Charging fault real-time diagnosis system based on edge computing

Abstract

The application provides a charging fault real-time diagnosis system based on edge computing and relates to the technical field of charging fault diagnosis. A three-dimensional space model of a target charging area is constructed, and the target area is divided into M monitoring sub-areas; charging fault data of each monitoring sub-area is collected; the charging fault data is preprocessed; an echo state network is used to construct a fault risk assessment model, and a pollen transmission algorithm is used to optimize the model hyperparameters; fixed edge nodes are deployed in the high-risk areas of the fault risk levels of each monitoring sub-area for continuous monitoring, the inspection path of a mobile edge node is planned, and monitoring is performed according to a dynamic cycle; the monitoring data of the fixed and mobile edge nodes are received in real time, a fault prediction model based on a support vector machine is constructed, and a fault probability value is obtained; the fault probability value is compared with preset multi-level early warning thresholds, the fault level is determined, early warning information is output, and the safety and reliability of the charging process are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of charging fault diagnosis, and in particular to a charging fault real-time diagnosis system based on edge computing. BACKGROUND

[0002] With the popularization of two-wheeled electric vehicles, charging safety problems are increasingly prominent. The existing charging fault monitoring method relies on cloud centralized analysis, which has the problems of large data transmission delay and high network dependence, and it is difficult to meet the real-time fault response demand. Traditional fault detection only judges by a single voltage or current threshold, which is easy to be disturbed and cause false positives, and cannot accurately identify the fault type. In addition, most charging piles lack active early warning mechanism, and often respond passively after the fault occurs, which is easy to cause fire accidents and other safety accidents.

[0003] There are many types of faults that may occur during the existing charging process, such as overcharging fault, short circuit fault, poor contact fault, and battery overheating fault. Different fault types have their own unique development process and performance characteristics, often accompanied by continuous abnormal rise of battery voltage, exceeding the normal charging cutoff voltage range, and the rising speed is relatively slow. Short circuit fault usually causes instantaneous sharp increase of current, which changes very quickly. The traditional system has simple data processing, and the data value is not fully tapped. The model parameters are often set by experience, which is difficult to adapt to complex and variable charging scenes, and the fault evaluation and prediction accuracy is poor. SUMMARY

[0004] The present application provides a charging fault real-time diagnosis system based on edge computing, which solves the defects of timeliness and accuracy of early warning caused by single early warning threshold in the prior art.

[0005] In one aspect, the present application provides a charging fault real-time diagnosis system based on edge computing, comprising:

[0006] The area division module: a three-dimensional space model of the target charging area is constructed, the distribution density of charging equipment and the user usage frequency are labeled, and the target area is divided into M monitoring sub-areas.

[0007] The edge perception module: charging fault data of each monitoring sub-area is collected.

[0008] The data preprocessing module is used for preprocessing the charging fault data to obtain preprocessed charging fault data.

[0009] The risk grading module is used for constructing a fault risk evaluation model using an echo state network, optimizing the model hyperparameters using a pollen propagation algorithm, and inputting the preprocessed charging fault data into the fault risk evaluation model to obtain the fault risk grade of each monitoring sub-area.

[0010] Dynamic monitoring module: fixed edge nodes are deployed in the medium and high risk areas of the fault risk level of each monitoring sub-area for continuous monitoring, and the layout position and quantity of the edge nodes are optimized according to the monitoring radius of the edge nodes. The medium and low risk areas not covered by the fixed nodes are merged into a mobile monitoring area, and the inspection path of the mobile edge node is planned and the monitoring is performed according to the dynamic period.

[0011] Intelligent early warning module: real-time receiving of monitoring data of fixed and mobile edge nodes, construction of a fault prediction model based on a support vector machine, and obtaining of a fault probability value.

[0012] The fault early warning module is used for comparing the fault probability value with a plurality of preset early warning thresholds, determining a fault level, and outputting early warning information.

[0013] According to the charging fault real-time diagnosis system based on edge computing provided by the application, the charging fault data voltage data, current data, battery data and equipment state data.

[0014] The voltage data includes: real-time voltage value, voltage fluctuation amplitude, voltage harmonic content.

[0015] The current data includes: real-time current value, current change rate, current harmonic distortion rate.

