Direct current charging pile detection system and method applied to field

Abstract

The invention discloses a direct current charging pile detection system and method applied to the field, and relates to the technical field of power equipment on-line monitoring, and the method comprises the steps: inputting a three-dimensional data set into a Toeplitz cyclic measurement matrix for compressed sampling, reconstructing the three-dimensional data set into a high-resolution signal through employing an improved orthogonal matching pursuit algorithm, and carrying out the detection of the high-resolution signal; analyzing and calculating a local variable coefficient through a sliding window, marking abnormal positions, and integrating the abnormal positions into an abnormal position set; inputting the abnormal position set into a pre-trained long and short term memory prediction model, evaluating the insulation residual life and the fault risk level, and calculating the maximum allowable output current of the DC charging pile in real time; the load impedance is dynamically adjusted and detected through the memristor array, the output current of the direct current charging pile is collected in real time, and synchronous compression wavelet transform is adopted to analyze a harmonic ridge line and extract a harmonic component; the accuracy of fault early warning is remarkably improved, rapid dynamic adjustment of the output current is achieved, and the technical problem of multi-physics field coupling monitoring of the rapid charging pile is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online monitoring of power equipment, and in particular to a DC charging pile detection system and method applied to a field. BACKGROUND

[0002] In the technical field of DC charging piles, real-time state monitoring and fault early warning are core links to ensure charging safety and equipment reliability. Distributed optical fiber sensors (DOFS) are used to synchronously collect the strain and temperature of the charging pile output cable, and an abnormal alarm is realized through a threshold triggering mechanism. By laying multiple sensing nodes, a two-dimensional temperature-strain distribution model of the cable surface is constructed, and the preliminary positioning of the fault area is realized in combination with the preset safety threshold. In addition, some improved schemes introduce Fourier transform or short-time window STFT to analyze the current harmonic components to evaluate the power quality of the charging pile output. This kind of method performs well in steady-state conditions, especially in the scene where a single physical quantity (such as temperature) suddenly changes.

[0003] The existing technology still has limitations in analyzing the coupling of multiple physical fields under dynamic conditions. The two-dimensional data model generated by the distributed sensing system is difficult to represent the three-dimensional degradation trajectory of the cable insulation layer under complex stress-thermal cycle, which leads to the local micro-strain accumulation or gradient anomaly (such as non-uniform deformation caused by mechanical vibration) being easily ignored by the threshold algorithm. In addition, the traditional harmonic analysis method is limited by the frequency domain resolution of the fixed basis function, and it is difficult to capture the correlation between transient harmonic components and insulation aging. This limitation is particularly significant in the high current density scenario of fast charging piles, and there is an urgent need for a detection method that can integrate multi-dimensional dynamic features and quantify the insulation degradation risk. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a DC charging pile detection method applied to a field to solve the problem of difficult accurate evaluation of cable insulation layer degradation risk.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a DC charging pile detection method applied to a field, which comprises collecting the strain gradient and temperature gradient of the DC charging pile output cable, and preprocessing to generate a three-dimensional data set;

[0008] The three-dimensional data set is input into a Toeplitz cyclic measurement matrix for compressed sampling, and an improved orthogonal matching pursuit algorithm is used to reconstruct a high-resolution signal. The local coefficient of variation is calculated by sliding window analysis, the abnormal position is marked, and the abnormal position set is integrated;

[0009] The abnormal position set is input into a pre-trained long short-term memory prediction model, insulation residual life and fault risk level are evaluated, and the maximum allowable output current of the direct current charging pile is calculated in real time;

[0010] The load impedance is dynamically adjusted through the memristor array, the output current of the direct current charging pile is collected in real time, and the harmonic ridge line is analyzed and the harmonic component is extracted by using synchronous compression wavelet transform;

[0011] The abnormal position set, the fault risk level and the harmonic component are fused, the power distribution network health index is output, and the optimal network reconstruction scheme is generated by quantum annealing algorithm.

[0012] As a preferred scheme of the direct current charging pile detection method applied to the field of the application, wherein: the strain gradient and the temperature gradient are time-aligned by using the GPS timestamp, and the noise interference is eliminated by using adaptive Kalman filtering and median filtering respectively, the filtered strain gradient and temperature gradient are interpolated to a unified space-time grid according to spatial position, time slice and physical quantity type, and a three-dimensional data set is generated.

[0013] As a preferred scheme of the direct current charging pile detection method applied to the field of the application, wherein: the three-dimensional data set is input into a Toeplitz cyclic measurement matrix for compressed sampling, and an improved orthogonal matching pursuit algorithm is used to reconstruct a high-resolution signal, the specific steps are as follows,

[0014] The three-dimensional data set is divided into a plurality of sub-blocks according to spatial position, time slice and physical quantity type;

[0015] Based on each sub-block, the first row elements are generated by random Gaussian distribution, and the Toeplitz cyclic measurement matrix is generated row by row through cyclic shift operation, the compression sampling rate is dynamically adjusted according to the local variance of the sub-block, and the compressed observation value is generated;

[0016] For the compressed observation value of each sub-block, the signal is reconstructed based on the Toeplitz cyclic measurement matrix and the improved orthogonal matching pursuit algorithm, and the noise interference is suppressed by dynamically adjusting the regularization parameter, and the sub-block reconstruction signal is generated;

[0017] The sub-block reconstruction signals are weighted and fused according to spatial position, the residual compensation is performed on the sub-block reconstruction signals in the overlapping area, and the reconstruction result is iteratively corrected, and the high-resolution signal is output.

[0018] As a preferred scheme of the direct current charging pile detection method applied to the field of the application, wherein: the local variation coefficient is calculated by sliding window analysis, the abnormal position is marked, and the abnormal position set is integrated, the specific steps are as follows,

[0019] The sliding window of the high-resolution signal is defined according to a spatial position dimension and a time slice dimension, the local variation coefficient of each sliding window in the time slice dimension and the spatial position dimension is calculated respectively, and the time and space coupling is performed according to a preset local variance fusion weight to generate a time and space coupling variation coefficient distribution map;

[0020] An abnormality judgment threshold is dynamically generated according to the time and space coupling variation coefficient distribution map, and the center point of the sliding window exceeding the abnormality judgment threshold is marked as a candidate abnormal point;

[0021] The morphological closing operation is performed on the marked candidate abnormal point to eliminate the spatial isolated point and fill the time continuous abnormal area, and the denoised abnormality map is generated;

[0022] The abnormal point in the denoised abnormality map is mapped to a time and space coordinate system, and the adjacent abnormal points are merged by using a time and space proximity density clustering algorithm, and an abnormal position set representing the abnormal position of the cable is output.

