Search

CN-120886687-B - Charging abnormality detection method and device for electric automobile and charging pile

CN120886687BCN 120886687 BCN120886687 BCN 120886687BCN-120886687-B

Abstract

The invention relates to a method and a device for detecting abnormal charging of an electric automobile and a charging pile. The method for detecting the charging abnormality of the electric automobile comprises the steps of preprocessing charging voltage data and charging current data to generate charging time domain information and charging frequency domain information, extracting time domain features of the charging time domain information, extracting frequency spectrum features of the charging frequency domain information to obtain a charging time domain feature vector and a charging frequency domain feature vector, splicing the charging time domain feature vector and the charging frequency domain feature vector to form a spliced charging feature vector, fusing attention features of the spliced feature vector according to dynamic environment factors to generate a fused charging feature vector, and carrying out abnormality identification on the fused charging feature vector to obtain an abnormality detection result corresponding to a current charging period. The method for detecting the charging abnormality of the electric automobile effectively reduces the false detection rate of the charging abnormality detection and improves the robustness and generalization capability of the abnormality detection under various environmental conditions.

Inventors

  • PAN ZHIJIE

Assignees

  • 广东拓杰机电科技有限公司

Dates

Publication Date
20260505
Application Date
20250806

Claims (9)

  1. 1. The method for detecting the abnormal charging of the electric automobile is characterized by comprising the following steps of: S1, preprocessing charging voltage data and charging current data to generate charging time domain information and charging frequency domain information; S2A, performing time domain feature extraction on the charging time domain information to form a charging time domain feature vector; S2B, extracting spectral features of the charging frequency domain information to form a charging spectral feature vector; S3, splicing the charging time domain feature vector and the charging frequency spectrum feature vector to obtain a spliced charging feature vector; S4, performing attention feature fusion on the spliced charging feature vectors according to the dynamic environment factors to generate fused charging feature vectors; S5, carrying out anomaly identification on the fused charging feature vector to generate an anomaly detection result corresponding to the current charging period; The step S2A comprises the following sub-steps: carrying out one-dimensional convolution operation on the voltage time domain information in the charging time domain information to form a local characteristic sequence of the voltage; Vector encoding is carried out on the local characteristic sequence of the voltage to form an embedded vector sequence with fixed dimension, so as to obtain an embedded vector of the voltage; carrying out one-dimensional convolution operation on the current time domain information in the charging time domain information to form a local characteristic sequence of the current; Vector encoding is carried out on the local characteristic sequence of the current, so that an embedded vector of the current is obtained; taking the voltage embedded vector as a query vector, taking the current embedded vector as a key value vector to generate weight, and multiplying and fusing the weight and the current embedded vector to obtain an enhanced current embedded vector; Meanwhile, taking the current embedded vector as a query vector, taking the voltage embedded vector as a key value vector, generating weight, and multiplying and fusing the weight and the voltage embedded vector to obtain an enhanced voltage embedded vector; Splicing and full-connection mapping are sequentially carried out on the enhanced voltage embedded vector and the enhanced current embedded vector, and a primarily fused charging time domain feature vector is obtained; and carrying out bidirectional time sequence feature extraction on the preliminarily fused charging time domain feature vector to extract the context feature representation reflected in the voltage and current combined evolution process so as to form a final charging time domain feature vector.
  2. 2. The method for detecting abnormal charge of an electric vehicle according to claim 1, wherein the step S2B includes the sub-steps of: Splicing the voltage spectrum information and the current spectrum information in the charging frequency domain information in the channel dimension to form a two-dimensional spectrum tensor; performing depth separable convolution feature extraction on the two-dimensional spectrum tensor to obtain a two-dimensional spectrum feature vector; carrying out point-by-point convolution on the two-dimensional spectrum feature vector to obtain compressed two-dimensional spectrum features; and performing feature mapping and nonlinear transformation on the compressed spectrum features to generate a spectrum embedded vector with fixed dimensions, and obtaining a charged spectrum feature vector.
  3. 3. The method for detecting abnormal charging of an electric vehicle according to claim 2, wherein the dynamic environmental factor is generated by a lightweight neural network, and the specific structure of the lightweight neural network is as follows: In the formula, Representing a dynamic environmental factor; Input representing environmental data; the multi-layer perceptron comprises a plurality of full-connection layers and nonlinear activation functions, and is used for extracting nonlinear characteristics implicit in charge characteristic vectors of input pairs of environmental data; Is a matrix transformation operation; The attention feature fusion in step S4 is configured to perform attention weight calculation by using the dynamic environmental factor as a query vector and the spliced charging feature vector as a key value vector, and perform dot multiplication on the weight of the attention weight calculation and the spliced charging feature vector to generate a fused charging feature vector.
  4. 4. The method for detecting abnormal charge of an electric vehicle according to any one of claims 1 to 3, further comprising the steps of, when an abnormality detection result corresponding to a current charging period is obtained: judging whether an abnormal detection result corresponding to the current charging period triggers an abnormal early warning threshold or not, if not, continuing to execute the step S1, if so, suspending the charging flow of the current electric automobile, and prompting manual intervention; The triggering judgment of the constant early warning threshold is expressed as follows: In the formula, An abnormality detection result corresponding to the current charging period T; a history window length for smoothing the reference; representing a minimum risk tolerance threshold; Representing a moderate risk threshold.
