A method and system for diagnosing and repairing abnormal data in low communication environments of charging stations

By combining the CNN-LSTM model and the Lagrange optimization algorithm, the problem of data anomaly diagnosis and repair in low-communication environments of charging stations was solved, achieving accurate identification and repair, and improving the intelligence level of data management and the stability of operation management of charging stations.

CN123069044APending Publication Date: 2026-11-10STATE GRID JIANGSU INTEGRATED ENERGY SERVICE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611593431.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-15
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Charging stations face problems such as data packet loss, transmission delay and signal interference in low communication environments, which affect the integrity, accuracy and consistency of data. Traditional methods have low diagnostic accuracy and high misjudgment rate, and the repaired data deviates greatly from the actual operating conditions, which cannot meet the needs of real-time data processing.

Method used

An anomaly identification is achieved by using a CNN-LSTM fusion deep learning model combined with historical data, and by using a Lagrange optimization repair algorithm to integrate physical hard constraints and communication quality soft constraints, accurate diagnosis and repair of abnormal data can be realized.

Benefits of technology

It enables accurate identification of various types of abnormal data in charging stations, adapts to the computing power requirements of low-communication environments, reduces operation and maintenance costs, improves the level of intelligent data management, and ensures the stable operation of the operation and management system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069044A_ABST
    Figure CN123069044A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for diagnosing and repairing abnormal data in low-communication environments at charging stations, relating to the field of charging station data processing technology. The method includes: segmenting standardized data into continuous time-series data segments; calculating the information entropy of each time-series data segment using an information entropy discrimination mechanism; determining a time-series data segment as an abnormal risk segment when any information entropy exceeds a preset information entropy threshold; extracting local spatial features and temporal dependencies from the abnormal risk data segments using a lightweight anomaly diagnosis model, outputting anomaly type data and diagnostic confidence; constructing an objective function using a Lagrange optimization repair algorithm, incorporating physical hard constraints and communication quality soft constraints into the optimization objective function, and iteratively solving for the optimal repair value that satisfies both physical hard constraints and communication quality soft constraints. This invention can dynamically adjust weights according to the communication environment, avoiding excessive reliance on unreliable data in repair results when communication quality is extremely poor.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of charging station data processing technology, and more specifically, to a method and system for diagnosing and repairing abnormal data in a low-communication environment of a charging station. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the construction scale and number of charging stations, as supporting infrastructure for new energy vehicles, have continued to grow. In actual operation, charging stations are mostly deployed outdoors or in remote areas, often facing the severe challenges of low-communication environments. Unstable communication links frequently lead to data packet loss, transmission delays, and signal interference, seriously affecting data integrity, accuracy, and consistency. If abnormal data cannot be diagnosed and repaired promptly and accurately, it will interfere with the charging station operation and management system's judgment of equipment status, reduce operational efficiency, and may even lead to safety accidents.

[0003] Traditional rule-matching-based anomaly diagnosis methods rely on manually preset thresholds, which are poorly adaptable to various types of data anomalies in complex scenarios, resulting in low diagnostic accuracy and a high false positive rate.

[0004] Traditional data repair methods often employ single-dimensional interpolation, failing to integrate the physical logic of the device with the business characteristics of the scenario, resulting in significant discrepancies between the repaired data and the actual operating conditions. Furthermore, most methods are not adapted and optimized for low-communication environments, leading to issues such as high computational load, high computing power requirements, and inability to resume interrupted data transmission, making it difficult to meet the real-time data processing needs of charging stations in low-communication scenarios.

[0005] Deep learning technology has powerful feature extraction and nonlinear fitting capabilities, and has outstanding advantages in the field of data anomaly diagnosis. However, existing deep learning-based solutions do not fully combine the multi-device coupling characteristics and low communication environment pain points of charging stations, and cannot achieve timely and accurate processing of abnormal data, making it difficult to support the stable operation of charging station operation and management systems.

[0006] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0007] To address the problems in related technologies, this invention proposes a method and system for diagnosing and repairing abnormal data in low communication environments of charging stations, in order to overcome the aforementioned technical problems existing in the existing related technologies.

[0008] Therefore, the specific technical solution adopted by the present invention is as follows:

[0009] In a first aspect, the present invention provides a method for diagnosing and repairing abnormal data in a low-communication environment of a charging station, comprising:

[0010] Collect multi-source operational data from various devices within the charging station and preprocess it to obtain standardized data;

[0011] Standardized data is segmented to form continuous time-series data segments. The information entropy of each time-series data segment is calculated through an information entropy discrimination mechanism. When any information entropy is greater than a preset information entropy threshold, the time-series data segment is determined to be an abnormal risk data segment.

[0012] Using a lightweight anomaly diagnosis model, local spatial features and temporal dependencies are extracted from anomaly risk data fragments, and anomaly type data and diagnostic confidence are output.

