A charging pile fault diagnosis system based on charging big data

Abstract

The application discloses a charging pile fault diagnosis system based on charging big data, comprising: a charging operation parameter acquisition module, which acquires charging operation parameter data and constructs a charging session data set; a charging stage division module, which generates a stage state vector and an energy closed-loop state vector; a component state mapping module, which generates a component-level stage state vector and a component-level energy closed-loop state vector; a counterfactual working condition constraint module, which generates counterfactual working condition constraint data; a stage transition prediction module, which generates a predicted energy closed-loop state vector and stage evolution offset data based on an improved CPC model; a fault source analysis module, which generates a fault source score; and an operation baseline updating module, which generates a fault diagnosis result and updates an operation baseline. The application improves the accuracy and stability of charging pile fault diagnosis under complex working conditions.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and new energy vehicle charging technology, and in particular to a charging pile fault diagnosis system based on charging big data. Background Technology

[0002] With the large-scale deployment of new energy vehicles and high-power DC charging facilities, charging piles operate under high frequency, high load, and complex conditions for extended periods. Complex coupling relationships gradually form between the power conversion components, thermal management components, insulation detection components, and communication components within the charging system. Changes in the operating state during the charging process directly affect charging safety and equipment stability. Regarding fault diagnosis during charging pile operation, existing technologies mainly employ threshold alarms, univariate anomaly detection, or traditional time-series neural networks to identify faults in operating parameters. However, the following problems commonly exist in actual charging scenarios: Traditional fault diagnosis methods mostly rely on single parameters such as voltage, current, or temperature for state judgment, which makes it difficult to effectively characterize the continuous evolution relationship between charging stages and the energy coupling relationship between different functional components. This results in insufficient ability to identify latent faults, gradual faults, and faults propagating across components. Existing deep learning-based diagnostic methods usually extract features directly from time-series data, lacking constraint modeling of the charging stage transition process and the energy closed-loop convergence relationship. This makes it difficult to accurately distinguish between normal convergence and pseudo-convergence states, and it is prone to misdiagnosis under complex operating conditions. In addition, existing methods generally do not establish counterfactual operating condition constraints under different vehicle states of charge, ambient temperatures, and output power conditions, resulting in insufficient adaptability of the model to operating condition disturbances. At the same time, existing fault diagnosis models lack the ability to model fault propagation paths and dynamic update mechanisms of post-maintenance operating baselines, making it difficult to achieve continuous evolution analysis of the long-term operating status of charging piles.

[0003] Therefore, how to provide a charging pile fault diagnosis system based on charging big data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a charging pile fault diagnosis system based on charging big data. This invention constructs the correlation between stage state vectors, energy closed-loop state vectors, and counterfactual operating condition constraint data. Combined with an improved CPC model, a closed-loop consistency constraint mechanism, and a dynamic negative sample generation mechanism, it achieves joint modeling of the evolution relationship of charging stages, the coupling relationship of functional components, and the fault propagation path. This improves the ability to identify latent faults, pseudo-convergence faults, and cross-component propagation faults under complex operating conditions, and enables dynamic updating of the operating baseline and intelligent diagnosis of fault sources.

[0005] A charging pile fault diagnosis system based on charging big data according to an embodiment of the present invention includes: The charging operation parameter acquisition module is used to collect charging operation parameter data of the charging pile during the charging process and to build a charging session dataset. The charging phase division module is used to divide the charging session dataset into charging phases and generate a phase state vector and an energy closed-loop state vector corresponding to each charging phase. The component state mapping module is used to map the stage state vector and energy closed-loop state vector to the charging pile functional components, and generate component-level stage state vector and component-level energy closed-loop state vector. The counterfactual operating condition constraint module is used to generate counterfactual operating condition constraint data for the corresponding charging stage based on the charging operation parameter data. The stage transition prediction module is used to input the component-level stage state vector, the component-level energy closed-loop state vector, and the counterfactual operating condition constraint data into the improved CPC model, generate the predicted energy closed-loop state vector corresponding to the next charging stage, and generate stage evolution offset data. The fault source analysis module is used to generate stage convergence slope and convergence consistency entropy based on the stage evolution offset data, and to generate the fault source score of the corresponding component by combining the component-level stage state vector. The baseline update module is used to generate fault diagnosis results based on the fault source score, stage convergence slope, convergence consistency entropy and stage evolution offset data, and update the operating baseline of the corresponding component based on the component status data after repair.

[0006] Optionally, the charging operation parameter acquisition module includes: The system collects voltage sampling data, current sampling data, output power sampling data, module temperature sampling data, insulation resistance sampling data, vehicle state of charge data, BMS communication message data, gun wire specifications and power battery parameters during the charging session, and generates a multi-source time-series data stream according to a unified time base. The multi-source time-series data stream is time-aligned, mapping data corresponding to different sampling frequencies to the same sampling time axis, and generating time-series synchronization data based on the time interval between adjacent sampling moments. The time-series synchronization data is subjected to abnormal fluctuation screening. Transient pulse data is identified based on the change amplitude, change direction and change duration between continuous sampling points. The abnormal sampling interval is reconstructed by interpolation of adjacent time-series intervals to generate reconstructed time-series data. The reconstructed timing data is sliced ​​into running states. The charging session boundaries are divided based on the charging start signal, BMS handshake message, output power change slope, and charging end identifier, and data segments of the corresponding charging session are generated. The data segments of the charging session are processed by multi-dimensional parameter association, and the voltage parameters, current parameters, output power parameters and module temperature parameters corresponding to the same sampling time are combined in time sequence to generate a charging operation parameter matrix; The charging operation parameter matrix is ​​processed by working condition parameter association, and the vehicle state of charge data, ambient temperature data, charging cable specification parameters and power battery parameters are mapped to the corresponding sampling interval to generate a working condition association dataset. An index mapping relationship is established for the aforementioned working condition associated dataset according to the charging session number, sampling time series, and charging device number to generate a charging session dataset.

[0007] Optionally, the charging stage division module includes: Joint timing analysis is performed on the output power timing data, current timing data, and BMS communication message timing data in the charging session dataset. The charging stage switching node is identified based on the output power change slope, current change slope, BMS handshake establishment flag, constant voltage request flag, and charging stop flag. The charging session dataset is divided into stages based on the charging stage switching nodes, generating handshake stage data segments, pre-charge stage data segments, constant current stage data segments, constant voltage stage data segments, power reduction stage data segments, and stop stage data segments. The average value, fluctuation amplitude, and rate of change of voltage parameters, current parameters, output power parameters, module temperature parameters, and insulation resistance parameters in each stage data segment are calculated to generate the corresponding stage operation parameter sequence for the charging stage. The time window is divided into the sequence of operating parameters for the stage, and the parameters are arranged in time sequence according to the magnitude, direction and rate of change of the parameters within the continuous sampling period to generate the stage state vector corresponding to the charging stage. The output power parameter, current parameter, and module temperature parameter in the stage state vector are correlated and calculated to generate the power convergence sequence, current convergence sequence, and thermal convergence sequence for the corresponding sampling period. The power convergence sequence, current convergence sequence, and thermal convergence sequence are time-coupled, and a stage energy correlation matrix is ​​established based on the convergence direction, convergence speed, and convergence amplitude between consecutive sampling periods. The stage energy correlation matrix is ​​processed by stage mapping, and the stage energy correlation matrices corresponding to each charging stage are combined in chronological order to generate the energy closed-loop state vector of the corresponding charging stage.