[0023] As a preferred scheme of the application, the method for detecting the DC charging pile applied in the field comprises the following steps,

[0024] The spatial distribution parameter, the time duration parameter, the maximum strain gradient and the average temperature gradient of each abnormal position of the cable are extracted from the abnormal position set, and a time and space feature vector sequence is generated;

[0025] The time series prediction is performed on the time and space feature vector sequence by using the pre-trained long short-term memory prediction model, the insulation remaining life prediction value is output, and the fault risk level is generated based on the mapping relationship between the insulation remaining life prediction value and the preset resistance decay rate threshold;

[0026] The insulation remaining life prediction value and the fault risk level are taken as state inputs, the policy iteration optimization is performed by using the reinforcement learning network, the maximum allowable output current adjustment amount is output, and the maximum allowable output current of the DC charging pile is output by linear superposition to the current reference current value.

[0027] As a preferred scheme of the application, the method for detecting the DC charging pile applied in the field comprises the following steps,

[0028] The deviation of the DC charging pile maximum allowable output current from the real-time output current is calculated, the dynamic target impedance value is calculated by using the Ohm's law, the duty cycle control parameter in the memristor array is decomposed by using the pulse coding algorithm, and the impedance matching control signal is generated.

[0029] The total resistance value of the memristor array is adjusted based on an impedance matching control signal, a direct current charging pile output current is collected in real time, and a synchronous compression wavelet transform analysis is performed, a harmonic ridge line concentrated in time-frequency energy is extracted, and the harmonic ridge line is segmented by K-means energy density clustering to obtain a harmonic component.

[0030] As a preferred scheme of the direct current charging pile detection method applied to the field of the application, the specific steps of generating the optimal network reconstruction scheme are as follows,

[0031] The spatial density distribution of each cable abnormal position in the abnormal position set is calculated, the fault risk level is normalized and encoded, and the harmonic components are weighted and summed according to the harmonic order, fused by a preset health degree evaluation fusion weight, and a power distribution network health degree index is generated;

[0032] The power distribution network health degree index is taken as a risk quantization item, and the minimum network loss is taken as an optimization target, and a quantum annealing optimization problem is constructed by combining topological constraints and voltage safety constraints;

[0033] The quantum annealing optimization problem satisfying the topological constraints and the voltage safety constraints is solved by the quantum annealing algorithm to generate the optimal network reconstruction scheme.

[0034] In the second aspect, the application provides a direct current charging pile detection system applied to the field, which comprises a data acquisition module for acquiring and preprocessing strain gradients and temperature gradients of a direct current charging pile output cable to generate a three-dimensional data set;

[0035] An anomaly detection module is configured to input the three-dimensional data set into a Toeplitz cyclic measurement matrix for compressed sampling, reconstruct a high-resolution signal by using an improved orthogonal matching pursuit algorithm, calculate a local variation coefficient by using a sliding window analysis, mark abnormal positions, and integrate the abnormal positions into an abnormal position set;

[0036] A health prediction module is configured to input the abnormal position set into a pre-trained long short-term memory prediction model, evaluate the insulation remaining life and the fault risk level, and calculate a maximum allowable output current of the direct current charging pile in real time;

[0037] A harmonic analysis module is configured to dynamically adjust a detection load impedance by using a memristor array, collect a direct current charging pile output current in real time, and analyze a harmonic ridge line and extract a harmonic component by using a synchronous compression wavelet transform;

[0038] A scheme generation module is configured to fuse the abnormal position set, the fault risk level, and the harmonic component, output a power distribution network health degree index, and generate an optimal network reconstruction scheme by using a quantum annealing algorithm.

[0039] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the method for detecting a direct-current charging pile in a field according to the first aspect of the present application.

[0040] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any step of the method for detecting a direct-current charging pile in a field according to the first aspect of the present application.

[0041] The present application has the following beneficial effects: through the compression sensing reconstruction of the three-dimensional strain-temperature gradient field and the dynamic impedance adjustment technology of the memristor, high-precision real-time monitoring of the insulation state of the direct-current charging pile is achieved. The Toeplitz compression sampling and the improved orthogonal matching pursuit algorithm are used to significantly reduce the data volume and greatly improve the micro-strain detection sensitivity; combined with the fast impedance matching of the memristor array and the synchronous compression wavelet transform, high-precision harmonic feature extraction is achieved; finally, through the LSTM life prediction and quantum annealing optimization, an intelligent closed-loop system is formed, which significantly improves the accuracy of fault warning and realizes the rapid dynamic adjustment of the output current, effectively solving the technical problems of multi-physical field coupling monitoring of fast charging piles. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Fig. 1 Flowchart of the method for detecting a direct-current charging pile in a field.

[0044] Fig. 2 Schematic diagram of the direct-current charging pile detection system applied in a field.

[0045] Fig. 3 Flowchart of the compression sampling and signal reconstruction process.

[0046] Fig. 4 Flowchart of the power distribution network health index generation and network reconstruction. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.

[0048] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0049] It will be apparent that systems and / or methods, described herein, can be implemented in various forms of hardware, software, firmware, special purpose processors, or a combination of these. The present disclosure has been described in relation to particular embodiments, which are intended to be illustrative only and not restrictive. The scope of the application is not limited to the exemplary embodiments set forth herein, but is only limited by the claims. Numerous modifications and adaptations will be apparent. Accordingly, the patent is not limited by the foregoing description, but is only limited by the scope of the patent claims.