[0013] When the diagnostic confidence level is greater than or equal to the preset confidence threshold, the abnormal data type is determined to be valid abnormal data, thus completing the diagnosis of the abnormal data;

[0014] Based on valid abnormal data, a Lagrange optimization repair algorithm is used to construct an objective function, and physical hard constraints and communication quality soft constraints are incorporated into the optimization objective function. The optimal repair value that satisfies the physical hard constraints and communication quality soft constraints is iteratively solved to complete the repair of abnormal data.

[0015] Establish a mapping relationship between valid abnormal data and gateway device identity information, and based on the mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.

[0016] Secondly, the present invention also provides a diagnostic and repair system for abnormal data in a low communication environment of a charging station, the system comprising:

[0017] The data acquisition and preprocessing module is used to collect multi-source operating data from various devices in the charging station and preprocess it to obtain standardized data.

[0018] The orderliness discrimination module is used to segment standardized data into continuous time-series data segments, and calculate the information entropy of each time-series data segment through the information entropy discrimination mechanism. When any information entropy is greater than the preset information entropy threshold, the time-series data segment is judged to be an abnormal risk data segment.

[0019] The anomaly data diagnosis module is used to extract local spatial features and temporal dependencies from anomaly risk data fragments using a lightweight anomaly diagnosis model, and output anomaly type data and diagnostic confidence.

[0020] The valid data discrimination module is used to determine the abnormal data as valid abnormal data when the diagnostic confidence is greater than or equal to the preset confidence threshold, so as to complete the diagnosis of abnormal data.

[0021] The abnormal data repair module is used to construct an objective function based on valid abnormal data using a Lagrange optimization repair algorithm, and incorporates physical hard constraints and communication quality soft constraints into the optimization objective function. Iteratively solves for the optimal repair value that satisfies the physical hard constraints and communication quality soft constraints to complete the repair of abnormal data.

[0022] The fault location module is used to establish a mapping between valid abnormal data and gateway device identity information, and based on the mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.

[0023] The beneficial effects of this invention are as follows:

[0024] 1) This invention uses a CNN-LSTM fusion deep learning model, combined with historical data, to achieve accurate identification of multiple types of abnormal data in charging stations. At the same time, through lightweight model optimization, it adapts to the computing power and communication requirements of the low communication environment of charging stations, thus solving the problem of low data processing efficiency in low communication environments.

[0025] 2) This invention proposes a physical constraint-based Lagrange optimization repair algorithm, which integrates physical hard constraints such as voltage, current, SOC, single-machine rated power, and power balance of multiple devices to fundamentally eliminate the problem of repair values ​​exceeding the range and violating electrical laws; it introduces communication quality soft constraints based on packet loss rate, verification error rate, and transmission delay, so that the repair process can dynamically adjust the weights according to the communication environment, avoiding the repair results from over-reliance on unreliable data when the communication quality is extremely poor.

[0026] 3) By establishing a mapping between valid abnormal data and gateway device identity information, this invention can locate faulty gateway devices, reduce operation and maintenance costs, and improve the intelligence level of charging station data management. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of a method for diagnosing and repairing abnormal data in a low communication environment of a charging station according to an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of a system for diagnosing and repairing abnormal data in a low communication environment of a charging station, according to an embodiment of the present invention.

[0030] Figure 3This is a partial sample image of the charging station operation history feature database according to an embodiment of the present invention;

[0031] Figure 4 This is a diagram of the CNN-LSTM fusion deep learning model architecture according to an embodiment of the present invention;

[0032] Figure 5 This is an iterative convergence curve of the Lagrange optimization repair algorithm under different communication environment qualities according to an embodiment of the present invention. Detailed Implementation

[0033] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0034] According to an embodiment of the present invention, a method for diagnosing and repairing abnormal data based on low communication environment of charging stations is provided.

[0035] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, a method for diagnosing and repairing abnormal data based on a low communication environment in a charging station, according to an embodiment of the present invention, includes:

[0036] Step S1: Collect multi-source operating data of various devices in the charging station and preprocess it to obtain standardized data;

[0037] In this optional embodiment, the process of collecting multi-source operational data from various devices within the charging station and preprocessing it to obtain standardized data includes:

[0038] Using intelligent data acquisition terminals, multi-source operation data of photovoltaic, energy storage, charging piles, and V2G piles are collected within the charging station;

[0039] The multi-source operation data includes: photovoltaic output voltage and current data, charging pile charging voltage and current data, V2G pile charging and discharging voltage and current data, energy storage charging and discharging voltage, current, and SOC data, photovoltaic working status, energy storage working status, charging pile working status, and V2G pile working status data.

[0040] The multi-source operational data is normalized to eliminate the influence of dimensions and remove invalid noise signals, resulting in standardized data.

[0041] Specifically, intelligent data acquisition terminals (including but not limited to DL / T645, RS485, CAN, and Ethernet interface protocol acquisition devices) deployed in the charging station collect operational data from photovoltaic, energy storage, charging piles, and V2G piles within the charging station. The main data collected includes photovoltaic output voltage and current data, charging pile charging voltage and current data, V2G pile charging and discharging voltage and current data, energy storage charging and discharging voltage, current, and SOC data, as well as photovoltaic, energy storage, charging pile, and V2G pile operating status data. All data is normalized to the [0,1] interval to eliminate dimensional influences and remove invalid noise signals, resulting in preprocessed data. The data normalization formula is as follows:

[0042] ;

[0043] In the formula, The original value; The maximum value within the sample; It is the minimum value within the sample.