[0008] Optionally, the component state mapping module includes: The voltage parameters, current parameters, output power parameters, module temperature parameters, insulation resistance parameters, and BMS communication message parameters in the stage state vector are analyzed for parameter change synchronicity, parameter change delay, and parameter change amplitude correlation, and a parameter change relationship matrix is ​​established. Based on the parameter coupling relationship in the parameter change relationship matrix, the voltage parameter, output power parameter and current parameter are mapped to the power conversion functional component, the module temperature parameter is mapped to the thermal management functional component, the insulation resistance parameter is mapped to the insulation detection functional component, and the BMS communication message parameter is mapped to the communication functional component, thus generating a component parameter mapping sequence. The component parameter mapping sequence is divided into time windows, and a component state correlation matrix is ​​established based on the parameter change amplitude, parameter change direction and parameter change rate within the same time window; Statistical processing is performed on the number of synchronous changes, the duration of delayed changes, and the difference in the magnitude of changes in the component state correlation matrix to establish a component state change sequence; The component state change sequence is subjected to component aggregation processing, and the parameter change sequence associated with the corresponding functional component is combined according to the sampling time order to generate a component-level stage state vector. Based on the parameter change relationship matrix and component state correlation matrix, the power convergence sequence, current convergence sequence and thermal convergence sequence in the energy closed-loop state vector are processed by component mapping to establish energy convergence correlation relationships corresponding to different functional components. The energy convergence correlations of different functional components are combined in a time series to generate a component-level energy closed-loop state vector.

[0009] Optionally, the counterfactual operating condition constraint module includes: The vehicle state of charge data, ambient temperature data, output power data, charging cable specifications, and power battery parameters in the charging operation parameter data are used to extract operating condition parameters and establish an operating condition parameter sequence. The operating condition parameter sequence is divided into time windows. An operating condition reference interval is established based on the vehicle charge state change interval, ambient temperature change interval, and output power change interval within the same time window, and a corresponding operating condition change matrix is ​​generated. The parameter matching process is performed on the gun wire specification parameters, power battery parameters and output power variation range in the operating condition change matrix to establish the operating condition coupling relationship between different operating condition parameters. Based on the correspondence between the charging stage switching nodes and the time window, the operating condition coupling relationship is processed by stage mapping, and different operating condition parameters are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage, generating a stage operating condition association sequence. The frequency, amplitude, and duration of fluctuations in the ambient temperature change range, output power change range, and vehicle state of charge change range in the stage operating condition correlation sequence are statistically processed to establish an operating condition change trend sequence. The time-series correlation processing is performed on the operating condition change trend sequence, and an operating condition constraint correlation matrix is ​​established based on the change direction, change amplitude and change duration between consecutive time windows; The working condition constraint correlation matrix is ​​processed by stage combination, and the stage working condition constraint relationship is established based on the working condition change trend within the same working condition reference interval, generating counterfactual working condition constraint data for the corresponding charging stage.

[0010] Optionally, the stage transition prediction module includes: The component-level stage state vector, the component-level energy closed-loop state vector, and the counterfactual operating condition constraint data are input into the improved CPC model. The improved CPC model includes a stage transition coding layer, a dual-path context coding layer, a counterfactual operating condition constraint layer, a convergence residual enhancement layer, a closed-loop consistency constraint layer, and a dynamic negative sample generation layer. The stage transition coding layer encodes the parameter change sequence corresponding to different charging stages in a stage sequence, and establishes a stage transition coding sequence based on the parameter change direction, parameter change rate and parameter change amplitude between adjacent charging stages. The dual-path context coding layer includes a time evolution coding path and a component association coding path. The time evolution coding path establishes a time evolution context sequence based on the continuous relationship of parameter changes between continuous charging stages. The component association coding path establishes a component association context sequence based on the synchronous change relationship and the delayed change relationship of parameters between different functional components. The temporal evolution context sequence and the component-associated context sequence are processed by vector concatenation and unified feature dimension mapping to generate a dual-path context feature sequence; The counterfactual operating condition constraint layer performs operating condition constraint mapping processing on the dual-path context feature sequence based on the operating condition change trend within the same operating condition reference interval, and generates an operating condition constraint feature sequence. The dual-path context feature sequence and the working condition constraint feature sequence are jointly encoded to generate a stage context encoding sequence; The convergence residual enhancement layer predicts the power convergence sequence, current convergence sequence and thermal convergence sequence corresponding to the next charging stage, establishes the predicted energy closed loop state sequence, and establishes the convergence residual enhancement sequence based on the residual change direction, residual change rate and residual change amplitude between the predicted energy closed loop state sequence and the actual energy closed loop state sequence corresponding to the next charging stage. The closed-loop consistency constraint layer establishes a closed-loop consistency constraint sequence based on the consistency of convergence direction, convergence rate, and convergence magnitude among the power convergence sequence, current convergence sequence, and thermal convergence sequence, and performs consistency correction processing on the stage context encoding sequence based on the closed-loop consistency constraint sequence. The dynamic negative sample generation layer performs stage misalignment combination processing on the stage context encoding sequences corresponding to different working condition reference intervals to establish a working condition disturbance negative sample sequence. The negative sample sequence of the working condition disturbance, the convergence residual enhancement sequence, the closed-loop consistency constraint sequence and the stage context encoding sequence are compared and predicted to generate the predicted energy closed-loop state vector corresponding to the next charging stage. The energy closed-loop evolution offset data is generated based on the predicted energy closed-loop state vector, the energy closed-loop state vector corresponding to the actual next charging stage, the closed-loop consistency constraint sequence, and the convergence residual enhancement sequence.