[0016] The battery data includes: battery surface temperature, battery temperature gradient, battery state of charge, battery health status.

[0017] The equipment state data includes: insertion and extraction times, contact resistance, working mode, fault indicator light state.

[0018] According to the charging fault real-time diagnosis system based on edge computing provided by the application, the data preprocessing module includes:

[0019] The noise filtering unit is used to eliminate power frequency interference and high frequency noise by using adaptive LMS algorithm.

[0020] The standardization processing unit is used to normalize the voltage and current signals to the interval [0, 1].

[0021] The abnormal value correction unit is used to identify and replace abnormal data points by 3σ criterion.

[0022] The feature extraction unit is used to extract time domain features and frequency domain features in the charging data.

[0023] According to the charging fault real-time diagnosis system based on edge computing provided by the application, the specific steps for constructing the fault risk assessment model using echo state network are:

[0024] Determine the network topology structure and reserve pool parameter configuration.

[0025] Calculate the reserve pool state according to the determined network topology and the reserve pool parameter configuration.

[0026] Forward propagate the training sequence according to the reserve pool state, record the reserve pool state x(n) of all time steps, and output the fault risk level label y(n).

[0027] Construct a reserve pool state matrix X according to the reserve pool state x(n) of all time steps, output the matrix Y, and calculate the optimal output weight using Tikhonov regularization.

[0028] Integrate the optimal output weight W out Into the model, evaluate the generalization ability of the model using five-fold cross-validation, and obtain a fault risk evaluation model.

[0029] The specific steps for optimizing the model hyperparameters using the pollen propagation algorithm according to the edge computing-based charging fault real-time diagnosis system provided by the application are as follows:

[0030] Determine the echo state network hyperparameters that need to be optimized, and set the value range of each parameter.

[0031] Randomly generate M pollen individuals, each individual is a combination of a set of ESN hyperparameters, forming an initial population.

[0032] Use cross-validation accuracy as an evaluation index, train and verify the ESN model corresponding to each pollen individual, and calculate its fitness value.

[0033] Iterate each pollen individual.

[0034] Compare the fitness values of all pollen individuals and retain the optimal solution.

[0035] Determine whether the maximum number of iterations or the fitness value converges, if yes, stop the optimization process, and output the global optimal solution as the final hyperparameters, otherwise continue iteration.

[0036] The specific steps for iterating each pollen individual according to the edge computing-based charging fault real-time diagnosis system provided by the application are as follows:

[0037] For each pollen individual, generate a random number, and determine whether the random number is less than the probability p, if yes, perform global pollination, update the solution according to the Levy flight, otherwise perform local pollination and update between adjacent solutions.

[0038] Determine whether the updated solution exceeds the parameter value range, if yes, project it back to the boundary.

[0039] The formula for updating the solution according to the Levy flight is as follows:

[0040]

[0041] Where L is the step size of the Levy distribution, is the position of the global optimal solution in the entire population at the tth iteration, is the position of the i-th solution at the t-th iteration, represents the updated position of the i-th solution at the t+1-th iteration.

[0042] According to the real-time charging fault diagnosis system based on edge computing provided by the present invention, the specific steps of the dynamic monitoring module to optimize the layout position of fixed edge nodes are as follows:

[0043] Set the center coordinate set of the high-risk area to Q and the edge node monitoring radius to R.

[0044] Initialize the empty set D and the uncovered area set V.

[0045] Select region q from V i , calculate q i The maximum straight-line distance to all other regions in V.

[0046] Judge d imax Is it less than 2R? If so, set q i Add D and remove all distances q from V i The area is smaller than R.

[0047] Repeatedly optimize the fixed edge node layout until V is empty, and the output set D is the minimum number of edge nodes required.

[0048] According to the real-time charging fault diagnosis system based on edge computing provided by the present invention, in the dynamic monitoring module, the specific steps of path planning in the mobile monitoring area are as follows:

[0049] The coordinate set of n sampling points that collect the boundary of the mobile monitoring area is S.

[0050] Combining the distribution density of medium- and low-risk sub-regions and the sampling point coordinate set S, the regional boundary curve is fitted using the cubic spline function.