[0044] Step S2: Segment the standardized data to form continuous time-series data segments, and calculate the information entropy of each time-series data segment through the information entropy discrimination mechanism. When any information entropy is greater than the preset information entropy threshold, the time-series data segment is determined to be an abnormal risk data segment.

[0045] In this optional embodiment, the step of segmenting standardized data to form continuous time-series data segments, and calculating the information entropy of each time-series data segment using an information entropy discrimination mechanism, wherein if any information entropy is greater than a preset information entropy threshold, the time-series data segment is determined to be an abnormal risk data segment, includes:

[0046] Using the sliding window histogram method, standardized data is segmented according to a preset sliding window to form continuous time-series data segments;

[0047] The probability of each time series data segment appearing within a preset value range is statistically analyzed, and the information entropy of each time series data segment is calculated using the Shannon entropy formula.

[0048] Acquire multi-source operation data of the charging station for a preset period (7 consecutive days), calculate the average information entropy of the multi-source operation data for the preset period, and use the average information entropy as the preset information entropy threshold.

[0049] The information entropy of each time series data segment is compared with the preset information entropy threshold one by one;

[0050] When the information entropy is less than or equal to the preset information entropy threshold, it is determined that the data distribution of the time series data segment is concentrated, the state is ordered, and the operation is normal.

[0051] When the information entropy is greater than the preset information entropy threshold, the time series data segment is determined to be an abnormal risk data segment.

[0052] Specifically, the multi-device collaborative system of a charging station, consisting of photovoltaic, energy storage, charging piles, and V2G piles, is considered as a unified information body. During normal operation, the power, voltage, and current of each device are mutually coupled and constrained, resulting in a stable and orderly state, corresponding to low information entropy. When communication interference, data interruption, or equipment failure occurs, the data distribution dispersion increases, and the state transitions become disordered, corresponding to high information entropy. The trend of entropy changes is used to determine whether the system deviates from its stable operating range, thereby triggering anomaly diagnosis. The application process of the system orderliness discrimination mechanism based on information entropy is as follows:

[0053] The sliding window histogram method was used to divide the preprocessed data into fixed time windows, with a window size of 5 minutes and a step size of 1 minute, forming continuous time series data segments. The probability of each data point appearing within a segment was then calculated. ,satisfy .

[0054] Calculate information entropy using the Shannon entropy formula. :

[0055] ;

[0056] Take seven consecutive days of data on the normal operation of the charging station and calculate the average information entropy over those seven days. Normal operation data selection criteria: no packet loss, equipment voltage, current, and SOC all within the equipment's rated operating range, and no records of manual shutdown for maintenance. Information entropy. The data is dynamically updated daily, and the average information entropy of the data from the previous 7 days' normal operating hours is recalculated daily. To avoid misjudgments caused by seasonal and climate changes. Less than The data is distributed centrally and is in an orderly manner, and the system is operating normally. Greater than The data is scattered and its state is chaotic, posing a risk of anomalies.

[0057] Step S3: Using a lightweight anomaly diagnosis model (in this embodiment, it refers to a CNN-LSTM fusion deep learning model), extract local spatial features and temporal dependencies from anomaly risk data segments, and output anomaly type data and diagnostic confidence.

[0058] In this optional embodiment, the step of using a lightweight anomaly diagnosis model to extract local spatial features and temporal dependencies from anomaly risk data fragments and outputting anomaly type data and diagnostic confidence includes:

[0059] An anomaly diagnosis model is constructed, which includes a convolutional layer, a pooling layer, a long short-term memory layer, and an output layer;

[0060] A feature database was constructed by selecting historical normal operation data of charging stations within a preset time period (3 to 6 months), and a labeled dataset containing mutation anomalies, constant value anomalies, missing value anomalies and deviation anomalies was constructed through simulation.

[0061] The feature database and labeled dataset are used as training samples to iteratively train the anomaly diagnosis model until the loss function converges, thus obtaining the trained anomaly diagnosis model.

[0062] The trained anomaly diagnosis model is lightweighted to obtain a lightweight anomaly diagnosis model; wherein, the lightweighting process includes pruning neurons whose absolute weight value is lower than a preset threshold (0.001) and quantizing the weights from floating-point to fixed-point.

[0063] Anomaly risk data segments are input into a lightweight anomaly diagnosis model. The convolutional layer extracts the local spatial features between the operating data of each device at the same time in the anomaly risk data segments. After feature dimensionality reduction by the pooling layer, the features are input into the long short-term memory layer to capture temporal dependencies. The output layer uses the softmax activation function to output the classification probabilities of the anomaly risk data segments corresponding to mutation anomalies, constant value anomalies, missing value anomalies, and deviation anomalies, thus obtaining the anomaly type and diagnostic confidence of each anomaly risk data segment (it should be noted that the diagnostic confidence value is calculated based on the softmax function, and the maximum classification probability of the target anomaly category is taken as the confidence value).