[0011] Optionally, the fault source analysis module includes: The power offset sequence, current offset sequence, and thermal offset sequence in the stage energy closed-loop evolution offset data are divided into time windows to establish a stage offset parameter sequence. The offset change analysis is performed on the continuous time windows in the stage offset parameter sequence, and a stage convergence sequence is established based on the offset change direction, offset change magnitude and offset change rate between adjacent time windows. The slope of the continuous offset change relationship in the stage convergence sequence is calculated, and the stage convergence slope is generated based on the correspondence between the offset change amplitude and the time interval between continuous time windows. Synchronous change relationship analysis is performed on the power offset sequence, current offset sequence and thermal offset sequence, and a closed-loop consistent correlation sequence is established based on the consistency of change direction, change rate and change amplitude among different offset sequences. Statistical processing is performed on the distribution relationship of different offset sequences in the closed-loop consistency association sequence, and a convergence consistency entropy is generated based on the distribution frequency, distribution dispersion and change direction of different offset sequences. Component correlation analysis is performed on the parameter change sequence and stage energy closed-loop evolution offset data in the component-level stage state vector to establish state correlation sequences and offset propagation correlation sequences corresponding to different functional components; The stage convergence slope, convergence consistency entropy, state correlation sequence, and offset propagation correlation sequence are jointly correlated and processed to generate the fault source score of the corresponding functional component based on the offset change intensity, offset duration, and offset propagation direction of the different functional components.

[0012] Optionally, the baseline update module includes: Time window alignment processing is performed on the fault source score, stage convergence slope, convergence consistency entropy and stage energy closed-loop evolution offset data to establish a fault diagnosis parameter sequence. Correlation analysis is performed on the fault source score, stage convergence slope and convergence consistency entropy in the fault diagnosis parameter sequence, and a fault evolution trend sequence is established based on the change direction, change amplitude and change duration between continuous time windows. Propagation relationship analysis was performed on the power offset sequence, current offset sequence and thermal offset sequence in the stage energy closed-loop evolution offset data. Fault propagation correlation sequence was established based on the temporal sequence relationship and offset change correlation relationship between different offset sequences. The fault evolution trend sequence and fault propagation correlation sequence are processed to construct propagation paths, and fault propagation path sequences are established based on the offset propagation direction, offset propagation duration and offset propagation intensity corresponding to different functional components. The continuous offset propagation relationships in the fault propagation path sequence are processed by stage mapping, and different offset propagation relationships are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage to generate a stage fault association sequence. The corresponding relationships of different functional components in the stage fault association sequence are statistically processed, and the fault stage, fault propagation path and fault diagnosis result of the corresponding functional component are generated based on the fault source score, stage convergence slope and convergence consistency entropy. Collect the status data of the repaired components, perform offset correlation analysis between the status data of the repaired components and the historical normal operation status data, perform baseline offset correction processing based on the difference in the change of corresponding parameters before and after the repair, and update the operating baseline of the corresponding functional components.

[0013] The beneficial effects of this invention are: This invention addresses the problems of insufficient latent fault identification capability, weak adaptability to complex operating conditions, and difficulty in modeling fault propagation relationships during charging pile operation by constructing multidimensional correlations among charging stage state vectors, energy closed-loop state vectors, and counterfactual operating condition constraint data. It combines an improved CPC model with a collaborative design of energy closed-loop consistency constraint mechanisms. The invention proposes a stage evolution prediction strategy based on stage transition coding, dual-path context modeling, and counterfactual operating condition constraints, significantly improving the modeling capability of state evolution relationships between different charging stages and the feature stability under complex operating conditions. The model structure introduces time evolution coding paths and component association coding paths, achieving correlation alignment between charging stages and component states through joint coding of stage continuity relationships and functional component coupling relationships. In closed-loop modeling… In the first stage, a convergence residual enhancement layer and a closed-loop consistency constraint layer are introduced to jointly constrain the consistency of convergence direction, convergence rate, and convergence amplitude among power convergence, current convergence, and thermal convergence, effectively enhancing the model's ability to identify pseudo-convergence states and gradual fault states. In the negative sample construction stage, a dynamic negative sample generation layer is designed to establish negative samples of operating condition disturbances by combining different operating condition reference intervals and different charging stages, improving the model's ability to distinguish complex operating condition disturbances and cross-stage anomaly propagation. Finally, the fault propagation path is constructed by combining stage convergence slope, convergence consistency entropy, and stage energy closed-loop evolution offset data, and the operating baseline is dynamically updated based on the component status data after maintenance, realizing continuous evolution analysis of charging pile fault states, fault source localization, and intelligent diagnosis. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a charging pile fault diagnosis system based on charging big data proposed in this invention; Figure 2 This is a schematic diagram of the structure of an improved CPC model in a charging pile fault diagnosis system based on charging big data proposed in this invention. Detailed Implementation

[0015] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0016] refer to Figures 1-2 A charging pile fault diagnosis system based on charging big data includes: The charging operation parameter acquisition module is used to collect charging operation parameter data of the charging pile during the charging process and to build a charging session dataset. The charging phase segmentation module is used to segment the charging session dataset into charging phases and generate the phase state vector and energy closed-loop state vector corresponding to each charging phase. The component state mapping module is used to map the stage state vector and the energy closed-loop state vector to the charging pile functional components, and generate the component-level stage state vector and the component-level energy closed-loop state vector. The counterfactual operating condition constraint module is used to generate counterfactual operating condition constraint data for the corresponding charging stage based on the charging operation parameter data. The stage transition prediction module is used to input the component-level stage state vector, the component-level energy closed-loop state vector, and the counterfactual operating condition constraint data into the improved CPC model, generate the predicted energy closed-loop state vector corresponding to the next charging stage, and generate stage evolution offset data. The fault source analysis module is used to generate stage convergence slope and convergence consistency entropy based on stage evolution offset data, and to generate the fault source score of the corresponding component by combining the component-level stage state vector. The baseline update module is used to generate fault diagnosis results based on fault source scores, stage convergence slope, convergence consistency entropy, and stage evolution offset data, and to update the operating baseline of the corresponding component based on the component status data after repair.

[0017] In this embodiment, the charging operation parameter acquisition module includes: The system collects voltage sampling data, current sampling data, output power sampling data, module temperature sampling data, insulation resistance sampling data, vehicle state of charge data, BMS communication message data, gun wire specifications and power battery parameters during the charging session, and generates a multi-source time-series data stream according to a unified time base. Time alignment processing is performed on multi-source time-series data streams, mapping data corresponding to different sampling frequencies to the same sampling time axis, and generating time-series synchronization data based on the time interval between adjacent sampling moments; Abnormal fluctuations are filtered out from the time-series synchronization data. Transient pulse data are identified based on the change amplitude, change direction and change duration between consecutive sampling points. The abnormal sampling interval is reconstructed by interpolation between adjacent time-series intervals to generate reconstructed time-series data. The reconstructed timing data is sliced ​​into running states. The charging session boundaries are divided based on the charging start signal, BMS handshake message, output power change slope and charging end identifier, and data segments of the corresponding charging session are generated. Multi-dimensional parameter correlation processing is performed on the data segments of the charging session. The voltage parameters, current parameters, output power parameters and module temperature parameters corresponding to the same sampling time are combined in time sequence to generate a charging operation parameter matrix. The charging operation parameter matrix is ​​processed to correlate the operating condition parameters, and the vehicle state of charge data, ambient temperature data, charging cable specifications and power battery parameters are mapped to the corresponding sampling intervals to generate an operating condition correlated dataset. An index mapping relationship is established for the working condition associated dataset according to the charging session number, sampling time series, and charging device number to generate the charging session dataset.