[0050] Reference will now be made to specific implementations, examples of which are illustrated in the accompanying drawings. The following description will present these examples in the context of a method for detecting a direct current charging pile in the field. However, the method described herein can be applied to other fields. Figs. 1-4 For one embodiment of the present application, the embodiment provides a method for detecting a direct current charging pile in the field, comprising the following steps:

[0051] S1, collect the strain gradient and temperature gradient of the direct current charging pile output cable, and preprocess to generate a three-dimensional data set.

[0052] The strain gradient and temperature gradient are time-aligned using a GPS timestamp, and adaptive Kalman filtering and median filtering are used to eliminate noise interference. The filtered strain gradient and temperature gradient are interpolated to a unified space-time grid according to spatial position, time slice, and physical quantity type, and a three-dimensional data set is generated.

[0053] It should be noted that the original data of the strain gradient (example sampling rate 1 kHz) and the temperature gradient (example sampling rate 100 Hz) are recorded synchronously by the PPS signal trigger data acquisition card, and the time axes of the two are unified to the GPS second pulse clock reference, with a time alignment error controlled within 1 ms;

[0054] For the strain gradient, adaptive Kalman filtering is used, and the process noise covariance matrix is dynamically updated according to the local variance calculated by the sliding window (example window length 50), and the measurement noise covariance matrix is fixed as the sensor calibration value. For the temperature gradient, 5x5 neighborhood median filtering is used to eliminate impulse noise. The filtered strain gradient and temperature gradient are resampled to a unified grid point by cubic spline interpolation to fill in missing values according to three dimensions of spatial coordinates (example interval 1 cm), time slice (example interval 10 ms), and physical quantity type (strain gradient or temperature gradient), and finally a three-dimensional data set containing spatial position, timestamp, and physical quantity value is generated;

[0055] It should also be noted that the measurement noise covariance matrix directly uses the sensor factory calibration value (for example, the measurement noise variance of the strain sensor is calibrated to 1x10 -4 με 2where με represents micro-strain), the sensor factory calibration value is kept fixed in the filtering process; the process noise covariance matrix is obtained by real-time calculation of the local statistical properties of the strain gradient signal.

[0056] S2, input the three-dimensional data set into the Toeplitz cyclic measurement matrix for compressed sampling, and reconstruct into a high-resolution signal by using an improved orthogonal matching pursuit algorithm, calculate the local coefficient of variation by sliding window analysis, mark the abnormal position, and integrate into an abnormal position set.

[0057] The three-dimensional data set is divided into multiple sub-blocks according to spatial position, time slice and physical quantity type.

[0058] It should be noted that the spatial position dimension is segmented with a fixed size (for example, a rectangular grid of 10 cm x 10 cm), the time slice dimension is divided according to a fixed time length (for example, 100 ms is a time slice unit), and the physical quantity type dimension is kept independent of strain gradient and temperature gradient;

[0059] For the generated three-dimensional data set containing spatial coordinates, time stamps and physical quantity values, 10 cm x 10 cm spatial regions are sequentially intercepted along the x-y plane coordinate axes, and all three-dimensional data points within a 100 ms time window are extracted in each spatial region and classified into strain gradient sub-blocks and temperature gradient sub-blocks. A sliding window method with an overlap rate of 20% (example spatial overlap 2 cm, time overlap 20 ms) is used to ensure boundary continuity, and each sub-block data is stored as a five-dimensional tensor structure (x coordinate, y coordinate, time point, strain gradient value, temperature gradient value), wherein the missing data points are filled with nearest neighbor interpolation. The final output is a sub-block set named according to the spatial grid number-time slice number-physical quantity type.

[0060] Based on each sub-block, the first row elements are generated by random Gaussian distribution, and the Toeplitz cyclic measurement matrix is generated by row-by-row cyclic shift operation, and the compression sampling rate is dynamically adjusted according to the local variance of the sub-block to generate compressed observation values.

[0061] It should be noted that the three-dimensional data (spatial position, time slice, physical quantity type) in the sub-block is expanded into a one-dimensional vector. The sliding window method is used to move and calculate on the vector, and the window size is fixed (for example, 5 x 5 units of space combined with 20 time slices to form a window of 500 three-dimensional data points), and the window is gradually slid with an overlap rate of 50%. In each window position, the Welford online algorithm is used to calculate the unbiased variance of the sub-block sliding window (example: 5 x 5 space units x 20 time slices) in real time, and after traversing the entire sub-block vector, the arithmetic mean of all window calculated variance values is taken as the final local variance of the sub-block;

[0062] The first row elements are randomly generated with a zero-mean Gaussian distribution, and the number of elements is equal to the sub-block dimension d (for example, 1000 independent and identically distributed N(0, 1 / d) random numbers are generated when the sub-block is expanded into a 1000-dimensional vector), a Toeplitz cyclic measurement matrix is constructed by a cyclic right shift operation, specifically, the first row elements are shifted right by k-1 positions to generate the kth row elements (k is from 1 to the number of rows m of the measurement matrix), an mxd-dimensional Toeplitz cyclic measurement matrix is formed, the compression sampling rate is dynamically adjusted, and finally the Toeplitz cyclic measurement matrix is multiplied by the vectorized sub-block data to output the compressed observation value.

[0063] It should also be noted that the basic compression sampling rate is set according to the relationship between the measurement dimension and the signal sparsity in the compression sensing theory, and the compression sampling rate is dynamically adjusted according to the local variance of the sub-block, and the expression is:

[0064] η=1-0.5log2(1+σ 2 / σ 2 );

[0065] Wherein, η is the compression sampling rate (value range 0~1), σ 2 is the local variance of the sub-block, and σ 2 is the preset noise variance.

[0066] The preset noise variance is usually determined in advance based on hardware parameters or empirical data, and is used for noise calibration in the algorithm.

[0067] For each sub-block compression observation value, the signal is reconstructed based on the Toeplitz cyclic measurement matrix and the improved orthogonal matching pursuit algorithm, and the noise interference is suppressed by dynamically adjusting the regularization parameter to generate the sub-block reconstruction signal.