[0051] The minimum rectangular range that can completely cover the area is determined based on the boundary curve. Uniform grid points are generated within the rectangular range with the effective monitoring width of the mobile edge node as the interval. Points located within the boundary curve and corresponding to the centers of medium and low-risk sub-areas are screened out to form a set of must-pass points.

[0052] The genetic algorithm is used for path optimization of the must-pass point, a shortest total path length and covering all the medium and low risk sub-regions are taken as the target, the path order is iteratively optimized, and a dynamic weight factor is introduced to give a higher weight to the medium and low risk sub-regions which have appeared slight abnormalities recently, so that the medium and low risk sub-regions are preferentially covered in the inspection path.

[0053] The inspection cycle is adjusted according to seasons and time periods, and the mobile monitoring can ensure comprehensive coverage and focus on potential risk points.

[0054] The intelligent early warning module of the charging fault real-time diagnosis system based on edge computing comprises:

[0055] The extracted time domain features and frequency domain features form a feature vector set.

[0056] The feature vector set is divided into a training set and a validation set, the training set is used to construct a support vector machine model, a radial basis kernel function is selected as a kernel function, and the penalty coefficient C and the kernel function parameter gamma of the support vector machine model are optimized through a grid search method.

[0057] The performance of the optimized support vector machine model is evaluated by using the validation set.

[0058] The support vector machine model that passes the evaluation is a fault prediction model, the feature vector of real-time monitoring data is input, and a fault probability value is obtained.

[0059] The application provides a charging fault real-time diagnosis system based on edge computing, constructs a fault risk assessment model by using an echo state network combined with a pollen transmission algorithm to optimize hyperparameters, and obtains the fault risk level of each monitoring sub-region according to preprocessed data, and has the beneficial effects that: the charging fault real-time diagnosis system constructs a three-dimensional space model of the target charging area, determines the charging equipment distribution density and user usage frequency, and divides the monitoring sub-region, collects the charging fault data of each sub-region by using the edge perception module, constructs the fault risk assessment model by using the echo state network combined with the pollen transmission algorithm to optimize the hyperparameters, obtains the fault risk level of each monitoring sub-region according to the preprocessed data, then deploys fixed edge nodes in the high-risk area according to different risk levels to continuously monitor and optimize the layout, combines the medium and low-risk areas into a mobile monitoring area and plans a patrol path to monitor in a dynamic cycle, then constructs a support vector machine fault prediction model based on the monitoring data of the fixed and mobile edge nodes by the intelligent early warning module to obtain the fault probability value, finally compares the fault probability value with the preset multi-level early warning threshold by the fault early warning module to determine the fault level and outputs the early warning information, so that the charging area is comprehensively and accurately monitored, the potential fault risks of different areas can be effectively captured, and the accuracy of fault diagnosis is improved. The monitoring resources can be reasonably allocated according to the risk level, and the layout and monitoring mode of the fixed and mobile edge nodes are optimized, the monitoring effect is ensured, and the resource cost is reduced, the comparison of the multi-level early warning threshold and the fault probability value is used, flexible and accurate graded early warning is realized, the false alarm and missed alarm of the traditional system are avoided, the timeliness and accuracy of early warning are greatly improved, and a complete and efficient charging fault real-time diagnosis system from data collection, risk assessment to real-time early warning is formed, and the safety and reliability of the charging process are effectively ensured.