[0018] In this implementation, voltage sampling data, current sampling data, output power sampling data, and module temperature sampling data in the multi-source time-series data stream are periodically collected using different sampling frequencies. During time alignment processing, a unified time axis is established based on the sampling timestamps, and linear time mapping is performed according to the time interval between adjacent sampling moments. During abnormal fluctuation screening processing, data segments with a change amplitude between consecutive sampling points exceeding twice the average change amplitude of the corresponding sampling interval and a duration of less than three sampling cycles are identified as transient pulse data. In the operation status slicing processing, the charging session boundary is divided based on the handshake establishment identifier, output power change slope, and charging end message status in the BMS handshake message. In the operating condition parameter association processing, vehicle state of charge data, ambient temperature data, charging cable specifications, and power battery parameters are mapped to the corresponding sampling intervals according to the sampling time axis, and a time-series correspondence is established with the charging operation parameter matrix.

[0019] In this embodiment, the charging stage division module includes: Joint timing analysis is performed on the output power timing data, current timing data, and BMS communication message timing data in the charging session dataset. The charging stage switching node is identified based on the output power change slope, current change slope, BMS handshake establishment flag, constant voltage request flag, and charging stop flag. The charging session dataset is divided into stages based on the charging stage switching nodes, generating data segments for the handshake stage, pre-charging stage, constant current stage, constant voltage stage, power reduction stage, and stop stage. The average value, fluctuation amplitude, and rate of change of voltage parameters, current parameters, output power parameters, module temperature parameters, and insulation resistance parameters in each stage data segment are calculated to generate the corresponding stage operation parameter sequence for the charging stage. The time window is divided into the sequence of phase operation parameters, and the parameters are arranged in time sequence according to the magnitude, direction and rate of change of parameter changes within the continuous sampling period to generate the phase state vector of the corresponding charging phase. The output power parameters, current parameters, and module temperature parameters in the stage state vector are correlated and calculated to generate the power convergence sequence, current convergence sequence, and thermal convergence sequence for the corresponding sampling period. Time-series coupling processing is performed on the power convergence sequence, current convergence sequence and thermal convergence sequence, and a stage energy correlation matrix is ​​established based on the convergence direction, convergence speed and convergence amplitude between continuous sampling periods. The stage energy correlation matrix is ​​processed by stage mapping, and the stage energy correlation matrices corresponding to each charging stage are combined in chronological order to generate the energy closed-loop state vector of the corresponding charging stage.

[0020] In this implementation, the slope of output power change is calculated as the ratio of the difference in output power between adjacent sampling moments to the corresponding time interval, and the slope of current change is calculated as the ratio of the difference in current between adjacent sampling moments to the corresponding time interval. During the time window division process, a continuous sliding time window is established based on the duration of the charging phase and the sampling frequency, and overlapping sampling intervals are retained between adjacent time windows. The magnitude of parameter change is characterized by the difference between the maximum and minimum parameter values ​​within the same time window, and the rate of parameter change is characterized by the ratio of the parameter change between consecutive sampling points to the sampling time interval. The power convergence sequence, current convergence sequence, and thermal convergence sequence are constructed using the decreasing trend of the corresponding parameter change rate within the continuous time window. The matrix elements in the stage energy correlation matrix are characterized by the consistency of the direction of change and the correlation of the magnitude of change between different convergence sequences within the same time window.

[0021] In this embodiment, the component state mapping module includes: The voltage, current, output power, module temperature, insulation resistance, and BMS communication message parameters in the stage state vector are analyzed for synchronicity, delay, and correlation of parameter changes, and a parameter change magnitude is established. Based on the parameter coupling relationships in the parameter change relationship matrix, voltage parameters, output power parameters, and current parameters are mapped to power conversion functional components, module temperature parameters are mapped to thermal management functional components, insulation resistance parameters are mapped to insulation detection functional components, and BMS communication message parameters are mapped to communication functional components, generating a component parameter mapping sequence. Divide the component parameter mapping sequence into time windows, and establish a component state correlation matrix based on the parameter change amplitude, parameter change direction and parameter change rate within the same time window; Statistical processing is performed on the number of synchronous changes, the duration of delayed changes, and the difference in the magnitude of changes in the component state correlation matrix to establish a component state change sequence; The component state change sequence is aggregated by combining the parameter change sequence associated with the corresponding functional component according to the sampling time order to generate a component-level stage state vector. Based on the parameter change relationship matrix and the component state correlation matrix, the power convergence sequence, current convergence sequence and thermal convergence sequence in the energy closed loop state vector are processed by component mapping to establish the energy convergence correlation relationship corresponding to different functional components. The energy convergence correlations of different functional components are combined in a time series to generate a component-level energy closed-loop state vector.

[0022] In this implementation, parameter change synchronicity is characterized by the proportion of sampling points with consistent change directions of different parameters within the same time window to the total number of sampling points; parameter change delay is characterized by the time difference between different parameters reaching their peak values; and parameter change amplitude correlation is characterized by the trend of the difference between the change amplitudes of different parameters. The component state association matrix is ​​updated according to the time window order, and continuous sampling intervals are retained between adjacent time windows. In the component aggregation process, the parameter change sequences associated with corresponding functional components are vector-concatenated according to the sampling time order, and a time series association relationship is established based on the sampling time interval. In the energy convergence association construction process, the power convergence sequence, current convergence sequence, and thermal convergence sequence correspond to the operating state change processes of the power conversion functional component, thermal management functional component, and insulation detection functional component, respectively.

[0023] In this embodiment, the counterfactual operating condition constraint module includes: Operating condition parameters are extracted from the vehicle state of charge data, ambient temperature data, output power data, charging cable specifications and power battery parameters in the charging operation parameter data, and an operating condition parameter sequence is established. The operating condition parameter sequence is divided into time windows. Based on the vehicle's state of charge change range, ambient temperature change range, and output power change range within the same time window, an operating condition reference range is established, and a corresponding operating condition change matrix is ​​generated. Parameter matching processing is performed on the gun wire specification parameters, power battery parameters and output power variation range in the operating condition change matrix to establish the operating condition coupling relationship between different operating condition parameters. Based on the correspondence between the charging stage switching nodes and time windows, the operating condition coupling relationship is processed by stage mapping, and different operating condition parameters are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage, generating a stage operating condition association sequence. The frequency, amplitude, and duration of fluctuations in the ambient temperature change range, output power change range, and vehicle state of charge change range in the stage operating condition correlation sequence are statistically processed to establish an operating condition change trend sequence. The time-series correlation processing is performed on the operating condition change trend sequence, and an operating condition constraint correlation matrix is ​​established based on the change direction, change amplitude and change duration between consecutive time windows; The working condition constraint correlation matrix is ​​processed by stage combination, and the stage working condition constraint relationship is established based on the working condition change trend within the same working condition reference interval, generating counterfactual working condition constraint data for the corresponding charging stage.

