Method for Detecting Abnormal Charging and Discharging of Electric Vehicles Based on Multivariate Gaussian Distribution Model

Abstract

The present invention relates to a method for detecting abnormal charging and discharging of electric vehicles based on a multivariate Gaussian distribution model, which includes the following steps: S1: Obtain the charging and discharging data of electric vehicles, and perform data preprocessing, bad data processing, data conversion, and data feature mining on the obtained charging and discharging data of electric vehicles; S2: Evaluate the processed data, and construct a quality evaluation index system and a quality evaluation model for the charging and discharging data of electric vehicles according to the characteristics of the charging and discharging data of electric vehicles; S3: Perform visual processing on the charging and discharging data to obtain abnormal detection categories and standard curves, and predict whether new data is abnormal by the determined abnormal detection model and standard curves. The present invention improves the efficiency and accuracy of charging and discharging fault detection, and improves the safety of using electric vehicles.

Description

Technical Field

[0001] The present invention relates to a method for detecting abnormal charging and discharging of electric vehicles based on a multivariate Gaussian distribution model, belonging to the technical field of electrical equipment and electrical engineering. Background Technique

[0002] As one of the seven major fields of new infrastructure, the rapid development direction of electric vehicles and supporting charging equipment is clear. With the increasingly dense charging and discharging service network, various types of faults and even fires during the charging and discharging process occur frequently, causing significant harm and losses to electric vehicle users and the charging and discharging operation and maintenance service network, and at the same time affecting consumers' consumption choices of electric vehicles. At the same time, the abnormal state diagnosis system for the charging and discharging process of electric vehicles is not yet perfect, resulting in the actual safety emergency effect of electric vehicle supporting facilities being difficult to meet expectations.

[0003] Most of the currently used electric vehicle charging piles adopt the method of regular manual maintenance. This method requires a large amount of manpower, material and financial resources, bringing a lot of trouble to the charging pile operators, and it is difficult to ensure the timeliness of fault discovery and troubleshooting. At present, relevant work has been done in China on the faults of electric vehicle charging and discharging, and a variety of strategies and models have been proposed, while there is less research on intelligent diagnosis methods for faults. The research on electric vehicle charging and discharging fault diagnosis in China needs to be further deepened. Summary of the Invention

[0004] The purpose of the present invention is to overcome the problems existing in the prior art and provide a method for detecting abnormal charging and discharging of electric vehicles based on a multivariate Gaussian distribution model, with reduced algorithm calculation amount and enhanced practicability, which can greatly improve the efficiency, accuracy and safety of electric vehicle charging and discharging fault detection.

[0005] In order to solve the above problems, the method for detecting abnormal charging and discharging of electric vehicles based on a multivariate Gaussian distribution model of the present invention includes the following steps:

[0006] S1: Obtain electric vehicle charging and discharging data, and perform data preprocessing, bad data processing, data conversion and data feature mining on the obtained electric vehicle charging and discharging data;

[0007] S2: Evaluate the processed data, and construct a quality evaluation index system and a quality evaluation model for electric vehicle charging and discharging data according to the characteristics of electric vehicle charging and discharging data;

[0008] S3: Perform visualization processing on the charging and discharging data to obtain abnormal detection categories and standard curves, and predict whether new data is abnormal by the determined abnormal detection model and standard curves.

[0009] Further, the data preprocessing in step S1 includes:

[0010] ①Using the tagging method, the attributes of the hybrid energy data are uniquely identified by using a unified standard tag;

[0011] ②Perform encapsulation processing on the data. Based on the unique identifier of the electric vehicle charging and discharging system data, combined with the data cleaning rules, encapsulate the electric vehicle charging and discharging system data and the specific data cleaning rules to form an electric vehicle charging and discharging system data encapsulation package associated with specific cleaning rules that can be recognized and controlled.

[0012] Furthermore, the attributes of the hybrid energy data include source, type, and structure.

[0013] Furthermore, the processing of bad data specifically includes the following steps:

[0014] 1) Train the BP neural network with the historical normal data in a period of time before the occurrence of bad data;

[0015] 2) Use the normal data at the moment before the occurrence of bad data to obtain a set of output data through the trained neural network;

[0016] 3) Use the output data obtained in step 2) as the data for correcting the bad data.