[0068] It should be noted that the Toeplitz cyclic measurement matrix is combined with the improved orthogonal matching pursuit algorithm to gradually reconstruct the signal by iteratively selecting atoms and calculating the residual. In each iteration, the improved orthogonal matching pursuit algorithm calculates the inner product of the current residual and each column of the Toeplitz cyclic measurement matrix, selects the column with the largest absolute value of the inner product as the new atom of the support set, updates the signal estimate value by least squares method and recalculates the residual. The regularization parameter is dynamically adjusted according to the ratio of the local variance of the sub-block to the preset noise variance, for example, when the local variance of the sub-block is less than the preset noise variance, the regularization parameter is increased to enhance noise suppression. The iteration termination condition is set to the residual norm reaching a preset maximum iteration number (for example, 100 times), and finally the sub-block reconstruction signal is output.

[0069] All sub-block reconstruction signals are fused according to the spatial position, the residual of the sub-block reconstruction signal in the overlapping area is compensated and the reconstruction result is iteratively corrected, and the high-resolution signal is output.

[0070] It should be noted that the spatial weight mapping table is established according to the spatial position coordinates of the sub-block reconstructed signals, and the weight value is inversely proportional to the distance from the center point of the sub-block to the fusion point (the distance from the center point of the sub-block to the fusion point is calculated by the existing three-dimensional Euclidean distance method). For multiple sub-block reconstructed signals in the overlapping area, a weighted average algorithm is used for preliminary fusion, and the weight coefficient is determined by the spatial weight mapping. For the fusion boundary area, the residual error vector of the adjacent sub-block reconstructed signals is calculated, and the residual error compensation coefficient is solved by the least square method, and the sub-block reconstructed signals in the overlapping area are compensated and corrected. The fusion result is optimized in an iterative manner, the residual error compensation coefficient is updated and the weighted fusion value is recalculated each time, and the mean square error of the adjacent sub-block reconstructed signals in the overlapping area reaches the maximum iteration times (for example, 10 times), and the finally output high-resolution signal;

[0071] It should also be noted that the center point spatial coordinates (such as three-dimensional coordinates (x, y, z)) of each sub-block reconstructed signal are obtained, the weight of the target fusion point and each sub-block center point is calculated using a Gaussian kernel function, all related sub-block weights are normalized, and finally a spatial weight mapping table containing normalized weights and corresponding coordinate indexes is generated.

[0072] The sliding window of the high-resolution signal is defined in the spatial position dimension and the time slice dimension, the local variation coefficient of each sliding window in the time slice dimension and the spatial position dimension is calculated respectively, and the spatio-temporal coupling is performed according to the preset local variance fusion weight to generate a spatio-temporal coupling variation coefficient distribution map.

[0073] It should be noted that a rectangular window (for example, 5x5 units) is set in the spatial position dimension, a fixed length window (for example, 20 slices) is set in the time slice dimension, and the window slides on the three-dimensional data body with an overlap rate of 50%. In each window position, the local variation coefficient of the time slice dimension (time slice standard deviation divided by time slice mean) and the local variation coefficient of the spatial position dimension (spatial position standard deviation divided by spatial position mean) are calculated respectively;

[0074] The local variation coefficients of the two dimensions are weighted and summed according to the preset local variance fusion weight (for example, the time dimension weight is 0.6 and the spatial dimension weight is 0.4) to obtain the spatio-temporal coupling variation coefficient. After traversing the complete data body, the spatio-temporal coupling variation coefficients of all window positions are arranged according to the spatial coordinates and time stamps to generate a spatio-temporal coupling variation coefficient distribution map;

[0075] It should be noted that the preset local variance fusion weight is determined by analyzing the signal stability in the time slice dimension and the spatial position dimension, and a larger fluctuation dimension is assigned a lower weight (for example, when the time dimension fluctuates significantly, the time weight is taken as 0.3-0.5); secondly, referring to the sensor accuracy parameter, the weight of the dimension with higher measurement noise is correspondingly reduced (for example, the strain gradient space measurement noise variance is calibrated as 1x10 -4με 2 When the spatial weight is increased to 0.6-0.7), the weight combination that optimizes the signal-to-noise ratio of the spatiotemporal coupling variation coefficient distribution map is selected through experimental verification.

[0076] The anomaly determination threshold is dynamically generated according to the distribution diagram of spatiotemporal coupling variation coefficient, and the center point of the sliding window exceeding the anomaly determination threshold is marked as a candidate anomaly point.

[0077] It should be explained that the global mean and global standard deviation of the spatiotemporal coupling coefficient of variation distribution map are calculated by arithmetic mean and standard deviation, and the global mean plus three times the global standard deviation is used as the initial anomaly judgment threshold; a three-dimensional sliding window statistical analysis is performed on the spatiotemporal coupling coefficient of variation distribution map, and the window size is set to three time slices multiplied by three spatial position units multiplied by three spatial position units. The local mean and local standard deviation of the spatiotemporal coupling coefficient of variation within each window range are calculated. When the coefficient value of the window center point (the value of the spatiotemporal coupling coefficient of variation corresponding to the center position in the spatiotemporal area covered by the sliding window) exceeds the local mean plus two times the local standard deviation, the center point is marked as a candidate anomaly point.

[0078] A morphological closing operation is performed on the marked candidate anomaly points to eliminate spatial isolated points and fill the temporal continuous anomaly areas to generate a denoised anomaly map.

[0079] It should be explained that the occurrence of candidate anomaly points at each spatial position is checked in the time slice dimension. When the same spatial coordinate is marked as a candidate anomaly point in multiple consecutive time slices (for example, 3 slices), it is determined that there is a time-continuous anomaly in this area and it is marked as a time-continuous anomaly area. A circular structure element (for example, a radius of 5 cm) is defined in the spatial position dimension, and an expansion operation is performed on all candidate anomaly points. Isolated points whose spatial distance is less than the diameter of the structure element are merged into connected areas; then an erosion operation is performed to restore the original anomaly area boundary but eliminate unconnected isolated points. A linear structure element (for example, a length of 3 consecutive slices) is defined in the time slice dimension, and expansion and erosion operations are performed on the anomaly point sequence at each spatial position to fill the short-interrupted anomaly time segments, and finally a denoised anomaly map is generated.

[0080] The abnormal points in the denoised anomaly map are mapped to the spatiotemporal coordinate system, and the spatiotemporal proximity density clustering algorithm is used to merge adjacent abnormal points to output an abnormal location set representing the abnormal location of the cable.