[0024] In this implementation, the operating condition reference interval is jointly divided using the range of vehicle state of charge variation, ambient temperature variation, and output power variation within a continuous time window, with overlapping parameter intervals retained between adjacent operating condition reference intervals. During parameter matching, the rated current carrying capacity range is used to match the charging cable specifications with the output power variation interval, and the charging capacity variation range is used to match the power battery parameters with the vehicle state of charge variation interval. The operating condition change trend sequence is sorted according to the consistency of the changing direction of operating condition parameters within a continuous time window, and a continuous relationship of operating condition changes is established based on the duration of fluctuations. During the establishment of stage operating condition constraint relationships, the operating condition change trends of corresponding charging stages within the same operating condition reference interval are time-aligned, and stage operating condition constraint relationships are established based on the difference in change amplitude and the consistency of change direction.

[0025] In this embodiment, the stage transition prediction module includes: The component-level stage state vector, component-level energy closed-loop state vector, and counterfactual operating condition constraint data are input into the improved CPC model. The improved CPC model includes a stage transition coding layer, a dual-path context coding layer, a counterfactual operating condition constraint layer, a convergence residual enhancement layer, a closed-loop consistency constraint layer, and a dynamic negative sample generation layer. The stage transition coding layer encodes the parameter change sequence corresponding to different charging stages in a stage sequence, and establishes a stage transition coding sequence based on the parameter change direction, parameter change rate and parameter change amplitude between adjacent charging stages. The dual-path context coding layer includes a time evolution coding path and a component association coding path. The time evolution coding path establishes a time evolution context sequence based on the continuous relationship of parameter changes between continuous charging stages, while the component association coding path establishes a component association context sequence based on the synchronous change relationship and the delayed change relationship of parameters between different functional components. Vector concatenation and unified feature dimension mapping are performed on the temporal evolution context sequence and the component association context sequence to generate a dual-path context feature sequence; The counterfactual operating condition constraint layer performs operating condition constraint mapping processing on the dual-path context feature sequence based on the operating condition change trend within the same operating condition reference interval, and generates an operating condition constraint feature sequence. The dual-path context feature sequence and the working condition constraint feature sequence are jointly encoded to generate the stage context encoding sequence. The convergence residual enhancement layer predicts the power convergence sequence, current convergence sequence and thermal convergence sequence corresponding to the next charging stage, establishes the predicted energy closed loop state sequence, and establishes the convergence residual enhancement sequence based on the residual change direction, residual change rate and residual change amplitude between the predicted energy closed loop state sequence and the actual energy closed loop state sequence corresponding to the next charging stage. The closed-loop consistency constraint layer establishes a closed-loop consistency constraint sequence based on the consistency of convergence direction, convergence rate, and convergence magnitude among the power convergence sequence, current convergence sequence, and thermal convergence sequence, and performs consistency correction processing on the stage context encoding sequence based on the closed-loop consistency constraint sequence. The dynamic negative sample generation layer performs stage misalignment combination processing on the stage context encoding sequences corresponding to different working condition reference intervals to establish a working condition disturbance negative sample sequence. Comparative prediction processing is performed on the negative sample sequence of operating condition disturbance, the enhanced sequence of convergence residual, the closed-loop consistency constraint sequence and the stage context encoding sequence to generate the predicted energy closed-loop state vector corresponding to the next charging stage. The generation stage energy closed-loop evolution offset data is based on the predicted energy closed-loop state vector, the energy closed-loop state vector corresponding to the actual next charging stage, the closed-loop consistency constraint sequence, and the convergence residual enhancement sequence.

[0026] In this implementation, both the improved CPC model and the traditional CPC model use a contrastive predictive coding mechanism to learn features from time series data. Both predict the state features corresponding to the next time step through context coding sequences and establish time series correlations based on the differences between the prediction results and the actual results, thereby realizing the modeling of the time series state change process. The improved CPC model introduces a stage transition coding layer, a dual-path context coding layer, a counterfactual operating condition constraint layer, a convergence residual enhancement layer, a closed-loop consistency constraint layer, and a dynamic negative sample generation layer on the basis of the traditional CPC model. The stage transition coding layer encodes the parameter change relationship between different charging stages in a stage sequence. The dual-path context coding layer synchronously establishes the time evolution relationship and the functional component association relationship. The counterfactual operating condition constraint layer establishes the operating condition constraint relationship corresponding to the same operating condition reference interval. The closed-loop consistency constraint layer establishes the consistency constraint relationship between power convergence, current convergence, and thermal convergence. The improved CPC model utilizes stage transition relationships, operating condition constraint relationships, and energy closed-loop consistency relationships to jointly establish a charging stage evolution model, reducing the interference of different operating condition changes on the stage prediction process, improving the ability to identify stage evolution offsets under latent fault conditions, and enhancing the ability to distinguish fault propagation relationships between different functional components.

[0027] In this embodiment, the fault source analysis module includes: Time windows are divided into power offset sequences, current offset sequences, and thermal offset sequences in the stage energy closed-loop evolution offset data to establish stage offset parameter sequences. Migration change analysis is performed on continuous time windows in the stage migration parameter sequence, and a stage convergence sequence is established based on the migration change direction, migration change magnitude and migration change rate between adjacent time windows. The slope of the continuous offset change relationship in the stage convergence sequence is calculated, and the stage convergence slope is generated based on the correspondence between the offset change amplitude and the time interval between continuous time windows. Synchronous variation relationship analysis was performed on the power offset sequence, current offset sequence and thermal offset sequence, and a closed-loop consistent correlation sequence was established based on the consistency of change direction, change rate and change amplitude among different offset sequences. Statistical processing is performed on the distribution relationship of different offset sequences in the closed-loop consistency association sequence, and convergence consistency entropy is generated based on the distribution frequency, distribution dispersion and change direction of different offset sequences. Component correlation analysis is performed on the parameter change sequence and stage energy closed-loop evolution offset data in the component-level stage state vector to establish state correlation sequences and offset propagation correlation sequences corresponding to different functional components; The stage convergence slope, convergence consistency entropy, state correlation sequence, and offset propagation correlation sequence are jointly correlated. Based on the offset change intensity, offset duration, and offset propagation direction of different functional components, the fault source score of the corresponding functional component is generated.