[0017] Furthermore, the bad data processing in step S1 includes a bad data identification algorithm based on GSA, and the specific steps are as follows:

[0018] a) Cluster the measurement data, where the number of clusters k = 1, 2, ··· K, and calculate the cluster dispersion W corresponding to different numbers of clusters k ;

[0019] b) Use the uniform distribution as the reference distribution. When clustering the reference data with k = 1, 2, ··· K, calculate the expectation E(W r,k ) of the cluster dispersion corresponding to different numbers of clusters;

[0020] c) Calculate the Gap(k) value corresponding to k = 1, 2, ··· K, and the calculation formula is Gap(k) = ln(E(W r,k )) - ln(W k );

[0021] d) Finally, determine the smallest k value that makes the inequality Gap(k) ≥ Gap(k + 1) hold. At this time, the k value is the optimal number of clusters.

[0022] Furthermore, in step S3, the visualization processing of the charging and discharging data to obtain the anomaly detection category and the standard curve specifically includes the following steps:

[0023] S3.1: Remove the incorrect data from all charge and discharge data according to the following conditions:

[0024] 1) I <= -500; U <= 100; Umax <= 0.5; Umin <= 0.5; Tmin <= 0; R < 0;

[0025] 2) Charge data: Discharge data:

[0026] SOC db1 (1) = 0; U max，db1 (1) = 0;

[0027] SOC db2 (1) = 0; SOC db2 (2) = 0; U max，db2 (1) = 0; U max，db2 (2) = 0; SOC db3 (1) = 0; SOC db3 (2) = 0; SOC db3 (3) = 0; U max，db3 (1) = 0; U max，db3 (2) = 0; U max，db3 (3) = 0;

[0028] SOCda1(1) = 0; U max，da1 (1) = 0;

[0029] SOCda2(1) = 0; SOCda2(2) = 0; U max，da2 (1) = 0; U max，da2 (2) = 0;

[0030] SOCda3(1) = 0; SOCda3(2) = 0; SOCda3(3) = 0;

[0031] Umax, da3(1) = 0; Umax, da3(2) = 0; Umax, da3(3) = 0;

[0032] Among them, SOC db1 (t) = SOC(t) - SOC(t - 1); Umax, db1 (t) = Umax, db1 (t) - Umax, db1 (t - 1) is the forward first-order difference; SOCdb2(t) = SOC(t) - SOC(t - 2); Umax, db2(t) = Umax, db2(t) - Umax, db2(t - 1) is the forward second-order difference; SOC db3 (t) = SOC(t) - SOC(t - 3); Umax，db3 U(t) = max，db3 U(t) - max，db3 U(t - 1) is the forward third-order difference; SOCda1(t) = SOC(t) - SOC(t + 1); max，da1 U(t) = max，da1 U(t) - max，da1 U(t - 1) is the backward first-order difference; SOCda2(t) = SOC(t) - SOC(t + 2); max，da2 U(t) = max，da2 U(t) - max，da2 U(t + 2) is the backward second-order difference; SOCda3(t) = SOC(t) - SOC(t + 3); Umax,da3(t) = Umax,da3(t) - Umax,da3(t + 3) is the backward third-order difference, where for any t, t >= 0.

[0033] S3.2: Visualize the retained data: Represent the obtained charge and discharge current, charge and discharge voltage, maximum voltage of a single cell, minimum voltage of a single cell, maximum temperature of a single cell, minimum temperature of a single cell, charge and discharge battery power, and charge and discharge insulation resistance data on the coordinate axes.

[0034] Furthermore, in step S3, predicting whether new data is abnormal by the determined anomaly detection model and standard curve includes the anomaly point detection method of multivariate Gaussian distribution, and the calculation formula is as follows:

[0035]

[0036]

[0037]

[0038] Among them, m is the number of samples; x is a set of feature data; n is the number of input features; μ is a vector, and each of its elements is the mean of a row of data in the original feature matrix; Σ represents the covariance matrix; calculate the value of p(x) and compare it with the predetermined threshold ε. If the calculated p(x) is less than ε, it is an abnormal point, otherwise it is a normal point.