[0081] It should be noted that the denoised anomaly map after morphological closing operation processing is read, wherein the position with pixel value of 1 represents a potential abnormal point; the three-dimensional data structure of the entire anomaly map is scanned, and each spatial position is checked on the marked state of each time slice, and the coordinates of all pixel points marked as 1 are recorded; the complete space-time coordinate information of these abnormal points is extracted, including spatial position (x, y, z) and corresponding time slice number; finally, the coordinate information is matched and verified with the original cable position mapping table to obtain the abnormal points in the denoised anomaly map;

[0082] According to the spatial coordinates and time slice information recorded in each abnormal point in the denoised anomaly map, the accurate position of the abnormal point is reconstructed in the three-dimensional space-time coordinate system, a space-time clustering algorithm based on density is used, a spatial neighborhood radius (for example, 10 cm) and a time neighborhood range (for example, 5 continuous slices) are set as clustering parameters, and adjacent abnormal points that meet the density connection condition are merged; in the clustering process, the spatial dimension uses Euclidean distance to calculate the proximity of abnormal points, and the time dimension uses the number of slice intervals to calculate the continuity. When the abnormal point meets the neighborhood condition in spatial distance and time interval, it is merged into the same abnormal cluster. The final output abnormal position set contains the center coordinates, time span and spatial coverage range information of each abnormal cluster, wherein the spatial coverage range is determined by the minimum bounding rectangle of all abnormal points in the cluster, and the time span is determined by the earliest and latest time slice of the abnormal points in the cluster.

[0083] S3, input the abnormal position set into a pre-trained long short-term memory prediction model, evaluate the insulation remaining life and fault risk level, and calculate the maximum allowed output current of the direct current charging pile in real time.

[0084] From the abnormal position set, the spatial distribution parameters, time duration parameters, maximum strain gradient and average temperature gradient of each cable abnormal position are extracted to generate a space-time feature vector sequence.

[0085] It should be noted that for each cable abnormal position in the abnormal position set, the spatial distribution parameters include the abnormal region center coordinates, the spatial coverage range length, width and height (the abnormal region center coordinates and the spatial coverage range length, width and height are calculated by the minimum bounding rectangle algorithm); the time duration parameters include the start time slice, the end time slice and the total number of time slices; the strain gradient data of all time slices of the abnormal position in the duration are queried from the original three-dimensional data set, and the maximum strain gradient value is extracted; the average temperature gradient value of the spatial region corresponding to the abnormal position in the duration is calculated;

[0086] The four-dimensional characteristics of the spatial distribution parameter, the time duration parameter, the maximum strain gradient, and the average temperature gradient are arranged in a fixed order (the fixed order is: the spatial distribution parameter (center coordinates, length, width, and height), the time duration parameter (start and end slices, duration), the maximum strain gradient, and the average temperature gradient) to form a space-time feature vector representing the position of a single cable anomaly. The above operation is repeated for all anomaly positions, and the space-time feature vectors of each anomaly position are arranged in chronological order to ultimately form a complete space-time feature vector sequence.

[0087] The space-time feature vector sequence is input into the pre-trained long short-term memory prediction model for time series prediction, and an insulation remaining life prediction value is output. A fault risk level is generated based on the mapping relationship between the insulation remaining life prediction value and the preset resistance decay rate threshold.

[0088] It should be noted that the space-time feature vector sequence is input into the long short-term memory prediction model in chronological order, and the feature vector is processed step by step in time by the memory unit and the gating mechanism, and a sequence of insulation remaining life prediction values is output. Subsequently, a mapping table of insulation remaining life prediction values and resistance decay rates is established (for example, an insulation remaining life prediction value of 2000 hours corresponds to a resistance decay rate of 5%, and 1000 hours corresponds to a resistance decay rate of 10%), and a risk level is divided according to a preset resistance decay rate threshold (for example, a resistance decay rate < 8% is low risk, 8%-15% is medium risk, and > 15% is high risk). Finally, an evaluation report containing the insulation remaining life prediction value and the corresponding fault risk level is generated.

[0089] It should also be noted that the training process of the long short-term memory prediction model uses historical cable monitoring data, with the space-time feature vector sequence as input and the measured insulation remaining life value as label. The prediction error is minimized by the Adam optimizer, the number of training iterations is set to 200 rounds, the batch size is set to 32, and the optimal long short-term memory prediction model parameters are determined by 5-fold cross-validation.

[0090] The insulation remaining life prediction value and the fault risk level are input as states, and the reinforcement learning network is used for policy iteration optimization to output the maximum allowed output current adjustment amount, which is linearly superimposed on the current reference current value to output the maximum allowed output current of the DC charging pile.

[0091] It should be noted that the insulation remaining life prediction value and the fault risk level form a two-dimensional state vector, which is input into the pre-trained reinforcement learning network. The reinforcement learning network outputs a continuous action value, which is the maximum allowed output current adjustment amount, with an adjustment range limited to the example value ±50A interval. The rated output current value (for example, 250A) calibrated on the charging pile nameplate is used as a basic reference value. The output current adjustment amount is algebraically added to the current reference current value, and the expression is:

[0092] I max = Ibase +ΔI;

[0093] Among them, I max is the maximum allowable output current of the DC charging pile (unit: ampere), I base is the current baseline current value (unit: ampere), ΔI is the current adjustment amount (unit: ampere) output by the deep deterministic policy gradient network, ΔI∈[-50,+50];

[0094] It should also be explained that the state space of the reinforcement learning network consists of the predicted remaining insulation life and the fault risk level, the action space is the current adjustment, and the reward function is designed as a weighted sum of the cable safety factor and the charging efficiency, with the weight coefficients taken as examples of 0.7 and 0.3, respectively. The reinforcement learning network is trained using an experience replay mechanism, sampling state-action-reward samples from historical operating data for batch training. The reinforcement learning network parameters are optimized using a policy gradient algorithm. During training, a target reinforcement learning network and a soft update strategy are used to ensure stability. The discount factor is set to the example value of 0.9, and the policy network learning rate is set to the example value of 0.001. Each training round is updated 10,000 times.