[0028] In this implementation, the stage convergence slope is characterized by the ratio of the magnitude of the offset change between consecutive time windows to the corresponding time interval, and the stage convergence relationship is established based on the slope change trend corresponding to multiple consecutive time windows; during the construction of the closed-loop consistency correlation sequence, the power offset sequence, current offset sequence, and thermal offset sequence are synchronously aligned according to the same time window, and the consistency correlation relationship is established based on the number of time windows with consistent change directions among different offset sequences; during the construction of the convergence consistency entropy, the frequency of the consistency distribution of change directions corresponding to different offset sequences is mapped to the corresponding probability interval, and the convergence consistency entropy is established based on the degree of dispersion of the distribution between different probability intervals; during the construction of the offset propagation correlation sequence, the offset propagation direction correlation relationship is established based on the temporal sequence relationship between the parameter change sequences of different functional components.

[0029] In this embodiment, the baseline update module is run, including: Time window alignment processing is performed on the fault source score, stage convergence slope, convergence consistency entropy and stage energy closed-loop evolution offset data to establish a fault diagnosis parameter sequence. Correlation analysis is performed on the fault source score, stage convergence slope and convergence consistency entropy in the fault diagnosis parameter sequence, and a fault evolution trend sequence is established based on the change direction, change amplitude and change duration between continuous time windows. Propagation relationship analysis was performed on the power offset sequence, current offset sequence and thermal offset sequence in the stage energy closed-loop evolution offset data. Fault propagation correlation sequence was established based on the temporal sequence relationship and offset change correlation relationship between different offset sequences. The fault evolution trend sequence and fault propagation correlation sequence are processed to construct propagation paths. Fault propagation path sequences are established based on the offset propagation direction, offset propagation duration and offset propagation intensity corresponding to different functional components. A stage mapping process is performed on the continuous offset propagation relationship in the fault propagation path sequence, and different offset propagation relationships are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage to generate a stage fault association sequence. Statistical processing is performed on the correspondence between different functional components in the stage fault association sequence. Based on the fault source score, stage convergence slope and convergence consistency entropy, the fault stage, fault propagation path and fault diagnosis result of the corresponding functional component are generated. Collect the status data of the repaired components, perform offset correlation analysis between the status data of the repaired components and the historical normal operation status data, perform baseline offset correction processing based on the difference in the change of corresponding parameters before and after the repair, and update the operating baseline of the corresponding functional components.

[0030] In this implementation, the fault evolution trend sequence is sorted chronologically according to the direction of change of fault source fractions corresponding to continuous time windows, and a continuous relationship of fault evolution is established based on the change amplitude between multiple consecutive time windows. During the construction of the fault propagation correlation sequence, the power offset sequence, current offset sequence, and thermal offset sequence are synchronously aligned according to the same time window, and an offset propagation relationship is established based on the order of the peak occurrence times of different offset sequences. During the construction of the fault propagation path sequence, a propagation intensity correlation is established based on the number of consecutive occurrences of the offset propagation relationship corresponding to different functional components. In the baseline offset correction process, the post-repair component status data and historical normal operation status data are time-sequentially aligned according to the same charging stage, and the operating baseline of the corresponding functional component is updated based on the average change trend of the difference in corresponding parameter changes before and after repair.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to the operation of a 120kW dual-gun DC charging pile in a public DC fast charging station for new energy vehicles. This charging station is located in an urban transportation hub area, with an average of over 380 vehicles charging daily, operating at high frequency and high load for extended periods. The charging pile includes power conversion components, thermal management components, insulation detection components, and communication components. These components are susceptible to changes in ambient temperature, vehicle state of charge, and high-power charging impacts during long-term operation, leading to problems such as abnormal energy convergence, abnormal temperature drift, and abnormal power fluctuations between charging stages. Traditional fault diagnosis methods based on threshold alarms and univariate detection can only identify explicit faults, and are difficult to effectively identify gradual faults, pseudo-convergence faults, and faults propagating across components during the charging stage, easily resulting in false alarms and missed diagnoses.

[0032] In practical applications, the charging pile is first equipped with voltage sampling data, current sampling data, output power sampling data, module temperature sampling data, insulation resistance sampling data, vehicle state of charge data, and BMS communication message data during continuous operation. A charging session dataset is then established based on a unified time reference. Subsequently, the charging process is divided into stages based on the output power change slope, current change slope, and stage switching identifiers in the BMS communication messages, forming stage state vectors and energy closed-loop state vectors corresponding to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, and power reduction stage.

[0033] In the stage evolution modeling process, the stage state vector, energy closed-loop state vector, and counterfactual operating condition constraint data are input into the improved CPC model. A continuous evolutionary relationship between charging stages is established through a stage transition coding layer, and a dual-path context coding layer synchronously establishes the temporal evolution relationship and the functional component association relationship. A closed-loop consistency constraint layer is used to jointly constrain the consistency of convergence direction and convergence rate among the power convergence sequence, current convergence sequence, and thermal convergence sequence. During system operation, it was found that some charging piles exhibited stable output power in the constant voltage stage, but the module temperature continued to rise, and the current convergence sequence showed abnormal drift. Traditional methods identify this state as a normal stable state, while this invention can identify the inconsistency between power convergence and thermal convergence based on the closed-loop consistency constraint mechanism, thereby identifying pseudo-convergence fault states.

[0034] Furthermore, the system establishes a fault propagation correlation sequence based on the stage energy closed-loop evolution offset data, and analyzes the fault propagation path by combining the stage convergence slope and convergence consistency entropy. When an abnormal temperature rise occurs in the power conversion functional component, the thermal offset sequence first experiences abnormal drift, followed by the current offset sequence and output power offset sequence gradually showing synchronous anomalies. The system identifies the fault as spreading from the thermal management functional component to the power conversion functional component based on the offset propagation direction, and generates the corresponding fault source score for the functional component. After maintenance, the system continues to collect component status data after maintenance, and dynamically corrects the operating baseline based on the difference in parameter changes before and after maintenance, enabling the subsequent fault diagnosis process to adapt to changes in the operating status of the component after aging.

[0035] The improved CPC model, based on the traditional contrastive prediction coding structure, introduces a stage transition coding mechanism, a counterfactual operating condition constraint mechanism, and a closed-loop consistency constraint mechanism. By simultaneously establishing the evolution relationship of charging stages, the coupling relationship of functional components, and the energy convergence consistency relationship, it achieves joint modeling of latent faults and pseudo-convergence faults under complex operating conditions. At the same time, combined with the dynamic negative sample generation mechanism, it performs perturbation learning on the abnormal propagation relationship between different operating condition reference intervals and different charging stages, thereby improving the model's ability to identify cross-stage fault propagation and gradual fault states.