[0039] Furthermore, the feature data includes the current amplitude I, the maximum voltage amplitude Umax, and the forward and backward differences of each order of the maximum voltage amplitude ΔU max，da1 、ΔU max，da2 、ΔU max，da3 、ΔU max，db1 、ΔU max，db2 、ΔU max，db3, the minimum voltage amplitude Umin, the differences of each order of the minimum voltage amplitude Umin, the maximum temperature amplitude Tmax, the minimum temperature amplitude Tmin, the equivalent resistance amplitude R, the battery charge amplitude SOC, the forward and backward differences of each order of the battery charge amplitude ΔSOC db1 、ΔSOC db2 、ΔSOC db3 、ΔSOC da1 、ΔSOC da2 、ΔSOCda3.

[0040] Furthermore, in step S1, the data conversion includes data generalization, data normalization, and data attribute construction.

[0041] Furthermore, in step S1, the objectives to be achieved in the data feature mining include the data feature analysis objective, the objective of initially obtaining a set of feasible mining models, the objective of adjusting the feasible mining models, the objective of evaluating the feasible mining models, and the objective of recommending the mining models;

[0042] The tasks performed in the data feature mining include paying attention to the environmental changes related to the objectives and performing corresponding operations according to the actual situation.

[0043] The beneficial effects of the present invention are as follows: (1) When dealing with bad data, through the improved GSA algorithm, the errors caused by the original reference data set and sample estimation are removed, the computational complexity of the algorithm is reduced, the calculation speed of the algorithm is increased, and the practicability of the algorithm is enhanced; (2) A quality evaluation index system and a quality evaluation model for electric vehicle charging and discharging data are constructed to comprehensively evaluate the data; (3) In the data feature mining model, the research core has changed from the previous research and mining process for a specific field to analyzing the objectives to be achieved by the entities in data mining and the tasks to be performed, simplifying the research content; (4) The outlier detection method based on multivariate Gaussian distribution is used, considering the relationships between various parameters, and can objectively and accurately judge outliers; (5) The related algorithms are simplified, the efficiency and accuracy of charging and discharging fault inspection are improved, and the safety of using electric vehicles is enhanced. Description of the Drawings

[0044] Figure 1 is a flowchart of the method for detecting anomalies in electric vehicle charging and discharging based on the multivariate Gaussian distribution model of the present invention. Detailed Embodiments

[0045] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0046] As Figure 1As shown in the figure, the electric vehicle charging and discharging anomaly detection method based on the multivariate Gaussian distribution model of the present invention includes the following steps:

[0047] The purpose of the present invention is to overcome the problems existing in the prior art and provide an electric vehicle charging and discharging anomaly detection method based on the multivariate Gaussian distribution model, with reduced algorithm calculation amount, enhanced practicability, and capable of greatly improving the efficiency, accuracy, and safety of electric vehicle charging and discharging fault detection.

[0048] To solve the above problems, the electric vehicle charging and discharging anomaly detection method based on the multivariate Gaussian distribution model of the present invention includes the following steps:

[0049] S1: Obtain electric vehicle charging and discharging data, and perform data preprocessing, bad data processing, data conversion, and data feature mining on the obtained electric vehicle charging and discharging data;

[0050] The data preprocessing in step S1 includes:

[0051] ① By the marking method, use a unified specification mark to uniquely identify the attributes of the hybrid energy data; the attributes of the hybrid energy data include source, type, and structure.

[0052] ② Perform encapsulation processing on the data. According to the unique identifier of the electric vehicle charging and discharging system data, combine the data cleaning rules, and encapsulate the electric vehicle charging and discharging system data and the specific data cleaning rules to form an electric vehicle charging and discharging system data encapsulation package associated with the specific cleaning rules that can be recognized and controlled.

[0053] The bad data processing in step S1 includes a bad data identification algorithm based on GSA, and the specific steps are as follows:

[0054] a) Cluster the measurement data, where the number of clusters k = 1, 2, ··· K, and calculate the clustering dispersion W corresponding to different numbers of clusters k ;

[0055] b) Use a uniform distribution as the reference distribution. When clustering the reference data with k = 1, 2, ··· K, calculate the expectation E(W r,k ) of the clustering dispersion corresponding to different numbers of clusters;

[0056] c) Calculate the Gap(k) value corresponding to k = 1, 2, ··· K, and the calculation formula is Gap(k) = ln(E(W r,k )) - ln(W k );

[0057] d) Finally, determine the smallest k value that makes the inequality Gap(k) ≥ Gap(k + 1) hold, and this k value is the optimal number of clusters.