[0095] S4. Dynamically adjust the load impedance through the memristor array to collect the output current of the DC charging pile in real time, and use synchronous compression wavelet transform to analyze the harmonic ridges and extract the harmonic components.

[0096] The deviation from the real-time output current is calculated based on the maximum allowable output current of the DC charging pile, and the dynamic target impedance value is calculated using Ohm's law. The pulse coding algorithm is used to decompose it into the duty cycle control parameters in the memristor array to generate an impedance matching control signal.

[0097] It should be noted that the real-time output current is obtained by real-time sampling through a high-precision Hall current sensor connected in series at the output end of the DC charging pile, and the deviation from the maximum allowable output current of the DC charging pile is calculated. The expression is:

[0098] δI=|I max -I real ∣;

[0099] Where, δI is the current deviation (unit: A), I max is the maximum allowable output current of the DC charging pile (unit: ampere), I real is the real-time output current (unit: A);

[0100] The dynamic target impedance value is calculated using Ohm's law, and the expression is:

[0101]

[0102] Among them, Z target is the dynamic target impedance value, Vout is the output voltage of the direct current charging pile, which is measured in real time by a voltage sensor (for example, the accuracy is ±0.2%);

[0103] A mapping relationship table of the dynamic target impedance value and the duty cycle parameter is established, the table is obtained by experiment calibration, and contains 256 discrete levels (for example, the impedance value of 10Ω corresponds to the duty cycle of 50%); the dynamic target impedance value is quantitatively processed by using a pulse coding algorithm, the closest discrete level is matched according to the mapping relationship table, and a corresponding duty cycle parameter is output; the duty cycle parameter is converted into a pulse width modulation signal, the signal frequency is set to 1kHz, and the pulse width is proportional to the duty cycle parameter; and finally an impedance matching control signal is generated.

[0104] The total resistance value of the memristor array is adjusted based on the impedance matching control signal, the output current of the direct current charging pile is collected in real time, and synchronous compression wavelet transform analysis is performed, a harmonic ridge line concentrated in time-frequency energy is extracted, and K-means energy density clustering segmentation is performed on the harmonic ridge line to obtain harmonic components.

[0105] It should be noted that the real-time resistance value (example value 1kΩ-100kΩ) of each memristor unit in the memristor array under a set control voltage (for example, 1V) is measured, the admittance values (reciprocals of the resistance values) of all parallel units are added, and the reciprocal of the total resistance value of the memristor array is obtained;

[0106] The total resistance value of the memristor array is adjusted according to the impedance matching control signal, so that the output loop impedance of the direct current charging pile matches the dynamic target impedance value; the adjusted output current signal is collected synchronously, the sampling frequency is set to 10kHz, db4 wavelet base function is used for 5-layer synchronous compression wavelet transform, and a time-frequency energy distribution matrix is obtained; the continuous time-frequency points with energy exceeding an energy threshold (for example, 2 times the overall energy mean value) in the time-frequency energy distribution matrix are extracted to form a harmonic ridge line; K-means clustering analysis is performed on the extracted harmonic ridge line, the clustering number K is set to 3 (corresponding to the fundamental wave, the 3rd harmonic wave and the 5th harmonic wave), the time-frequency point energy value is used as a characteristic quantity for clustering segmentation, and finally the time-frequency distribution area of each harmonic component and the energy proportion thereof are output;

[0107] It should also be noted that the output current signal of the direct current charging pile under normal working conditions is collected, time-frequency energy distribution data is extracted through synchronous compression wavelet transform, the global mean value and the standard deviation of the energy of all time-frequency points are calculated, and the energy threshold is set to the global mean value plus twice the standard deviation (for example, when the global mean value is 50dB, the energy threshold is set to 70dB).

[0108] S5, fusion of the abnormal position set, the fault risk level and the harmonic component, output of the power distribution network health index, and generation of an optimal network reconstruction scheme through a quantum annealing algorithm.

[0109] The spatial density distribution of each cable anomaly position in the abnormal position set is calculated, the fault risk level is normalized and coded, and the harmonic components are weighted and summed according to the harmonic number, fused through the preset health degree evaluation fusion weight, and the power distribution network health degree index is generated.

[0110] It should be noted that the straight-line distance between the center points of adjacent cables (for example, the distance between cable A and cable B is 2 meters) is directly measured on the power distribution network topographic map as the physical distance d, the kernel function bandwidth is set to d / 2 (example value: when d = 2 meters, h = 1 meter), the Gaussian kernel function is applied to each cable anomaly position, the kernel function contribution of all abnormal positions is superimposed, the kernel density estimation algorithm is used to calculate the spatial density distribution value of the cable anomaly position, and the expression is:

[0111]

[0112] wherein, is the spatial density distribution value at the cable anomaly position coordinate (x, y), n is the total number of cable anomaly positions, h is the kernel function bandwidth (unit: meters), (x i ,y i ) is the two-dimensional spatial coordinate of the i-th cable anomaly position, i is the index variable of the cable anomaly position, K is the Gaussian kernel function, and (x, y) is the cable anomaly position coordinate;

[0113] The cable anomaly position coordinates are extracted from the denoised anomaly map by a three-dimensional time-space clustering algorithm (DBSCAN variant), and the specific centroid coordinates (space x, y, z) and time slice range of each anomaly cluster are obtained.

[0114] The fault risk level is normalized and coded according to the International Electrotechnical Commission insulation evaluation standard, the high-risk level is coded as a higher value corresponding to the significant insulation resistance drop range, the medium-risk level is coded as an intermediate value corresponding to the moderate insulation resistance drop range, and the low-risk level is coded as a lower value corresponding to the slight insulation resistance drop range. Secondly, the frequency domain weighting processing is performed on the harmonic components, the fundamental component is assigned the maximum weight, the third harmonic component is assigned the medium weight, and the fifth harmonic component is assigned the smaller weight. The weight distribution is based on the International Electrotechnical Commission harmonic standard for the effects of each harmonic. Finally, the health degree evaluation fusion weight coefficient determined by the expert consensus method is used for linear weighted fusion, wherein the cable anomaly spatial density distribution value is assigned the maximum weight reflecting the defect aggregation degree, the risk level coding value is assigned the medium weight reflecting the insulation deterioration degree, and the harmonic weighted sum is assigned the smaller weight reflecting the current distortion influence. The power distribution network health degree index in the standard range is calculated by weighted summation.