[0036] To verify the practical effect of this invention, it was compared with traditional threshold alarm methods, LSTM time-series diagnostic methods, and ordinary CPC model methods. Fault identification results were continuously statistically analyzed over 30 days. The experimental results are shown in Table 1. Table 1. Performance Comparison of Different Charging Pile Fault Diagnosis Methods

[0037] As can be seen from the comparison results in Table 1 above, the method of the present invention achieves a latent fault identification rate of 96.8%, which is 12.1% higher than the ordinary CPC model method and 18.2% higher than the LSTM timing diagnostic method; it achieves a pseudo-convergence fault identification rate of 94.5%, which is 14.3% higher than the ordinary CPC model method; and it achieves a cross-component propagation fault identification rate of 92.7%, which is 54.2% higher than the threshold alarm method. At the same time, the average false diagnosis rate is reduced to 2.3%, and the fault location time is shortened to 4.9s. This shows that the present invention can effectively improve the fault propagation identification capability and diagnostic stability under complex working conditions.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A charging pile fault diagnosis system based on charging big data, characterized in that, include: The charging operation parameter acquisition module is used to collect charging operation parameter data of the charging pile during the charging process and to build a charging session dataset. The charging phase division module is used to divide the charging session dataset into charging phases and generate a phase state vector and an energy closed-loop state vector corresponding to each charging phase. The component state mapping module is used to map the stage state vector and energy closed-loop state vector to the charging pile functional components, and generate component-level stage state vector and component-level energy closed-loop state vector. The counterfactual operating condition constraint module is used to generate counterfactual operating condition constraint data for the corresponding charging stage based on the charging operation parameter data. The stage transition prediction module is used to input the component-level stage state vector, the component-level energy closed-loop state vector, and the counterfactual operating condition constraint data into the improved CPC model, generate the predicted energy closed-loop state vector corresponding to the next charging stage, and generate stage evolution offset data. The fault source analysis module is used to generate stage convergence slope and convergence consistency entropy based on the stage evolution offset data, and to generate the fault source score of the corresponding component by combining the component-level stage state vector. The baseline update module is used to generate fault diagnosis results based on the fault source score, stage convergence slope, convergence consistency entropy and stage evolution offset data, and update the operating baseline of the corresponding component based on the component status data after repair.

2. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The charging operation parameter acquisition module includes: The system collects voltage sampling data, current sampling data, output power sampling data, module temperature sampling data, insulation resistance sampling data, vehicle state of charge data, BMS communication message data, gun wire specifications and power battery parameters during the charging session, and generates a multi-source time-series data stream according to a unified time base. The multi-source time-series data stream is time-aligned, mapping data corresponding to different sampling frequencies to the same sampling time axis, and generating time-series synchronization data based on the time interval between adjacent sampling moments. The time-series synchronization data is subjected to abnormal fluctuation screening. Transient pulse data is identified based on the change amplitude, change direction and change duration between continuous sampling points. The abnormal sampling interval is reconstructed by interpolation of adjacent time-series intervals to generate reconstructed time-series data. The reconstructed timing data is sliced ​​into running states. The charging session boundaries are divided based on the charging start signal, BMS handshake message, output power change slope, and charging end identifier, and data segments of the corresponding charging session are generated. The data segments of the charging session are processed by multi-dimensional parameter association, and the voltage parameters, current parameters, output power parameters and module temperature parameters corresponding to the same sampling time are combined in time sequence to generate a charging operation parameter matrix; The charging operation parameter matrix is ​​processed by working condition parameter association, and the vehicle state of charge data, ambient temperature data, charging cable specification parameters and power battery parameters are mapped to the corresponding sampling interval to generate a working condition association dataset. An index mapping relationship is established for the aforementioned working condition associated dataset according to the charging session number, sampling time series, and charging device number to generate a charging session dataset.

3. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The charging stage division module includes: Joint timing analysis is performed on the output power timing data, current timing data, and BMS communication message timing data in the charging session dataset. The charging stage switching node is identified based on the output power change slope, current change slope, BMS handshake establishment flag, constant voltage request flag, and charging stop flag. The charging session dataset is divided into stages based on the charging stage switching nodes, generating handshake stage data segments, pre-charge stage data segments, constant current stage data segments, constant voltage stage data segments, power reduction stage data segments, and stop stage data segments. The average value, fluctuation amplitude, and rate of change of voltage parameters, current parameters, output power parameters, module temperature parameters, and insulation resistance parameters in each stage data segment are calculated to generate the corresponding stage operation parameter sequence for the charging stage. The time window is divided into the sequence of operating parameters for the stage, and the parameters are arranged in time sequence according to the magnitude, direction and rate of change of the parameters within the continuous sampling period to generate the stage state vector corresponding to the charging stage. The output power parameter, current parameter, and module temperature parameter in the stage state vector are correlated and calculated to generate the power convergence sequence, current convergence sequence, and thermal convergence sequence for the corresponding sampling period. The power convergence sequence, current convergence sequence, and thermal convergence sequence are time-coupled, and a stage energy correlation matrix is ​​established based on the convergence direction, convergence speed, and convergence amplitude between consecutive sampling periods. The stage energy correlation matrix is ​​processed by stage mapping, and the stage energy correlation matrices corresponding to each charging stage are combined in chronological order to generate the energy closed-loop state vector of the corresponding charging stage.

4. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The component state mapping module includes: The voltage parameters, current parameters, output power parameters, module temperature parameters, insulation resistance parameters, and BMS communication message parameters in the stage state vector are analyzed for parameter change synchronicity, parameter change delay, and parameter change amplitude correlation, and a parameter change relationship matrix is ​​established. Based on the parameter coupling relationship in the parameter change relationship matrix, the voltage parameter, output power parameter and current parameter are mapped to the power conversion functional component, the module temperature parameter is mapped to the thermal management functional component, the insulation resistance parameter is mapped to the insulation detection functional component, and the BMS communication message parameter is mapped to the communication functional component, thus generating a component parameter mapping sequence. The component parameter mapping sequence is divided into time windows, and a component state correlation matrix is ​​established based on the parameter change amplitude, parameter change direction and parameter change rate within the same time window; Statistical processing is performed on the number of synchronous changes, the duration of delayed changes, and the difference in the magnitude of changes in the component state correlation matrix to establish a component state change sequence; The component state change sequence is subjected to component aggregation processing, and the parameter change sequence associated with the corresponding functional component is combined according to the sampling time order to generate a component-level stage state vector. Based on the parameter change relationship matrix and component state correlation matrix, the power convergence sequence, current convergence sequence and thermal convergence sequence in the energy closed-loop state vector are processed by component mapping to establish energy convergence correlation relationships corresponding to different functional components. The energy convergence correlations of different functional components are combined in a time series to generate a component-level energy closed-loop state vector.

5. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The counterfactual operating condition constraint module includes: The vehicle state of charge data, ambient temperature data, output power data, charging cable specifications, and power battery parameters in the charging operation parameter data are used to extract operating condition parameters and establish an operating condition parameter sequence. The operating condition parameter sequence is divided into time windows. An operating condition reference interval is established based on the vehicle charge state change interval, ambient temperature change interval, and output power change interval within the same time window, and a corresponding operating condition change matrix is ​​generated. The parameter matching process is performed on the gun wire specification parameters, power battery parameters and output power variation range in the operating condition change matrix to establish the operating condition coupling relationship between different operating condition parameters. Based on the correspondence between the charging stage switching nodes and the time window, the operating condition coupling relationship is processed by stage mapping, and different operating condition parameters are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage, generating a stage operating condition association sequence. The frequency, amplitude, and duration of fluctuations in the ambient temperature change range, output power change range, and vehicle state of charge change range in the stage operating condition correlation sequence are statistically processed to establish an operating condition change trend sequence. The time-series correlation processing is performed on the operating condition change trend sequence, and an operating condition constraint correlation matrix is ​​established based on the change direction, change amplitude and change duration between consecutive time windows; The working condition constraint correlation matrix is ​​processed by stage combination, and the stage working condition constraint relationship is established based on the working condition change trend within the same working condition reference interval, generating counterfactual working condition constraint data for the corresponding charging stage.

6. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The stage transition prediction module includes: The component-level stage state vector, the component-level energy closed-loop state vector, and the counterfactual operating condition constraint data are input into the improved CPC model. The improved CPC model includes a stage transition coding layer, a dual-path context coding layer, a counterfactual operating condition constraint layer, a convergence residual enhancement layer, a closed-loop consistency constraint layer, and a dynamic negative sample generation layer. The stage transition coding layer encodes the parameter change sequence corresponding to different charging stages in a stage sequence, and establishes a stage transition coding sequence based on the parameter change direction, parameter change rate and parameter change amplitude between adjacent charging stages. The dual-path context coding layer includes a time evolution coding path and a component association coding path. The time evolution coding path establishes a time evolution context sequence based on the continuous relationship of parameter changes between continuous charging stages. The component association coding path establishes a component association context sequence based on the synchronous change relationship and the delayed change relationship of parameters between different functional components. The temporal evolution context sequence and the component-associated context sequence are processed by vector concatenation and unified feature dimension mapping to generate a dual-path context feature sequence; The counterfactual operating condition constraint layer performs operating condition constraint mapping processing on the dual-path context feature sequence based on the operating condition change trend within the same operating condition reference interval, and generates an operating condition constraint feature sequence. The dual-path context feature sequence and the working condition constraint feature sequence are jointly encoded to generate a stage context encoding sequence; The convergence residual enhancement layer predicts the power convergence sequence, current convergence sequence and thermal convergence sequence corresponding to the next charging stage, establishes the predicted energy closed loop state sequence, and establishes the convergence residual enhancement sequence based on the residual change direction, residual change rate and residual change amplitude between the predicted energy closed loop state sequence and the actual energy closed loop state sequence corresponding to the next charging stage. The closed-loop consistency constraint layer establishes a closed-loop consistency constraint sequence based on the consistency of convergence direction, convergence rate, and convergence magnitude among the power convergence sequence, current convergence sequence, and thermal convergence sequence, and performs consistency correction processing on the stage context encoding sequence based on the closed-loop consistency constraint sequence. The dynamic negative sample generation layer performs stage misalignment combination processing on the stage context encoding sequences corresponding to different working condition reference intervals to establish a working condition disturbance negative sample sequence. The negative sample sequence of the working condition disturbance, the convergence residual enhancement sequence, the closed-loop consistency constraint sequence and the stage context encoding sequence are compared and predicted to generate the predicted energy closed-loop state vector corresponding to the next charging stage. The energy closed-loop evolution offset data is generated based on the predicted energy closed-loop state vector, the energy closed-loop state vector corresponding to the actual next charging stage, the closed-loop consistency constraint sequence, and the convergence residual enhancement sequence.

7. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The fault source analysis module includes: The power offset sequence, current offset sequence, and thermal offset sequence in the stage energy closed-loop evolution offset data are divided into time windows to establish a stage offset parameter sequence. The offset change analysis is performed on the continuous time windows in the stage offset parameter sequence, and a stage convergence sequence is established based on the offset change direction, offset change magnitude and offset change rate between adjacent time windows. The slope of the continuous offset change relationship in the stage convergence sequence is calculated, and the stage convergence slope is generated based on the correspondence between the offset change amplitude and the time interval between continuous time windows. Synchronous change relationship analysis is performed on the power offset sequence, current offset sequence and thermal offset sequence, and a closed-loop consistent correlation sequence is established based on the consistency of change direction, change rate and change amplitude among different offset sequences. Statistical processing is performed on the distribution relationship of different offset sequences in the closed-loop consistency association sequence, and a convergence consistency entropy is generated based on the distribution frequency, distribution dispersion and change direction of different offset sequences. Component correlation analysis is performed on the parameter change sequence and stage energy closed-loop evolution offset data in the component-level stage state vector to establish state correlation sequences and offset propagation correlation sequences corresponding to different functional components; The stage convergence slope, convergence consistency entropy, state correlation sequence, and offset propagation correlation sequence are jointly correlated and processed to generate the fault source score of the corresponding functional component based on the offset change intensity, offset duration, and offset propagation direction of the different functional components.

8. The charging pile fault diagnosis system based on charging big data according to claim 1, characterized in that, The baseline update module includes: Time window alignment processing is performed on the fault source score, stage convergence slope, convergence consistency entropy and stage energy closed-loop evolution offset data to establish a fault diagnosis parameter sequence. Correlation analysis is performed on the fault source score, stage convergence slope and convergence consistency entropy in the fault diagnosis parameter sequence, and a fault evolution trend sequence is established based on the change direction, change amplitude and change duration between continuous time windows. Propagation relationship analysis was performed on the power offset sequence, current offset sequence and thermal offset sequence in the stage energy closed-loop evolution offset data. Fault propagation correlation sequence was established based on the temporal sequence relationship and offset change correlation relationship between different offset sequences. The fault evolution trend sequence and fault propagation correlation sequence are processed to construct propagation paths, and fault propagation path sequences are established based on the offset propagation direction, offset propagation duration and offset propagation intensity corresponding to different functional components. The continuous offset propagation relationships in the fault propagation path sequence are processed by stage mapping, and different offset propagation relationships are mapped to the handshake stage, pre-charge stage, constant current stage, constant voltage stage, power reduction stage and stop stage to generate a stage fault association sequence. The corresponding relationships of different functional components in the stage fault association sequence are statistically processed, and the fault stage, fault propagation path and fault diagnosis result of the corresponding functional component are generated based on the fault source score, stage convergence slope and convergence consistency entropy. Collect the status data of the repaired components, perform offset correlation analysis between the status data of the repaired components and the historical normal operation status data, perform baseline offset correction processing based on the difference in the change of corresponding parameters before and after the repair, and update the operating baseline of the corresponding functional components.

Images