[0064] Specifically, such as Figure 4 As shown, a CNN-LSTM fusion deep learning model is constructed. This model includes convolutional layers, pooling layers, long short-term memory layers, and an output layer. There are three convolutional layers: the first convolutional layer uses 32 3×3 convolutional kernels with ReLU activation to extract local spatial features; the second and third convolutional layers use 64 3×3 convolutional kernels with ReLU activation to enhance feature extraction. The pooling layer uses max pooling with a stride of 2 to reduce dimensionality and prevent overfitting. The long short-term memory layer consists of two layers: the first layer contains 128 neurons, and the second layer contains 64 neurons, used to capture the long-term dependencies and temporal correlations of data such as voltage, current, and SOC over time.

[0065] like Figure 3 As shown, historical feature data of charging station equipment over the past 3-6 months were collected to construct a feature database. A labeled dataset containing mutation anomalies, constant value anomalies, missing value anomalies, and deviation anomalies was constructed through simulation.

[0066] Mutation anomalies are defined as numerical jumps exceeding a threshold ±15% within a short period; constant value anomalies refer to data remaining constant over a long period; missing value anomalies are NaN values ​​caused by lost data packets; and deviation anomalies are numerical values ​​deviating from the physical logic range (e.g., energy storage SOC exceeding 100%). Training samples are input into the CNN-LSTM fusion deep learning model for training. The initial learning rate is set to 0.001, and the Adam optimization algorithm is used to iteratively optimize the parameters of the CNN-LSTM fusion deep learning model until the loss function converges, resulting in the trained CNN-LSTM fusion deep learning model.

[0067] After the model training is completed, the CNN-LSTM fusion deep learning model is pruned and quantized. Redundant neurons with absolute weight values ​​below 0.001 are pruned, and the weights of the CNN-LSTM fusion deep learning model are quantized from 32-bit floating-point to 16-bit fixed-point. Redundant neurons and weight parameters are deleted to reduce the computational load and memory usage of the CNN-LSTM fusion deep learning model, thus adapting to the computing power requirements of the low-communication environment of the charging station.

[0068] Data segments identified as having anomaly risk by information entropy are input into a trained CNN-LSTM fusion deep learning model. The CNN-LSTM fusion deep learning model extracts local spatial features of the data through three convolutional layers, including data mutation points and numerical fluctuation features. After feature dimensionality reduction by pooling layers, the feature vectors are input into the LSTM layer to capture the dependencies of data over time, including the temporal change trend of equipment operation data and the correlation features of data in adjacent time periods. Finally, the output layer uses the softmax activation function to output the classification probability of the data segment corresponding to various anomalies (mutation anomaly, constant value anomaly, missing value anomaly, and deviation anomaly).

[0069] Step S4: When the diagnostic confidence is greater than or equal to the preset confidence threshold (0.8), the abnormal data is determined to be valid abnormal data, so as to complete the diagnosis of abnormal data;

[0070] Specifically, considering the low communication environment of charging stations, a confidence threshold of 0.8 is set as the baseline. If the confidence level is ≥0.8, it is determined to be a valid anomaly and enters the repair process; if the confidence level is <0.8, it is determined to be a false alarm or noise data and is filtered out.

[0071] Step S5: Based on the valid abnormal data, construct the objective function using the Lagrange optimization repair algorithm, and incorporate the physical hard constraints and communication quality soft constraints into the optimization objective function. Iteratively solve the optimal repair value that satisfies the physical hard constraints and communication quality soft constraints to complete the repair of the abnormal data.

[0072] In this optional embodiment, the step of constructing an objective function based on valid anomaly data using a Lagrange optimization repair algorithm, incorporating physical hard constraints and communication quality soft constraints into the optimization objective function, and iteratively solving for the optimal repair value that satisfies both physical hard constraints and communication quality soft constraints to complete the repair of anomaly data includes:

[0073] The objective function is constructed by minimizing the deviation between the unrepaired value of valid abnormal data and historical similar working condition data, and minimizing the deviation from the Gaussian weighted mean of the same type of equipment in the same region.

[0074] Constructing physical hard constraints and communication quality soft constraints;

[0075] The Lagrange multiplier method is used to fuse the objective function, physical hard constraints, and communication quality soft constraints to construct a Lagrange optimization function;

[0076] The intermediate repair value is obtained by taking the partial derivative of the Lagrange optimization function and setting the partial derivative to zero.

[0077] Substitute the intermediate repair values ​​into the physical hard constraints and communication quality soft constraints for verification, and determine one by one whether the repaired voltage, current, SOC and power meet the physical hard constraints and communication quality soft constraints.

[0078] If the intermediate repair value satisfies both the physical hard constraint and the communication quality soft constraint, then the intermediate repair value shall be taken as the optimal repair value.