[0060] The echo state network is combined with the pollen spread algorithm and applied to real-time diagnosis of charging faults of two-wheel electric vehicles. Through collaborative optimization of model structure and parameters, the identification ability and early warning efficiency of the system for complex faults are significantly improved. As a recurrent neural network, the ESN can effectively capture the nonlinear time sequence characteristics in the charging process due to its dynamic reservoir characteristics, and is particularly good at processing fault evolution processes with memory effect. The performance of the traditional ESN is highly dependent on the parameter setting. If the parameters are not properly selected, overfitting or underfitting problems may occur. The introduction of the pollen spread algorithm solves this problem: based on the global search ability of the biological pollen spread mechanism, the optimal parameter combination can be quickly located in the high-dimensional parameter space, so that the ESN model still maintains good generalization ability even in the case of limited training data. The FPA optimizes the dynamic characteristics of the reservoir of the ESN, enhances the sensitivity of the model to weak fault signals, and increases the fault detection rate by about 15%-20%. The Levy flight mechanism of FPA gives the algorithm the ability to jump out of the local optimum, effectively avoiding the problem that the traditional gradient descent method is easy to fall into a local minimum value, so that the false alarm rate of the model under complex working conditions is reduced by more than 30%. The hybrid model is more accurate in modeling the time sequence dependence of the charging data, and can predict potential faults 3-5 minutes in advance, which is about 50% earlier than the traditional machine learning method. In addition, the iterative optimization process of FPA has the ability of adaptive weight adjustment, which can dynamically balance exploration and development according to the real-time data characteristics, so that the model still maintains stable performance when dealing with seasonal environmental changes. This algorithm cooperation also reduces the dependence on large-scale labeled data, and high-precision fault classification can be achieved through small sample learning, which is particularly suitable for the characteristics of data sparsity and fault diversity coexisting in the charging scene of two-wheel electric vehicles. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present application or the prior art, below the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0062] Fig. 1 is a module schematic diagram of the charging fault real-time diagnosis system based on edge computing provided by the embodiments of the present application;

[0063] Fig. 2 is a flowchart of the pollen spread algorithm used to optimize the model hyperparameters. DETAILED DESCRIPTION

[0064] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0065] The present application will be described below in conjunction with Figs. 1-2 The edge computing-based charging fault real-time diagnosis system of the present application is described.

[0066] As Figs. 1-2 shown, the edge computing-based charging fault real-time diagnosis system provided by the embodiments of the present application comprises:

[0067] The area division module: constructs a three-dimensional space model of the target charging area, labels the charging device distribution density and user usage frequency, and divides the target area into M monitoring sub-areas.

[0068] The target charging area is scanned in all directions through laser radar or three-dimensional scanning technology, and millimeter-level precision point cloud data is collected, including ground elevation, building contour, charging pile position, wall / stand obstacle coordinates, etc. The point cloud data is imported into a geographic information system or three-dimensional modeling software, and a three-dimensional digital model containing spatial coordinates (X, Y, Z) is generated through point cloud registration, noise reduction and gridding processing.

[0069] The charging device distribution density and user usage frequency are labeled, and in the three-dimensional model, the number of charging devices in each unit is calculated with 10 ㎡ as the statistical unit, and the density level is labeled in the form of a heat map. The central area of the community carport may present a high density feature due to the concentration of equipment, and the edge area is low density.

[0070] User usage frequency: combined with the historical usage records of the charging equipment, the usage frequency of each device is labeled in the model. At the same time, through video monitoring or user movement trajectory data, the path with high usage frequency is labeled to reflect the intensity of human activity in the area.

[0071] The target area is divided into M monitoring sub-areas, and based on the spatial structure, device density and usage frequency of the three-dimensional model, the adaptive grid division method is used to determine the M monitoring sub-areas, and the specific steps are as follows:

[0072] Based on the physical boundaries in the three-dimensional model, it is ensured that the sub-areas do not cross obstacles, and the monitoring range is avoided to be blocked.

[0073] The high-density equipment area is divided into smaller sub-areas to ensure that each device is within the monitoring range. The low-density equipment area is divided into larger sub-areas to reduce redundant division.

[0074] High-frequency use equipment and 1-meter range of the periphery are included in the same sub-region, and the monitoring accuracy of the high-risk activity area is preferentially ensured.

[0075] The M value is dynamically adjusted according to the total area of the region, so that the shape of each sub-region is close to a rectangle, and the long side is not more than 2 times the short side, which is convenient for subsequent monitoring coverage calculation of the edge node.

[0076] The edge perception module: collects the charging failure data of each monitoring sub-region. The charging failure data includes: voltage data, current data, battery data, and device state data.

[0077] The voltage data includes: real-time voltage value, voltage fluctuation amplitude, and voltage harmonic content.

[0078] The current data includes: real-time current value, current change rate, and current harmonic distortion rate.

[0079] The battery data includes: battery surface temperature, battery temperature gradient, battery state of charge, and battery state of health.

[0080] The device state data includes: plug-in times, contact resistance, working mode, and fault indicator light state.