  5. 5. A charging abnormality detection apparatus for an electric vehicle for performing the charging abnormality detection method for an electric vehicle according to claim 1, characterized by comprising a charging data preprocessing unit, a time domain feature extraction unit, a frequency spectrum feature extraction unit, a time-frequency feature stitching unit, a time-frequency feature dynamic fusion unit, and an abnormality identification unit; The charging data preprocessing unit is used for preprocessing charging voltage data and charging current data to generate charging time domain information and charging frequency domain information; the time domain feature extraction unit is used for extracting time domain features of the charging time domain information to form a charging time domain feature vector; the frequency spectrum feature extraction unit is used for extracting frequency spectrum features of the charging frequency domain information to form a charging frequency spectrum feature vector; The time-frequency characteristic splicing unit is used for splicing the charging time domain characteristic vector and the charging frequency spectrum characteristic vector to obtain a spliced charging characteristic vector; the time-frequency characteristic dynamic fusion unit is used for carrying out attention characteristic fusion on the spliced charging characteristic vectors according to the dynamic environment factors to generate fused charging characteristic vectors; The anomaly identification unit is used for carrying out anomaly identification on the fused charging characteristic vector and generating an anomaly detection result corresponding to the current charging period.
  6. 6. The device for detecting abnormal charge of an electric vehicle according to claim 5, wherein the time domain feature extraction unit comprises a voltage convolution layer, a voltage embedding layer, a current convolution layer, a current embedding layer, a bidirectional cross attention layer, a time domain feature fusion layer and a bidirectional long-short term memory neural network layer; The voltage convolution layer is used for carrying out one-dimensional convolution operation on the voltage time domain information in the charging time domain information to form a local characteristic sequence of the voltage; The voltage embedding layer is used for carrying out vector coding on the local characteristic sequence of the voltage to form an embedded vector sequence with fixed dimension, so as to obtain an embedded vector of the voltage; the current convolution layer is used for carrying out one-dimensional convolution operation on the current time domain information in the charging time domain information to form a local characteristic sequence of the current; The current embedding layer is used for carrying out vector coding on the local characteristic sequence of the current so as to obtain an embedding vector of the current; The bidirectional cross attention layer is used for taking the voltage embedded vector as a query vector, taking the current embedded vector as a key value vector to generate weight, multiplying and fusing the weight and the current embedded vector to obtain an enhanced current embedded vector; meanwhile, taking the current embedded vector as a query vector, taking the voltage embedded vector as a key value vector, generating weight, multiplying and fusing the weight and the voltage embedded vector, and obtaining an enhanced voltage embedded vector; The time domain feature fusion layer is used for sequentially splicing and fully connecting the enhanced voltage embedded vector and the enhanced current embedded vector to obtain a primarily fused charging time domain feature vector; The bidirectional long-short-term memory neural network layer is used for carrying out bidirectional time sequence feature extraction on the preliminarily fused charging time domain feature vector so as to extract the context feature representation reflected in the voltage and current combined evolution process and form a final charging time domain feature vector.
  7. 7. The apparatus according to claim 6, wherein the spectrum feature extraction unit includes a spectrum input reconstruction module, a depth separable convolution module, a channel compression module, and a spectrum embedding layer; the frequency spectrum input reconstruction module is used for splicing the voltage frequency spectrum information and the current frequency spectrum information in the charging frequency domain information in the channel dimension to form a two-dimensional frequency spectrum tensor; The depth separable convolution module is used for extracting depth separable convolution characteristics of the two-dimensional spectrum tensor to obtain a two-dimensional spectrum characteristic vector; the channel compression module is used for carrying out point-by-point convolution on the two-dimensional spectrum feature vector to obtain compressed two-dimensional spectrum features; The spectrum embedding layer is used for carrying out feature mapping and nonlinear transformation on the compressed spectrum features to generate a spectrum embedding vector with fixed dimension, and a charged spectrum feature vector is obtained.
  8. 8. The device for detecting abnormal charge of an electric vehicle according to claim 7, wherein the dynamic environmental factor is generated by a lightweight neural network, and the specific structure of the lightweight neural network is as follows: In the formula, Representing a dynamic environmental factor; Input representing environmental data; the multi-layer perceptron comprises a plurality of full-connection layers and nonlinear activation functions, and is used for extracting nonlinear characteristics implicit in charge characteristic vectors of input pairs of environmental data; Is a matrix transformation operation; The attention feature fusion in step S4 is configured to perform attention weight calculation by using the dynamic environmental factor as a query vector and the spliced charging feature vector as a key value vector, and perform dot multiplication on the weight of the attention weight calculation and the spliced charging feature vector to generate a fused charging feature vector.
  9. 9. The charging pile is characterized by comprising an electric parameter acquisition module, an environment data acquisition sensor and a charging abnormality detection device of an electric automobile; The electric parameter acquisition module comprises a voltage sensor and a current sensor which are arranged in a charging path, and the voltage sensor and the current sensor are respectively used for acquiring output voltage data and output current data of the charging gun in the charging process in real time, digitizing the output voltage data and the output current data to form charging voltage data and charging current data, and transmitting the charging voltage data and the charging current data to a charging abnormality detection device of the electric automobile; The environment data acquisition sensors are respectively arranged in the charging pile shell and the charging gun and are used for acquiring the temperature data of the charging pile and the temperature data of the charging gun, digitizing the temperature data and transmitting the digitized temperature data to the charging abnormality detection device of the electric automobile; the electric vehicle charging abnormality detection apparatus according to any one of claims 5 to 8.