[0079] If the intermediate repair value does not meet the physical hard constraints or the communication quality soft constraints, the penalty factor for not meeting the physical hard constraints or the communication quality soft constraints is iteratively updated, and the updated penalty factor is substituted back into the Lagrange optimization function to solve until the obtained intermediate repair value meets the physical hard constraints and the communication quality soft constraints, and then the optimal repair value is output.

[0080] In this optional embodiment, the construction of physical hard constraints and communication quality soft constraints includes:

[0081] The physical hard constraints are that the voltage repair value, current repair value, energy storage SOC repair value, and power repair value of the valid abnormal data satisfy their respective preset physical operating boundaries;

[0082] The communication quality soft constraint is an adaptive constraint based on the deviation between the value to be repaired and the physical median value of the device, with a higher constraint strength for lower communication quality.

[0083] Specifically, for the abnormal data obtained from model diagnosis, this invention proposes a Lagrange optimization repair algorithm with physical constraints. It integrates voltage and current limits, SOC range, and power balance to form physical hard constraints, and introduces adaptive soft constraints constructed from communication quality coefficients. The Lagrange multiplier method is used to incorporate the hard and soft constraints into the objective function to construct a Lagrange optimization equation. Solving the optimization equation by taking the derivative yields the globally optimal repair value that satisfies all constraints. The specific steps are as follows:

[0084] (1) Construct the optimization objective function The optimization objective is to minimize the deviation between the required repair value of valid abnormal data and historical similar operating condition data, and to minimize the deviation between the required repair value and the Gaussian weighted mean of normal equipment of the same type in the same area, so that the required repair value can simultaneously conform to the historical time series pattern and the operating characteristics of the equipment in the same period of the station.

[0085] ;

[0086] In the formula, These are the values ​​to be repaired for valid outlier data;

[0087] This is based on historical similar operating condition data using dynamic time warping matching. The current data segment is compared with a feature database of historical normal operating data using a dynamic time warping algorithm. The historical normal data segment with the smallest cumulative distance is calculated, and the data corresponding to the position of this historical normal data segment is extracted. ;

[0088] minimize This means minimizing the deviation between the required repair value of valid abnormal data and historical similar data. To obtain the Gaussian weighted average of the operating values ​​of similar normal equipment in the same area at the current time, we first filter out similar normal equipment in the same area at the current time, obtaining a set of operating values ​​for normal equipment. Then, we use a Gaussian function to calculate the weight of each operating value for normal equipment, and finally, we calculate the weighted average. ;

[0089] minimize This means minimizing the deviation between the required repair value of valid abnormal data and the data of normal equipment of the same type in the same area;

[0090] and The weights for the time and space dimensions are dynamically allocated based on the anomaly diagnosis confidence level: when the confidence level is ≥0.9, =0.6、 =0.4; when 0.8 ≤ confidence level < 0.9, =0.5、 =0.5.

[0091] (2) Equipment physical hard constraints: (In order to avoid the solution always falling outside the feasible region, the maximum number of iterations is set to 20. If the constraint is still violated after reaching the maximum number of iterations, the equipment physical boundary value is directly used as the optimal repair value to ensure that the optimal repair value strictly meets the equipment physical constraints); the repair value to be found for valid abnormal data must meet the equipment electrical boundary and system power balance constraints.

[0092] ≤ ≤ The required repair value for abnormal voltage data should meet the constraints of the equipment's highest and lowest operating voltages.

[0093] ≤ ≤ The value to be repaired for abnormal current data should meet the constraints of the equipment's maximum and minimum operating currents.

[0094] 0≤SOC≤100%, meaning the SOC value of the energy storage system should be within the range of 0-1.

[0095] 0≤P≤ This means that the operating power of the equipment must not exceed its rated power. Rated power for equipment such as photovoltaics, energy storage, charging piles, and V2G piles.

[0096] (3) Power balance constraints for multiple devices: ;in, The power grid interaction power is collected in real time at the power station's grid connection point; (The power consumption of auxiliary loads such as station lighting, air conditioning, and monitoring is collected in real time by the station's auxiliary power consumption data acquisition terminal.)

[0097] (4) Soft constraints on communication quality: To address the problem of unreliable data in low communication environments, a communication quality coefficient is introduced. (When poor communication quality causes a constant value stuck abnormal value, which remains at a fixed value, similar historical operating conditions will be encountered.) Same equipment space (It plays a leading role; the communication quality constraint is only an auxiliary constraint and not the sole basis for repair). Judgment rule: Based on gateway packet loss rate Message verification error rate Transmission delay The three types of indicators were normalized to interval;

[0098] ;

[0099] ;

[0100] ;

[0101] In the formula, 0.3, 0.3, and 1 are the threshold values ​​for the three types of indicators (gateway packet loss rate, etc.). >30%, Message verification error rate >30%, transmission delay (The normalized value is 0 when the latency is >1000ms), and the weighted average is used to obtain the final communication quality coefficient. .

[0102] ;

[0103] When the gateway packet loss rate 0, message check error rate 0, transmission delay When ≤200ms, (Optimal communication);

[0104] When the gateway packet loss rate >30%, message verification error rate >30%, transmission delay When >1s, Approaching 0 (extremely poor communication).