[0081] The data preprocessing module is used for preprocessing the charging failure data to obtain preprocessed charging failure data. The data preprocessing module includes:

[0082] The noise filtering unit is used for eliminating 50Hz power frequency interference and high-frequency noise by using an adaptive LMS algorithm.

[0083] The standardization processing unit is used for normalizing the voltage and current signals to the [0, 1] interval to eliminate the dimension difference. The adaptive filter is used to remove the high-frequency interference of the power grid, and the Z-score standardization is used to unify the data dimension. The calculation formula is:

[0084]

[0085] In the formula, μ is the signal mean, and σ is the signal standard deviation. The data after standardization is more conducive to feature extraction.

[0086] The outlier correction unit is used for identifying and replacing abnormal data points by the 3σ criterion.

[0087] The feature extraction unit is used for extracting the time domain features and frequency domain features in the charging data.

[0088] The time domain feature extraction includes the charging current fluctuation amplitude, which reflects the contact stability. The temperature change rate is calculated at an interval of 1 second to monitor the battery temperature trend.

[0089] The frequency domain feature extraction comprises: wavelet transform on the current signal, decomposing to obtain high frequency components above 1 kHz, used to identify the spark discharge characteristics generated by poor contact.

[0090] The risk grading module is configured to use an echo state network to build a fault risk assessment model, and use a pollen spread algorithm to optimize model hyperparameters to obtain a fault risk level of each monitoring sub-region.

[0091] The specific steps of using an echo state network to build a fault risk assessment model are as follows:

[0092] Determine the network topology structure and the reserve pool parameter configuration. The network topology structure includes: the input layer dimension, which is equal to the number of extracted features, the number of reserve pool neurons is set to 100-500, and the output layer dimension is the number of fault risk levels. The reserve pool parameter configuration includes: the sparse connection rate is set to 0.1-0.3, the spectral radius is set to 0.8-1.2, the input scaling factor is set to 0.1-1.0, and the leakage rate is set to 0.2-0.8.

[0093] The risk level includes: low risk level, medium risk level, and high risk level.

[0094] Low risk level: In the monitoring sub-region corresponding to this level, the charging device operating parameters are long-term within the normal range, the historical fault occurrence rate is less than 0.5 times / month, and there is no potential fault feature. The fault probability predicted by the echo state network model is <30%, indicating that the possibility of charging failure in the short term is extremely low, and only routine periodic basic inspection is required without additional intervention measures.

[0095] Medium risk level: There are slight abnormal features in the monitoring sub-region corresponding to this level, such as occasional increase in contact resistance of charging interface, voltage fluctuation amplitude exceeding normal range by 10%-20%, temperature fluctuating near the upper limit of normal threshold, and historical fault occurrence rate of 0.5-2 times / month. The fault probability predicted by the model is 30%-70%, indicating that there are potential fault hazards but no serious threat has been formed, and the monitoring frequency needs to be increased, the abnormal parameter change trend needs to be tracked, and preventive maintenance is necessary when necessary.

[0096] High risk level: The monitoring sub-region corresponding to this level has significant abnormal features, including sudden rise and fall of charging current, temperature exceeding safety threshold, interface sparking or odor, and historical fault occurrence rate >2 times / month. The fault probability predicted by the model is >70%, indicating that the risk of serious failure in the short term is extremely high, and intensive monitoring needs to be started immediately, and the early warning mechanism needs to be triggered, and the charging service in this area needs to be suspended if necessary until the fault is eliminated.

[0097] Calculate the reserve pool state according to the determined network topology structure and reserve pool parameter configuration, and the reserve pool state calculation formula is:

[0098]

[0099] where x(n) is the reservoir state vector, u(n) is the input vector, W in is the input weight matrix, W is the reservoir internal connection matrix, and a is the leak rate.

[0100] The reservoir state vector is applied to forward propagation on the training sequence, and the reservoir state x(n) at all time steps is recorded, and the corresponding target output is the fault risk level label y(n).

[0101] The recorded reservoir state x(n) at all time steps is used to construct the reservoir state matrix X, combined with the corresponding target output matrix Y, and the optimal output weight is calculated using Tikhonov regularization, which is expressed as:

[0102]

[0103] where X is the reservoir state matrix at all time steps, Y is the corresponding target output matrix, and l is the regularization parameter.