[0058] The processing of bad data specifically includes the following steps:

[0059] 1) Training the BP neural network with the historical normal data in a period of time before the appearance of bad data; enabling the trained neural network to have the prediction ability;

[0060] 2) Using the normal data at the moment before the appearance of bad data to obtain a set of output data through the trained neural network;

[0061] 3) Taking the output data obtained in step 2) as the data for correcting the bad data.

[0062] Data conversion includes data generalization, data normalization and data attribute construction.

[0063] In step S1, the objectives to be achieved in data feature mining include the data feature analysis objective, the objective of initially obtaining a set of feasible mining models, the objective of adjusting the feasible mining models, the objective of evaluating the feasible mining models, and the objective of recommending mining models;

[0064] In step S1, the tasks performed in data feature mining include paying attention to the environmental changes related to the objectives and performing corresponding operations according to the actual situation.

[0065] S2: Evaluating the processed data, and constructing a quality evaluation index system and a quality evaluation model for electric vehicle charging and discharging data based on the characteristics of electric vehicle charging and discharging data;

[0066] S3: Performing visual processing on the charging and discharging data to obtain the anomaly detection categories and standard curves, and predicting whether new data is abnormal by the determined anomaly detection model and standard curves. The quality evaluation indicators for electric vehicle charging and discharging data include: the inherent quality, the accessible quality, the context quality, and the representation quality of electric vehicle charging and discharging data. The quality evaluation model for electric vehicle charging and discharging data includes: decision relationships, detection associations, business rules, algorithm libraries, and rule weights.

[0067] In step S3, performing visual processing on the charging and discharging data to obtain the anomaly detection categories and standard curves specifically includes the following steps:

[0068] S3.1: Removing the error data from all charging and discharging data according to the following conditions: In step S3.1, the unit of current I is defaulted to ampere; the unit of voltage U is defaulted to volt.

[0069] 1) I <= -500; U <= 100; Umax <= 0.5; Umin <= 0.5; Tmin <= 0; R < 0;

[0070] 2) Charging data: Discharging data:

[0071] SOC db1 U(1)=0; max，db1 (1)=0;

[0072] SOC db2 SOC(1)=0; db2 U(2)=0; max，db2 U(1)=0; maxdb2 (2)=0;

[0073] SOC db3 SOC(1)=0; db3 SOC(2)=0; db3 (3)=0;

[0074] U max，db3 U(1)=0; max，db3 U(2)=0; max，db3 (3)=0; [[ID=3)7]]

[0075] SOC da1 U(1)=0; max，da1 (1)=0;

[0076] SOC da2 SOC(1)=0; da2 U(2)=0; max，da2 U(1)=0; max，da2 (2)=0;

[0077] SOC da3 SOC(1)=0; da3 SOC(2)=0; da3 (3)=0;

[0078] U max，da3 U(1)=0; max，da3 U(2)=0; max，da3 (3)=0;

[0079] Among them, SOC db1 ΔSOC(t)=SOC(t)-SOC(t - 1); U max，db1 ΔU(t)=U max，db1 ΔU(t)-U max，db1 ΔU(t - 1) is the forward first-order difference; SOC db2 Δ²SOC(t)=SOC(t)-SOC(t - 2); U max，db2 Δ²U(t)=U max，db2 Δ²U(t)-U max，db2 Δ²U(t - 2) is the forward second-order difference; SOC db3 Δ³SOC(t)=SOC(t)-SOC(t - 3); U max，db3 Δ³U(t)=Umax，db3 (t)-U max，db3 (t - 3) is the forward third-order difference; SOC da1 (t)=SOC(t)-SOC(t + 1); U max，da1 (t)=U max，da1 (t)-U max，da1 (t - 1) is the backward first-order difference; SOC da2 (t)=SOC(t)-SOC(t + 2); U max，da2 (t)=U max，da2 (t)-U max，da2 (t + 2) is the backward second-order difference; SOC da3 (t)=SOC(t)-SOC(t + 3); U max，da3 (t)=U max，da3 (t)-U max，da3 (t + 3) is the backward third-order difference, where for any t, t >= 0.

[0080] S3.2: Visualize the retained data: Represent the obtained charge and discharge current, charge and discharge voltage, maximum voltage of single cell, minimum voltage of single cell, maximum temperature of single cell, minimum temperature of single cell, charge and discharge battery power, and charge and discharge insulation resistance data on the coordinate axes.