[0105] (5) Constructed Lagrangian optimization function:

[0106] ;

[0107] In the formula, For Lagrange optimization functions; These are the values ​​to be repaired for valid outlier data; and These are the weights for the time dimension and the weights for the space dimension, respectively. Data based on similar historical operating conditions; It is the Gaussian weighted mean; As a penalty factor; As a voltage lower limit constraint penalty factor, the constraint repair voltage shall not be lower than the minimum operating voltage of the equipment. ; As a voltage upper limit constraint penalty factor, the constraint repair voltage shall not exceed the rated maximum voltage of the equipment. ; As a current lower limit constraint penalty factor, the constraint repair current shall not be lower than the minimum operating current of the equipment. ; The current upper limit constraint penalty factor ensures that the constraint repair current does not exceed the equipment's rated maximum current. ; This is the physical median value of the equipment; This is a penalty factor for communication quality constraints; the lower the communication quality, the lower the penalty factor. The higher the weight of the penalty item, the less likely the value to be repaired will deviate from the physically reasonable range. This refers to the communication quality coefficient. , , , , Penalty factor growth coefficient It should be noted that the penalty factor growth coefficient of 1.2 was obtained through simulation parameter tuning. If the value is too small, the penalty intensity will increase slowly and the iteration will converge slowly; if the value is too large, numerical oscillations are likely to occur. 1.2 is a moderate convergence speed and solution stability.

[0108] beg right The first-order partial derivative, and let the partial derivative number =0: ( (Intermediate repair value)

[0109] ;

[0110] intermediate repair value (The intermediate repair value can be any of the voltage, current, or SOC data to be repaired.) Substitute this into the equipment electrical boundary and system power balance constraint verification and communication quality soft constraint, and perform a judgment on the repaired voltage, current, SOC, and power one by one; if the intermediate repair value... Once the physical constraints and communication quality constraints are satisfied, the intermediate repair value is taken as the optimal repair value.

[0111] ≤ ≤ , ≤ ≤ , 0≤SOC≤100%, 0≤P≤ , If any of the above conditions are violated, it means that the current penalty factor value cannot simultaneously satisfy the constraints, then iterative update. , , , , ;

[0112] If the intermediate repair value Violation of the lower voltage limit constraint, i.e. Other constraint penalty factors remain unchanged. Updated to ;

[0113] If the voltage upper limit constraint is violated, i.e. Other constraint penalty factors remain unchanged. Updated to ;

[0114] If the lower current limit constraint is violated, i.e. Other constraint penalty factors remain unchanged. Updated to ;

[0115] If the current limit constraint is violated, i.e. Other constraint penalty factors remain unchanged. Updated to ;

[0116] If communication soft constraints or power balance constraints are violated, the penalty factors for other constraints remain unchanged. Updated to ;

[0117] The Lagrange optimization function is re-substituted until all constraints are satisfied, and the optimal repair value is output. If the constraints are still not satisfied after the maximum number of iterations (20), the physical boundary value of the equipment is directly used as the optimal repair value. The repaired data can be pushed to the charging station's energy management system (EMS) and operation and maintenance management platform to replace the original abnormal data, supporting core business decisions such as photovoltaic output prediction, energy storage charging and discharging scheduling, and charging pile load management, thereby improving the intelligent and safe and stable operation and management capabilities of the charging station.

[0118] Step S6: Establish a mapping between valid abnormal data and gateway device identity information, and based on the mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.

[0119] In this optional embodiment, establishing the association mapping between valid abnormal data and gateway device identity information, and extracting the source feature set from the valid abnormal data based on the association mapping relationship to determine the faulty device and fault type includes:

[0120] A device information index library is constructed, which contains the identity information of each device in the charging station and the identity information of its respective gateway.

[0121] Each piece of multi-source operational data collected is attached with a source tag; wherein, the source tag contains device identity information and the identity information of the gateway to which it belongs;

[0122] Once the lightweight anomaly diagnosis model identifies valid anomaly data, it extracts the anomaly features of each valid anomaly data, and establishes an association mapping between valid anomaly data and gateway device identity information using the anomaly features and timestamp as index keys and the device identity information and its associated gateway identity information as mapping values.

[0123] Based on the association mapping, the gateway device identity information and abnormal characteristics of each valid abnormal data are extracted. When all devices under the same gateway simultaneously experience missing anomalies, the information entropy is greater than 1.5 times the preset information entropy threshold, and the communication quality coefficient is less than the preset value (the preset value is 0.3), it is determined to be a communication link failure.

[0124] When only a single device exhibits a constant value anomaly, it is determined that the device's interface is faulty.

[0125] By comparing the timestamps of the abnormal features with the device logs, the faulty device and the type of fault can be located.

[0126] Specifically, (1) a basic information database for equipment is constructed. A charging station equipment information index table is pre-established in the system. Each record includes: equipment number, gateway number, gateway MAC address, communication port number, physical location of the equipment and data upload link identifier, forming a unique index for the equipment identity.