[0115] The distribution network health index is used as a risk quantification item, and together with minimizing network loss, it constitutes the optimization goal. Combined with topological constraints and voltage safety constraints, a quantum annealing optimization problem is constructed.

[0116] It should be explained that the distribution network health index is used as a risk quantification item, and together with minimizing network losses, it constitutes a multi-objective optimization function. The linear weighting method is used to transform the dual objectives into a single objective, and the weight coefficient is determined according to the degree of risk preference. Secondly, topological constraints are established, including node connectivity constraints, branch capacity constraints, and radial operation constraints. At the same time, voltage safety constraints are set, requiring that the voltage deviation of each node does not exceed the allowable fluctuation range of the rated voltage. The above objective function and constraints are mapped into the Hamiltonian of the quantum annealing problem, where the objective function is converted into an energy term and the hard constraints are implemented through the penalty term. The final quantum annealing optimization problem is formed.

[0117] It should also be noted that topology constraints are implemented by establishing a node-branch association matrix, which requires that three conditions be met: network connectivity constraints, ensuring that all load nodes are connected to power nodes; branch capacity constraints, ensuring that the power of each line does not exceed its maximum transmission capacity; radial operation constraints, using a depth-first search algorithm to verify that the network is loop-free;

[0118] Voltage safety constraints are implemented by setting upper and lower limits for node voltages, requiring the voltage amplitude of each node to be maintained within a specified percentage of the rated voltage (e.g., ±5%), and the Newton-Raphson method is used for power flow calculation verification.

[0119] The quantum annealing algorithm is used to solve the quantum annealing optimization problem that satisfies topological constraints and voltage safety constraints, and the optimal network reconstruction plan is generated.

[0120] It should be explained that the distribution network reconstruction problem is mapped into a quantum annealing model, and the network switch state is represented by quantum bits, the closed state is encoded as 1, and the open state is encoded as 0; secondly, a Hamiltonian containing objective functions and constraints is constructed. The objective function is composed of minimizing network losses and optimizing the health of the distribution network. The constraints include topological connectivity constraints, branch capacity constraints, and voltage safety constraints; the annealing operation is performed through a quantum annealing processor, the initial temperature is set to a high value, and the temperature is gradually lowered to make the quantum bit state converge to the optimal solution; finally, the quantum bit state is decoded to obtain the switching operation plan, and the optimal network reconstruction plan that meets all constraints is generated.

[0121] This embodiment also provides a DC charging pile detection system for on-site application, comprising: a data acquisition module for collecting strain gradients and temperature gradients of the DC charging pile output cable, and performing preprocessing to generate a three-dimensional data set;

[0122] An anomaly detection module is configured to input a three-dimensional data set into a Toeplitz cyclic measurement matrix for compressed sampling, and reconstruct a high-resolution signal by using an improved orthogonal matching pursuit algorithm, calculate a local coefficient of variation by using a sliding window analysis, mark an abnormal position, and integrate the abnormal position into an abnormal position set.

[0123] A health prediction module is configured to input the abnormal position set into a pre-trained long short-term memory prediction model, evaluate an insulation remaining life and a fault risk level, and calculate a maximum allowable output current of a direct-current charging pile in real time.

[0124] A harmonic analysis module is configured to dynamically adjust a detection load impedance by using a memristor array, collect an output current of the direct-current charging pile in real time, and analyze a harmonic ridge line and extract a harmonic component by using a synchronous compressive wavelet transform.

[0125] A scheme generation module is configured to fuse the abnormal position set, the fault risk level, and the harmonic component, output a power distribution network health index, and generate an optimal network reconstruction scheme by using a quantum annealing algorithm.

[0126] The embodiment also provides a computer device suitable for the direct-current charging pile detection method applied to a field, including a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the direct-current charging pile detection method applied to the field proposed in the above embodiment.

[0127] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (near field communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, a trackball, or a touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, a touchpad, or a mouse, etc.

[0128] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for detecting a direct-current charging pile applied in a field as proposed in the above embodiment; and the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0129] To sum up, the application realizes high-precision real-time monitoring of the insulation state of a direct-current charging pile by means of compression sensing reconstruction of a three-dimensional strain-temperature gradient field and dynamic impedance adjustment technology of a memristor. The use of Toeplitz compression sampling and an improved orthogonal matching pursuit algorithm significantly reduces the amount of data and greatly improves the micro-strain detection sensitivity; in combination with the rapid impedance matching of a memristor array and synchronous compression wavelet transform, high-precision harmonic feature extraction is realized; finally, an intelligent closed-loop system is formed through LSTM life prediction and quantum annealing optimization, which significantly improves the accuracy of fault warning and realizes rapid dynamic adjustment of the output current, effectively solving the technical problem of multi-physical field coupling monitoring of a fast-charging pile.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the application but not limit the application. Although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the application, and all of them should be covered in the scope of the claims of the application.

Claims

1. A DC charging pile detection method applied on site, characterized by: include, Collect the strain gradient and temperature gradient of the DC charging pile output cable, perform preprocessing, and generate a three-dimensional data set; The 3D data set is input into the Toeplitz cycle measurement matrix for compressive sampling and reconstructed into a high-resolution signal using an improved orthogonal matching pursuit algorithm. The local coefficient of variation is calculated through sliding window analysis, abnormal locations are marked, and the abnormal locations are integrated into a set. The abnormal location set is input into a pre-trained long-short-term memory prediction model to evaluate the remaining insulation life and fault risk level, and calculate the maximum allowable output current of the DC charging pile in real time; The load impedance is dynamically adjusted by a memristor array to detect the output current of the DC charging pile in real time. Synchronous compression wavelet transform is used to analyze the harmonic ridges and extract the harmonic components. By integrating the abnormal location set, fault risk level and harmonic components, the distribution network health index is output, and the optimal network reconstruction plan is generated through the quantum annealing algorithm.