[0081] In step S3, predicting whether new data is abnormal by the determined anomaly detection model and standard curve includes the anomaly point detection method of multivariate Gaussian distribution, and the calculation formula is as follows:

[0082]

[0083]

[0084]

[0085] Among them, m is the number of samples; x is a set of characteristic data; the characteristic data includes current amplitude I, maximum voltage amplitude Umax, forward and backward differences of each order of the maximum voltage amplitude ΔU max，da1 、ΔU max，da2 、ΔU max，db3 、ΔU max，db1 、ΔU max，db2 、ΔU max，db3 , minimum voltage amplitude Umin, differences of each order of the minimum voltage amplitude Umin, maximum temperature amplitude Tmax, minimum temperature amplitude Tmin, equivalent resistance amplitude R, battery power amplitude SOC, forward and backward differences of each order of the battery power amplitude ΔSOC db1 、ΔSOC db2 、ΔSOC db3 、ΔSOC da1, ΔSOC da2 , ΔSOC da3 ; n is the number of input features; μ is a vector, each element of which is the mean of a row of data in the original feature matrix; Σ represents the covariance matrix; calculate the value of p(x) and compare it with a predetermined threshold ε. If the calculated p(x) is less than ε, it is an outlier; otherwise, it is a normal point.

[0086] Inspired by the ideal embodiments of the present invention described above, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. An abnormal charging and discharging inspection method for electric vehicles based on a multivariate Gaussian distribution model, characterized in that, It includes the following steps: S1: Obtain the charging and discharging data of electric vehicles, and perform data preprocessing, bad data processing, data conversion, and data feature mining on the obtained charging and discharging data of electric vehicles; The bad data processing described in step S1 includes a bad data identification algorithm based on GSA, and the specific steps are as follows: a) Cluster the measurement data. The number of clusters \(k = 1, 2,\cdots, K\), and calculate the cluster dispersion \(W\) for different numbers of clusters k ; b) When using the uniform distribution as the reference distribution and performing k = 1, 2, ··· K clustering on the reference data, calculate the expected value E(W r,k ) of the clustering dispersion corresponding to different numbers of clusters; c) Calculate the corresponding Gap(k) values when the number of clusters k = 1, 2, ··· K. The calculation formula is Gap(k) = ln(E(W r,k )) - ln(W k ); d) Finally, determine the smallest k value to make the inequality Gap(k)≥Gap(k + 1) hold. At this time, the k value is the optimal number of clusters; S2: Evaluate the processed data, and construct a quality evaluation index system and a quality evaluation model for the charging and discharging data of electric vehicles according to the characteristics of the charging and discharging data of electric vehicles; S3: Perform visual processing on the charging and discharging data to obtain abnormal detection categories and standard curves, and predict whether new data is abnormal by the determined abnormal detection model and standard curves; In step S3, the visual processing of the charging and discharging data to obtain abnormal detection categories and standard curves specifically includes the following steps: S3.1: For all charging and discharging data, remove the wrong data according to the following conditions: 1) I <= -500; U <= 100; Umax <= 0.5; Umin <= 0.5; Tmin <= 0; R < 0; 2) Charging data: Discharging data: SOC db1 U(1) = 0; max，db1 (1) = 0; SOC db2 (1) = 0; SOC db2 (2) = 0; U max，db2 (1) = 0; U max，db2 (2) = 0; SOC db3 (1) = 0; SOC db3 (2) = 0; SOC db3 (3) = 0; U max，db3 U(1) = 0; max，db3 U(2) = 0; max，db3 U(3) = 0; SOC da1 (1) = 0; U max，da1 (1) = 0; SOC da2 (1) = 0; SOC da2 (2) = 0; U max，da 2(1) = O; U max，da2 (2)＝0； SOC da3 (1) = 0; SOC da3 (2) = 0; SOC da3 (3) = 0; U max，da3 (1) = 0; U max，da3 (2) = 0; U max，da3 (3) = 0; Among them, SOC db1 (t) = SOC(t) - SOC(t - 1); U max，db1 (t) = U max，db1 (t) - U max，db1 (t - 1) is the forward first-order difference; SOC db2 (t) = SOC(t) - SOC(t - 2); U max，db2 (t) = U max，db2 (t) - U max，db2 (t - 2) is the forward second-order difference; SOC db3 (t) = SOC(t) - SOC(t - 3); U max，db3 (t) = U max，db3 (t) - U max，db3 (t - 3) is the forward third-order difference; SOC da1 (t) = SOC(t) - SOC(t + 1); U max，da1 (t) = U max，da1 (t) - U max，da1 (t - 1) is the backward first-order difference; SOC da2 (t) = SOC(t) - SOC(t + 2); U max，da2 (t) = U max，da2 (t) - U max，da2 (t + 2) is the backward second-order difference; SOC da3 (t) = SOC(t) - SOC(t + 3); U max，da3 (t) = U max，da3 (t) - U max，da3 (t + 3) is the backward third-order difference, where for any t, t >= 0; S3.2: Visualize the remaining data: Represent the data of the obtained charging and discharging current, charging and discharging voltage, maximum voltage of a single battery, minimum voltage of a single battery, maximum temperature of a single battery, minimum temperature of a single battery, charging and discharging battery power, and charging and discharging insulation resistance on the coordinate axis; In step S3, predicting whether new data is abnormal by the determined abnormal detection model and standard curves includes an abnormal point detection method for multivariate Gaussian distribution, and the calculation formula is as follows: Among them, m is the number of samples; x is a set of feature data; n is the number of input features; μ is a vector, and each of its elements is the mean of a row of data in the original feature matrix; ∑ represents the covariance matrix; calculate the value of p(x) and compare it with a predetermined threshold ε. If the calculated p(x) is less than ε, it is an abnormal point, otherwise it is a normal point.