[0127] (2) Label each piece of data with a source tag. During the data collection and uploading stage, add source metadata to each piece of collected running data, including the device number that generated the data, the gateway MAC address that forwarded the data, the data transmission link identifier and the timestamp, to form a standardized data frame with a source tag.

[0128] (3) Extract the feature vector of abnormal data. After the diagnostic model identifies abnormal data, it extracts the multi-dimensional features corresponding to the abnormal data to form an abnormal feature vector, including: abnormal type, time of occurrence of abnormality, abnormal magnitude, data continuity features, communication packet loss features, data jitter features, etc.

[0129] (4) Establish a one-to-one association mapping. Using the abnormal data feature vector as the index item and the device number and gateway MAC address as the mapping value, establish an association mapping table in the form of key-value pairs in the system. The key is the abnormal data feature vector and timestamp, and the value is the device number, gateway MAC address and communication link information. For each abnormal data identified, a new record is added to the mapping table to realize the real-time binding and fast retrieval of abnormal features and gateway device identity information.

[0130] (5) Form an anomaly-device association map, cluster multiple anomaly data appearing in the same time period and under the same gateway, and statistically analyze the frequency and type distribution of anomalies to form an anomaly profile at the gateway level. Based on the above association mapping relationship, extract the source features such as transmission protocol identifier and device number of the anomaly data, locate the faulty gateway device and distinguish the fault type (including gateway hardware failure, communication link failure, and device interface failure). The specific location method is based on anomaly propagation path analysis. If all photovoltaic, energy storage, charging pile, and V2G devices under the same gateway MAC simultaneously show missing anomalies, the information entropy Communication quality coefficient If the value is abnormal, it is determined to be a communication link failure; if only a single device shows a constant value abnormality, it is determined to be a sensor or interface failure of that device.

[0131] By comparing the timestamps of the anomalies with the device logs, the faulty gateway devices and specific fault types can be accurately located. Furthermore, the specific manifestations of various faults can be distinguished by the transmission stability characteristics of the abnormal data.

[0132] like Figure 2 As shown, according to another embodiment of the present invention, a diagnostic and repair system for abnormal data in a low communication environment of a charging station is also provided. The system includes:

[0133] Data acquisition and preprocessing module 1 is used to collect multi-source operating data of various devices in the charging station and preprocess it to obtain standardized data.

[0134] Orderliness discrimination module 2 is used to segment standardized data into continuous time-series data segments, and calculate the information entropy of each time-series data segment through the information entropy discrimination mechanism. When any information entropy is greater than the preset information entropy threshold, the time-series data segment is judged to be an abnormal risk data segment.

[0135] Anomaly data diagnosis module 3 is used to extract local spatial features and temporal dependencies from anomaly risk data fragments using a lightweight anomaly diagnosis model, and output anomaly type data and diagnostic confidence.

[0136] The valid data discrimination module 4 is used to determine the abnormal data as valid abnormal data when the diagnostic confidence is greater than or equal to the preset confidence threshold, so as to complete the diagnosis of abnormal data.

[0137] The abnormal data repair module 5 is used to construct an objective function based on valid abnormal data using the Lagrange optimization repair algorithm, and incorporate physical hard constraints and communication quality soft constraints into the optimization objective function. Iteratively solves the optimal repair value that satisfies the physical hard constraints and communication quality soft constraints to complete the repair of abnormal data.

[0138] The fault location module 6 is used to establish an association mapping between valid abnormal data and gateway device identity information, and based on the association mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.

[0139] To facilitate understanding of the above technical solutions of the present invention, the working principle or operation method of the present invention in actual process will be described in detail below.

[0140] like Figure 5As shown, this embodiment selects a charging station containing 4 charging piles, 1 energy storage system, 1 photovoltaic system, and 2 V2G devices; collects real-world station operation data for 30 consecutive days, and artificially injects four types of anomalies (mutation, constant value, missing value, and deviation) to simulate good communication (…). ), general communication ( Poor communication () Simulation verification was conducted under three operating conditions, and the specific process is as follows:

[0141] Obtain the baseline dataset. Read the raw operational data collected continuously over 30 days from the charging station, including voltage, current, SOC, and equipment operating status of photovoltaic, energy storage, charging piles, and V2G piles. Simultaneously collect the raw communication metrics of the gateway (gateway packet loss rate). Message verification error rate Transmission delay ).

[0142] Three control simulation groups were set up: Group A: Good communication condition. Group B: General communication conditions Group C: Extremely poor communication conditions. .

[0143] Four types of abnormal data were artificially injected. For each dataset A / B / C, different time segments were randomly selected to artificially inject four types of anomalies: Sudden change anomaly: Voltage / current values ​​jump by more than ±15% relative to the true value within a short period; Constant value anomaly: Data was forcibly fixed at a certain constant for 15-30 minutes; Missing data anomaly: Some sampling points were set to NaN, simulating packet loss; Deviation anomaly: Sampling points were modified to illegal values ​​exceeding the physical boundaries of the device (e.g., energy storage voltage SOC=110%).