2. The method for detecting a DC charging pile on site according to claim 1, wherein: The strain gradient and temperature gradient are time-aligned using GPS timestamps, and adaptive Kalman filtering and median filtering are used to eliminate noise interference, respectively. The filtered strain gradient and temperature gradient are interpolated to a unified space-time grid according to spatial position, time slice and physical quantity type to generate a three-dimensional data set.

3. The method for detecting a DC charging pile on site according to claim 2, wherein: The three-dimensional data set is input into the Toeplitz cycle measurement matrix for compression sampling, and the improved orthogonal matching pursuit algorithm is used to reconstruct it into a high-resolution signal. The specific steps are as follows: Divide the three-dimensional data set into multiple sub-blocks according to spatial location, time slice and physical quantity type; Based on each sub-block, the first row elements are generated through random Gaussian distribution, and the Toeplitz cyclic measurement matrix is ​​generated row by row through cyclic shift operation. The compressed sampling rate is dynamically adjusted according to the local variance of the sub-block to generate compressed observation values. For each sub-block's compressed observation value, the signal is reconstructed based on the Toeplitz cyclic measurement matrix and the improved orthogonal matching pursuit algorithm. The noise interference is suppressed by dynamically adjusting the regularization parameter to generate the sub-block reconstructed signal. All sub-block reconstructed signals are weighted and fused according to spatial position, residual compensation is performed on the sub-block reconstructed signals in the overlapping area, and the reconstruction results are iteratively corrected to output a high-resolution signal.

4. The method for detecting a DC charging pile on site according to claim 3, wherein: The sliding window analysis is used to calculate the local coefficient of variation, mark abnormal locations, and integrate them into an abnormal location set. The specific steps are as follows: Define the sliding window of the high-resolution signal according to the spatial position dimension and the time slice dimension, calculate the local variation coefficient of each sliding window in the time slice dimension and the spatial position dimension respectively, and perform spatiotemporal coupling according to the preset local variance fusion weight to generate a spatiotemporal coupling variation coefficient distribution map; Dynamically generate anomaly determination thresholds based on the spatiotemporal coupling variation coefficient distribution map, and mark the center points of the sliding window that exceed the anomaly determination thresholds as candidate anomaly points; Perform morphological closing operations on the marked candidate anomaly points to eliminate spatial isolated points and fill in the temporal continuous anomaly areas to generate a denoised anomaly map; The abnormal points in the denoised anomaly map are mapped to the spatiotemporal coordinate system, and the spatiotemporal proximity density clustering algorithm is used to merge adjacent abnormal points to output an abnormal location set representing the abnormal location of the cable.

5. The method for detecting a DC charging pile on site according to claim 1, wherein: The specific steps of real-time calculation of the maximum allowable output current of the DC charging pile are as follows: Extract the spatial distribution parameters, time duration parameters, maximum strain gradient and average temperature gradient of each cable abnormal position from the abnormal position set to generate a spatiotemporal feature vector sequence; Through the pre-trained long-short-term memory prediction model, time series prediction is performed on the spatiotemporal feature vector sequence, and the predicted value of the remaining insulation life is output. The fault risk level is generated based on the mapping relationship between the predicted value of the remaining insulation life and the preset resistance decay rate threshold; The predicted value of the remaining insulation life and the fault risk level are used as state inputs, and the strategy is iteratively optimized through the reinforcement learning network to output the maximum allowable output current adjustment amount. This is then linearly superimposed on the current reference current value to output the maximum allowable output current of the DC charging pile.

6. The method for detecting a DC charging pile on site according to claim 1, wherein: The specific steps of extracting harmonic components are as follows: The deviation from the real-time output current is calculated based on the maximum allowable output current of the DC charging pile, and the dynamic target impedance value is calculated using Ohm's law. This is then decomposed into the duty cycle control parameters in the memristor array using a pulse coding algorithm to generate an impedance matching control signal. The total resistance of the memristor array is adjusted based on the impedance matching control signal, and the output current of the DC charging pile is collected in real time. Synchronous compression wavelet transform analysis is performed to extract the harmonic ridges where the time-frequency energy is concentrated. The harmonic ridges are then clustered and segmented using K-means energy density to obtain the harmonic components.

7. The method for detecting a DC charging pile on site according to claim 1, wherein: The specific steps of generating the optimal network reconstruction solution are as follows: The spatial density distribution of each abnormal cable location in the abnormal location set is calculated, the fault risk level is normalized and encoded, and the harmonic components are weighted and summed according to the harmonic order. The results are then fused using the preset health assessment fusion weights to generate the distribution network health index. The distribution network health index is used as a risk quantification item, and together with minimizing network loss, it constitutes the optimization goal. Combined with topological constraints and voltage safety constraints, a quantum annealing optimization problem is constructed. The quantum annealing algorithm is used to solve the quantum annealing optimization problem that satisfies topological constraints and voltage safety constraints, and the optimal network reconstruction plan is generated.

8. A DC charging pile detection system for on-site use, based on the DC charging pile detection method for on-site use according to any one of claims 1 to 7, characterized in that: include, The data acquisition module is used to collect the strain gradient and temperature gradient of the DC charging pile output cable, perform preprocessing, and generate a three-dimensional data set; The anomaly detection module is used to input the 3D data set into the Toeplitz cycle measurement matrix for compression sampling, and reconstruct it into a high-resolution signal using an improved orthogonal matching pursuit algorithm. The local coefficient of variation is calculated through sliding window analysis, anomaly locations are marked, and the anomaly locations are integrated into a set. The health prediction module is used to input the abnormal location set into the pre-trained long-short-term memory prediction model to evaluate the remaining insulation life and fault risk level, and calculate the maximum allowable output current of the DC charging pile in real time; The harmonic analysis module is used to dynamically adjust the load impedance through the memristor array, collect the output current of the DC charging pile in real time, and use synchronous compression wavelet transform to analyze the harmonic ridges and extract the harmonic components; The solution generation module integrates the abnormal location set, fault risk level and harmonic components, outputs the distribution network health index, and generates the optimal network reconstruction solution through the quantum annealing algorithm.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the on-site DC charging pile detection method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the on-site DC charging pile detection method according to any one of claims 1 to 7 are implemented.

Images