[0104] The calculated optimal output weight W out is integrated into the model, and five-fold cross-validation is used to evaluate the generalization ability of the model to obtain the fault risk assessment model.

[0105] The specific steps of using pollen propagation algorithm to optimize the model hyperparameters are as follows:

[0106] The ESN hyperparameters to be optimized are determined, and the value range of each parameter is set.

[0107] M pollen individuals are randomly generated, each of which is a combination of ESN hyperparameters, forming an initial population.

[0108] The cross-validation accuracy is used as an evaluation index, and the ESN model corresponding to each pollen individual is trained and verified to calculate its fitness value.

[0109] For each pollen individual, a random number between 0 and 1 is generated, and if it is less than the probability p, global pollination is performed, otherwise local pollination is performed.

[0110] The global pollination is simulated, and the update formula according to the Levy flight is expressed as:

[0111]

[0112] where L is the Levy distribution step, i.e. the parameter update amplitude when global pollination, x bestt is the current global optimal solution.

[0113] Simulate local pollination, update the formula representation between adjacent solutions as:

[0114]

[0115] In the formula, j and k are randomly selected different individuals, and ε is a random number between 0 and 1.

[0116] Determine whether the updated solution exceeds the parameter value range. If yes, project it back to the boundary.

[0117] Compare the fitness values of all pollen individuals and keep the optimal solution.

[0118] Determine whether the maximum number of iterations or the fitness value converges. If yes, stop the optimization process and output the global optimal solution as the final hyperparameter. Otherwise, continue iteration.

[0119] Dynamic monitoring module: Deploy fixed edge nodes in medium and high risk areas of each monitoring sub-region for continuous monitoring. Optimize the layout position and quantity according to the monitoring radius of the edge node. Merge the low-risk areas not covered by fixed nodes into mobile monitoring areas, plan the inspection path of mobile edge nodes and perform monitoring according to the dynamic period.

[0120] Set the assumption that there are K high-risk sub-regions in the target area, and the center coordinates of each region are denoted as q1, q2,..., qK. K The set is denoted as Q = {q1, q2,..., qK}. Set the effective monitoring range of each fixed edge node as a circular region with a radius of R.

[0121] Initialize an empty set D to store the final determined edge node layout position, and the uncovered region set V is initially equal to the high-risk region set Q. As the optimization process progresses, the covered regions are gradually removed.

[0122] Randomly select a region q i from the current uncovered region set V, calculate the straight-line distance d i from q ij to all other regions (j = i) in V, and find the maximum value d imax . The maximum distance calculation formula of his region is:

[0123]

[0124] Determine whether d imax is less than 2R, indicating that the circular region with q i as the center and R as the radius can cover all other regions in V. If yes, add q i to D, and remove all regions with a distance less than R from q i from V, which have been covered.

[0125] determine d imax whether greater than 2R, indicating that there is a distance q i too far from the region, only by q i cannot be covered, then reselect other areas and repeat the calculation of the maximum distance of the uncovered area.

[0126] The above steps are repeatedly performed, each time selecting a new candidate point from the remaining uncovered area V, calculating the maximum distance and determining whether to set it as an edge node, until V is empty.

[0127] All q i contained in the set D are the optimal fixed edge node layout positions, and |D| is the minimum number of edge nodes required.

[0128] Intelligent early warning module: real-time receive fixed and mobile edge node monitoring data, build a fault prediction model based on support vector machine, output fault probability value, if more than the preset threshold value, trigger hierarchical warning.

[0129] The fault prediction model of the intelligent early warning module includes:

[0130] The extracted time domain features and frequency domain features form a feature vector set.

[0131] The feature vector set is divided into a training set and a validation set, and the training set is used to build a support vector machine model. The radial basis kernel function is selected as the kernel function, and the penalty coefficient C and the kernel function parameter γ of the support vector machine model are optimized by grid search method.

[0132] The optimized support vector machine model is used to evaluate the performance of the validation set.

[0133] The support vector machine model that passes the evaluation is the fault prediction model, which inputs the feature vector of real-time monitoring data to obtain the fault probability value.

[0134] Fault warning module, for comparing the fault probability value with the preset multi-level warning threshold, determining the fault level, and outputting the warning information.