2. The method for detecting abnormal charging and discharging of an electric vehicle based on a multivariate Gaussian distribution model according to claim 1, wherein The data preprocessing described in step S1 includes: ① Through the marking method, use a unified specification mark to uniquely identify the attributes of the hybrid energy data; ② Perform encapsulation processing on the data. According to the unique identifier of the charging and discharging system data of electric vehicles, combined with the data cleaning rules, encapsulate the charging and discharging system data of electric vehicles and the specific data cleaning rules to form a recognizable and controllable encapsulated package of the charging and discharging system data of electric vehicles associated with specific cleaning rules.

3. The method for detecting abnormal charging and discharging of an electric vehicle based on the multivariate Gaussian distribution model according to claim 2, wherein The attributes of the hybrid energy data include source, type, and structure.

4. The method for detecting abnormal charging and discharging of an electric vehicle based on a multivariate Gaussian distribution model according to claim 1, characterized in that, The bad data processing specifically includes the following steps: 1) Train the BP neural network with the historical normal data in a period of time before the bad data appears; 2) Use the normal data at the moment before the bad data appears to obtain a set of output data through the trained neural network; 3) Use the output data obtained in step 2) as the data for correcting the bad data. It is characterized in that according to the method for detecting abnormal charging and discharging of electric vehicles based on the multivariate Gaussian distribution model described in claim 1: The characteristic data includes the current amplitude I, the maximum voltage amplitude Umax, and the forward and backward differences ΔU of each order of the maximum voltage amplitude max，da1 , ΔU max，da2 , ΔU max，da3 , ΔU max，db1 , ΔU max，db2 , ΔU max，db3 , the minimum voltage amplitude Umin, the differences ΔU of each order of the minimum voltage amplitude Umin min，da1 , ΔU min，da2 , ΔU min，da3 , ΔU min，db1 , ΔU min，db2 , ΔU min，db3 The maximum temperature amplitude Tmax, the minimum temperature amplitude Tmin, the equivalent resistance amplitude R, the battery charge amplitude SOC, and the forward and backward differences ΔSOC of each order of the battery charge amplitude db1 , ΔSOC db2 , ΔSOC db3 , ΔSOC da1 , ΔSOC da2 , ΔSOC da3 .

6. The method for detecting abnormal charging and discharging of an electric vehicle based on a multivariate Gaussian distribution model according to claim 1, characterized in that In step S1, the data conversion includes data generalization, data normalization, and data attribute construction.

7. The method for detecting abnormal charging and discharging of an electric vehicle based on a multivariate Gaussian distribution model according to claim 1, wherein In step S1, the objectives to be achieved in the data feature mining include the data feature analysis objective, the objective of initially obtaining a set of feasible mining models, the feasible mining model adjustment objective, the feasible mining model evaluation objective, and the mining model recommendation objective. The tasks performed in the data feature mining include paying attention to environmental changes related to the objective and performing corresponding operations according to the actual situation.

Images