Description

Charging abnormality detection method and device for electric automobile and charging pile Technical Field The invention relates to the field of electric vehicle charging safety protection, in particular to a charging abnormality detection method, a charging abnormality detection device and a charging pile for an electric vehicle. Background With the popularization of electric vehicles, the use frequency of the charging pile is gradually increased, and the safety problem in the charging process of the electric vehicles, especially the spontaneous combustion accident, has become a focus of social attention, especially when the electric vehicles are in a static state after being charged or fully charged, the Battery pack of the electric vehicles may be spontaneous-burned due to multiple reasons such as overcharge, overheat or faults of a Battery management system (Battery MANAGEMENT SYSTEM, BMS). The spontaneous combustion can not only cause damage to the vehicle, but also easily bring harm to the surrounding environment, especially to places with poor ventilation conditions such as underground parking lots, once the electric vehicle is spontaneous combustion, the common fire extinguisher is difficult to extinguish, and the harmful gas and temperature released during spontaneous combustion of the battery pack can rapidly spread in the low ventilation environment, so that the electric vehicle is harmful to surrounding batteries, charging piles and other electric vehicles. Therefore, the conventional technology generally relies on means such as overvoltage protection, overcurrent protection, leakage protection, insulation resistance detection, and the like, and monitors real-time data such as voltage, current, insulation resistance, and the like to identify an abnormal state occurring in real-time output in a charging process, and trigger a protection mechanism according to the abnormal state so as to prevent more serious abnormality. For example, abnormal fluctuations in current or voltage during charging may trigger an over-current protection or over-voltage protection mechanism to automatically disconnect the charging connection or reduce the charging power, thereby preventing damage to the battery or the charging post. However, the conventional technology is based on identifying the abnormality after the abnormality occurs on the real-time data, and performing corresponding early warning and triggering protection mechanism, and although the response can be timely performed, early warning cannot be performed in advance when slight or progressive abnormality occurs. For example, when there is a slight overheat, overcharge, or slight current fluctuation in the battery pack or the charging pile of the electric vehicle, the system may not be able to recognize in time, and when a major fault occurs, the reaction is usually too slow, and in addition, non-immediate faults such as battery aging and a hardware fault of the charging pile often cannot be detected in time. Based on this, in the prior art, the neural network is generally used to analyze and predict the charging power data to identify the abnormal state in the charging process, and the neural network can capture the fine fluctuation in the charging power data and model the fluctuation, so as to realize the abnormal early warning of the charging process, and expect to detect the potential problem in advance before the fault occurs. However, in the actual charging process, the protection mechanism built in the charging pile, such as overvoltage protection, overcurrent protection, temperature monitoring, etc., is usually executed based on a predetermined threshold or rule, for example, after the charging pile works for a long time, if the temperature reaches a certain set threshold, the charging power is automatically reduced, or when the voltage of the battery pack reaches a set upper limit, the charging pile stops charging to avoid safety problems such as overcharge. In this regard, the existing neural network cannot flexibly adjust its feature expression according to environmental changes in anomaly detection, so that normal protective anomalies and potential dangerous anomalies cannot be effectively distinguished, and a large amount of training data is usually required to be relied on, or different types of anomalies are respectively identified by constructing a plurality of labels, but the method still has a certain limitation. Because the normal protective anomalies (such as overhigh temperature, overcharging and the like) are similar to the characteristic expressions of the potential dangerous anomalies (such as overheat of a battery, system faults and the like), confusion is easy to cause, so that gradient interference occurs in the training process of using a large amount of training data of the neural network, the training efficiency is low, the model is difficult to effectively distinguish the normal protective anomalies from the potential dangero