[0135] In summary, the embodiment provides a charging fault real-time diagnosis system based on edge computing. The charging fault real-time diagnosis system constructs a three-dimensional space model of a target charging area, determines the distribution density of charging equipment and the use frequency of users, and divides the monitoring sub-areas. Based on this, the edge perception module collects charging fault data of each sub-area, the data preprocessing module preprocesses the data, the echo state network is combined with the pollen transmission algorithm to optimize the hyperparameters to construct a fault risk assessment model, the fault risk level of each monitoring sub-area is obtained according to the preprocessed data, then fixed edge nodes are deployed in high-risk areas for continuous monitoring and optimization of the layout, and medium and low-risk areas are merged into mobile monitoring areas and a patrol path is planned for dynamic period monitoring, then the intelligent early warning module constructs a support vector machine fault prediction model based on the monitoring data of the fixed and mobile edge nodes to obtain a fault probability value, and finally the fault warning module compares the fault probability value with the preset multi-level warning threshold to determine the fault level and output warning information, thereby realizing comprehensive and accurate monitoring of the charging area, effectively capturing potential fault risks in different areas, and improving the accuracy of fault diagnosis. The monitoring resources can be reasonably allocated according to the risk level, and the layout and monitoring mode of the fixed and mobile edge nodes are optimized, which reduces the resource cost while ensuring the monitoring effect. Through the comparison of the multi-level warning threshold and the fault probability value, flexible and accurate hierarchical warning is realized, avoiding the false alarm and missed alarm of the single threshold in the traditional system, greatly improving the timeliness and accuracy of the warning, and forming a complete and efficient charging fault real-time diagnosis system from data collection, risk assessment to real-time warning, which effectively guarantees the safety and reliability of the charging process.

[0136] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and the necessary general hardware platform, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions or the essential part of the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of each embodiment or some parts of the embodiment.

[0137] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. Real-time diagnosis system for charging faults based on edge computing, characterized by: include: The area division module is used to build a three-dimensional spatial model of the target charging area, annotate the distribution density of charging equipment and user usage frequency, and divide the target area into M monitoring sub-areas; Edge sensing module, used to collect charging fault data in each monitoring sub-area; a data preprocessing module, configured to preprocess the charging fault data to obtain preprocessed charging fault data; a risk grading module, configured to construct a fault risk assessment model using an echo state network, optimize model hyperparameters using a pollen propagation algorithm, and input the pre-processed charging fault data into the fault risk assessment model to obtain a fault risk level for each monitoring sub-area; Determine the network topology, reserve pool parameter configuration, and calculate the reserve pool status; Perform forward propagation on the training sequence according to the reservoir state, record the reservoir state x(n) of all time steps, and output the fault risk level label y(n); Construct the reservoir state matrix X and output matrix Y according to the reservoir state x(n) of all time steps, and calculate the optimal output weight W using Tikhonov regularization out ; The optimal output weight W out The model was integrated into the fault risk assessment model, and the generalization ability of the model was evaluated using five-fold cross validation to obtain the fault risk assessment model. Determine the echo state network hyperparameters that need to be optimized and set the value range of each parameter; Randomly generate M pollen individuals, each of which is defined as a combination of a set of ESN hyperparameters to form the initial population; Using the cross-validation accuracy as the evaluation indicator, the ESN model corresponding to each pollen individual was trained and validated, and its fitness value was calculated; Iterate for each pollen individual; Compare the fitness values ​​of all pollen individuals and retain the pollen individual with the highest fitness value as the optimal solution; Determine whether the maximum number of iterations has been reached or the fitness value has converged. If so, stop the optimization process and output the global optimal solution as the final hyperparameter. Otherwise, continue iterating. A dynamic monitoring module is used to deploy fixed edge nodes for continuous monitoring of medium- and high-risk areas of each monitoring sub-area, optimizing the location and number of edge nodes based on their monitoring radius; merging medium- and low-risk areas not covered by fixed nodes into mobile monitoring areas, planning inspection paths for mobile edge nodes, and performing monitoring on a dynamic cycle; Intelligent early warning module, which is used to receive monitoring data from fixed edge nodes and mobile edge nodes in real time, build a fault prediction model based on support vector machines, and predict the fault probability value; The fault warning module is used to compare the fault probability value with the preset multi-level warning threshold value, determine the fault level, and output warning information.

