number of principal components, 'a', is determined using the cumulative percentage variance, and is described as follows: Where CPV is the cumulative percentage variance; It is the covariance matrix The i-th diagonal element. Select the number of principal components, 'a', to achieve a predetermined cumulative percentage variance.

[0042] Based on dimensionality reduction and multi-step battery voltage analysis, battery anomalies are accurately identified, thereby enabling battery fault diagnosis. For example... Figure 4 As shown, all battery samples are labeled with three types: core points, boundary points, and noise points. Points with a neighborhood radius of Eps that have a number of sample points 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 certain core point are called boundary points. Points that are neither core points nor boundary points are noise points. Among them, noise points are outliers. Noise points are far away from each point cluster and are considered outliers that do not belong to any cluster category, i.e., batteries with abnormal voltage.

[0043] This method can mark outliers during clustering to identify abnormal battery voltages. Based on the current vehicle status and temperature, the corresponding pre-trained semi-supervised density model is invoked. When the dimensionality-reduced voltage data is input into the model for clustering, each battery corresponds to one point. Core points and boundary points have many adjacent points within Eps, indicating good battery voltage consistency, and are judged as normal. Noise points have no adjacent points within Eps, indicating poor battery voltage consistency, and are judged as abnormal.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such 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 fault diagnosis method with low false alarm rate, comprising core optimization of semi-supervised density clustering method, multi-model generation, and faulty battery diagnosis, characterized in that: The core optimization of the semi-supervised density clustering method is achieved through hyperparameter optimization, kernel domain design, and soft boundary definition. The multi-model generation method divides the vehicle into multiple categories based on ambient temperature and vehicle status, and trains each category using a semi-supervised density clustering method to generate multiple differentiated models. The fault battery diagnosis process involves preprocessing real-time data and, based on the current temperature and vehicle status, calling the corresponding pre-trained semi-supervised density clustering model to identify battery faults.

2. The battery low false alarm rate fault diagnosis method according to claim 1, characterized in that, The core optimization of the semi-supervised density clustering method is as follows: Voltage data for different battery cells and time steps are as follows: In the formula, n represents the number of batteries, and k represents the total time step of the data. Different time steps correspond to different moments. Let be the voltage of battery j at time step t; Construct a semi-supervised density clustering model, define the objective function, take the minimum value of Eps and the maximum value of MinPts; The objective function is: A multi-level L-shaped optimization method is designed to simultaneously optimize Eps and MinPts. The optimization starts with the minimum Eps and the maximum Minpts in the mesh, and the constraints are defined as follows: In the formula, Let be the voltage of battery j at time step t, k be the total time step of the data, and n be the number of batteries. Representing vectors The transpose of ; A voltage data matrix representing different battery cells and time steps; Six polygonal kernel regions were designed: circular, square, equilateral triangle, regular pentagon, regular hexagon, and regular octagon. By covering voltage data with different distribution characteristics through regions of different shapes, the clustering adaptability was improved. Define the soft boundary of Eps: Among them, and These are the minimum and maximum values ​​of the statistical noise in the training dataset, respectively. Expand the Eps domain into three intervals: Within the range, Minpts are counted as 1 when calculating; Outside the range, it is counted as 0; within In the middle, the equivalent sample number ES is calculated according to the Gaussian cumulative function, with a value of 0~1; The equivalent sample size ES is described as follows: in and The mean and standard deviation of the statistical noise are given.

3. The battery low false alarm rate fault diagnosis method according to claim 2, characterized in that, The multi-model generation specifically refers to: The ambient temperature was divided into 0~10℃, 10~20℃, 20~30℃, and 30~40℃. The vehicle status was divided into charging status and driving status. A large number of training samples with different MinPts and Eps were used. By using an optimized semi-supervised density clustering method, eight models with different temperatures and statuses were trained to generate multiple differentiated models.

4. The battery low false alarm rate fault diagnosis method according to claim 3, characterized in that, The faulty battery diagnosis includes: During each real-time battery fault diagnosis, a sliding window is used to extract voltage data for different battery cells and time steps. : In the formula, Let be the voltage of battery j at time step t, w be the length of the sliding window, and n be the number of batteries. Principal component analysis of high-dimensional voltage matrices Perform dimensionality reduction processing. Further zero averaging yields the matrix : in, Let be the zero-average voltage of battery j over time step t, and its calculation formula is: covariance matrix The calculation formula is: In the formula, Denotes the transpose of matrix D; calculate eigenvalues and the corresponding feature vector And arrange them as follows: In the formula, Represents the eigenvectors The feature matrix obtained by arranging; Select The first column a is , Representation matrix The transpose of the matrix is ​​used to calculate the dimensionality-reduced semi-supervised density clustering input matrix. : The number of principal components, 'a', is determined using the cumulative percentage variance, and is described as follows: Where CPV is the cumulative percentage variance; It is the covariance matrix The i-th diagonal element. Select the number of principal components, 'a', to achieve a predetermined cumulative percentage variance.

5. The battery low false alarm rate fault diagnosis method according to claim 4, characterized in that, All battery 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. Noise points are far away from each point cluster and are considered outliers that do not belong to any cluster category. They are batteries with abnormal voltage.

6. The battery low false alarm rate fault diagnosis method according to claim 5, characterized in that, Based on the current vehicle status and temperature, the corresponding pre-trained semi-supervised density model is invoked. When the dimensionality-reduced voltage data is input into the model for clustering, each battery corresponds to one point. Core points and boundary points have many adjacent points within Eps, indicating good battery voltage consistency, and are judged as normal. Noise points have no adjacent points within Eps, indicating poor battery voltage consistency, and are judged as abnormal.

Images