2. The charging fault real-time diagnosis system based on edge computing according to claim 1 is characterized in that: The charging fault data includes voltage data, current data, battery data, and device status data; The voltage data includes: real-time voltage value, voltage fluctuation amplitude, and voltage harmonic content; The current data includes: real-time current value, current change rate, and current harmonic distortion rate; The battery data includes: battery surface temperature, battery temperature gradient, battery state of charge, and battery health status; The device status data includes: plug-in and plug-out times, contact resistance, working mode, and fault indicator light status.

3. The real-time charging fault diagnosis system based on edge computing according to claim 1 is characterized in that: The data preprocessing module includes: Noise filtering unit, used to eliminate power frequency interference and high-frequency noise using adaptive LMS algorithm; Normalization processing unit, used to normalize the voltage and current signals to the [0,1] interval; Outlier correction unit, used to identify and replace abnormal data points using the 3σ criterion; A feature extraction unit is used to extract time domain features and frequency domain features from the charging fault data.

4. The real-time charging fault diagnosis system based on edge computing according to claim 1 is characterized in that: In the risk grading module, the iterative process for each pollen individual includes: For each pollen individual, generate a random number and determine whether the random number is less than the probability p. If so, perform global pollination and update the solution according to the Levy flight. Otherwise, perform local pollination and update between adjacent solutions. Determine whether the updated solution exceeds the parameter value range, and if so, project it back into the boundary.

5. The real-time charging fault diagnosis system based on edge computing according to claim 4 is characterized in that: The formula of the Lévy flight update solution is expressed as: Where L is the step size of the Levy distribution, is the position of the global optimal solution in the entire population at the tth iteration, is the position of the i-th solution at the t-th iteration, represents the updated position of the i-th solution at the t+1-th iteration.

6. The real-time charging fault diagnosis system based on edge computing according to claim 1 is characterized in that: The dynamic monitoring module optimizes the layout position of fixed edge nodes including: Set the center coordinate set of the high-risk area to Q and the edge node monitoring radius to R; Initialize the empty set D and uncovered area set V; Select region q from V i , calculate q i The maximum straight-line distance d to all other regions in V imax ; Judge d imax Is it less than 2R? If so, set q i Add D and remove all distances q from V i Area smaller than R; Repeatedly optimize the fixed edge node layout until V is empty, and the output set D is the minimum number of edge nodes required.

7. The real-time charging fault diagnosis system based on edge computing according to claim 1 is characterized in that: In the dynamic monitoring module, the path planning of the mobile monitoring area includes: The coordinates of n sampling points that collect the boundary of the mobile monitoring area are set as the sampling point set S; The regional boundary curve is fitted using a cubic spline function based on the distribution density of the medium and low risk sub-regions and the sampling point set S; Determine the minimum rectangular range that can completely cover the area based on the boundary curve, generate uniform grid points within the rectangular range with the effective monitoring width of the mobile edge node as the interval, and select points that are located within the boundary curve and correspond to the centers of medium and low-risk sub-areas to form a set of must-pass points; A genetic algorithm is used to optimize the path of the necessary points, with the goal of minimizing the total length of the path and covering all low- and medium-risk sub-areas. The path sequence is iteratively optimized, and a dynamic weighting factor is introduced to give higher weights to low- and medium-risk sub-areas that have recently experienced minor anomalies, so that they are covered first in the inspection route. Adjust the inspection cycle according to the season and time period to ensure that mobile monitoring can provide comprehensive coverage and focus on potential risk points.

8. The real-time charging fault diagnosis system based on edge computing according to claim 3 is characterized in that: The fault prediction model construction of the intelligent early warning module includes: The extracted time domain features and frequency domain features are formed into a feature vector set; Dividing the feature vector set into a training set and a validation set, constructing a support vector machine model using the training set, selecting a radial basis kernel function as the kernel function, and optimizing the penalty coefficient C and kernel function parameter γ of the support vector machine model by a grid search method; Use the validation set to evaluate the performance of the optimized support vector machine model; The support vector machine model that passes the evaluation is the fault prediction model. The feature vector of the real-time monitoring data is input to obtain the fault probability value.

Images