[0144] Normalization was performed on the original simulation data after the injection of anomalies in groups A, B, and C, and a device number and the MAC source tag of the gateway to which each data point belongs were added.

[0145] Using ground truth data from 7 consecutive days without anomalies, calculate the mean information entropy of normal samples. Calculate the information entropy for each time segment. , The timing segment was determined to be running normally; The segment is labeled as an abnormal risk data segment and fed into the trained lightweight CNN-LSTM model.

[0146] Set a preset confidence threshold Diagnostic confidence : Determined as valid abnormal data, proceed to the Lagrange repair and fault location process; diagnostic confidence level If the alarm is determined to be a noise false alarm, it will be filtered out and discarded.

[0147] Assign time weights to the Lagrange optimization function based on diagnostic confidence. Spatial weights For, confidence level hour, , ; Confidence hour, , .

[0148] Construct an objective function, solve for the intermediate repair value, and then verify the intermediate repair value by substituting it into the equipment's physical constraints. If all constraints are met, the intermediate repair value is taken as the optimal repair value; if the constraints are not met, multiply the corresponding penalty factor by 1.2 and solve again. After reaching the maximum number of iterations (20), directly use the physical boundary value as the optimal repair value.

[0149] Read the source tags (device number, gateway) that come with the abnormal data, perform fault location judgment, compare the fault location results output by the simulation with the fault tags manually injected in the early stage of the simulation, and calculate the fault location accuracy.

[0150] To verify the effectiveness of the information entropy discrimination mechanism and Lagrange optimization repair algorithm proposed in this invention, a comparative experiment was conducted with the following traditional methods, as shown in Tables 1 and 2 below:

[0151] Table 1. Comparison between information entropy discrimination mechanism and traditional threshold discrimination method

[0152] Table 2. Comparison between Lagrange optimization repair algorithm and traditional methods

[0153] Specifically, Table 1 compares the performance of the information entropy discrimination mechanism used in this invention with that of the traditional threshold discrimination method in the task of identifying equipment anomalies in charging stations. The equipment anomaly identification rate characterizes the ability of the information entropy discrimination mechanism to accurately identify abnormal equipment samples; the false positive rate in low communication environment (C=0.2) represents the probability that the information entropy discrimination mechanism will incorrectly identify normal data as anomalies in scenarios with poor communication conditions. The lower the value of this index, the stronger the anti-interference ability of the information entropy discrimination mechanism. As can be seen from the data in Table 1, the equipment anomaly identification rate of the information entropy discrimination mechanism of this invention is 92.7%, which is higher than the 86.1% of the traditional threshold discrimination method; in low communication environment, the false positive rate of this method is only 5.1%, which is much lower than the 22.8% of the traditional threshold discrimination method. The traditional threshold discrimination method relies on a fixed threshold to determine anomalies. In scenarios with strong communication noise interference, the threshold adaptability is poor, and a large number of false positives are easily generated; the information entropy discrimination mechanism realizes anomaly discrimination by mining the uncertainty features of data, which has stronger robustness to communication environment disturbances and balances higher recognition accuracy with lower false positive risk.

[0154] Table 2 compares the data repair performance of the Lagrange optimization repair algorithm and the traditional linear interpolation repair method under different communication quality scenarios. The root mean square error (RMSE) is used as the evaluation index; a smaller RMSE indicates a smaller deviation between the repaired data and the actual data, and higher repair accuracy. The RMSE index shows that the Lagrange optimization repair algorithm is generally superior to the traditional linear interpolation repair method. Traditional linear interpolation relies solely on the linear relationship between adjacent data points to fill in missing data, without considering the inherent constraints of the device data. When communication deteriorates or data loss increases, the repair error significantly increases. The Lagrange optimization repair algorithm introduces the concept of constraint optimization, combining the physical constraints of the data to solve for the optimal estimate, effectively reducing data repair errors under various communication conditions.

[0155] The experimental results in Tables 1 and 2 show that, compared to traditional threshold discrimination methods, the information entropy discrimination mechanism of this invention can improve the accuracy of anomaly identification in charging station equipment, significantly suppress misjudgment in low-communication environments, and has stronger anti-communication interference capabilities. Compared to traditional linear interpolation repair methods, the Lagrange optimization repair algorithm can effectively reduce the root mean square error of data repair in various scenarios, including good, average, and poor communication environments, and its data repair performance is comprehensively superior to traditional linear interpolation methods. The two mechanisms work together to ensure both the reliability of anomaly identification and the accuracy of missing data repair in complex and variable communication environments, verifying the effectiveness of the proposed information entropy discrimination mechanism and Lagrange optimization repair algorithm, and demonstrating their ability to better adapt to the needs of anomaly data diagnosis and repair in low-communication environments of charging stations.

[0156] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing and repairing abnormal data in a low-communication environment of a charging station, characterized in that, include: Collect multi-source operational data from various devices within the charging station and preprocess it to obtain standardized data; Standardized data is segmented to form continuous time-series data segments. The information entropy of each time-series data segment is calculated through an information entropy discrimination mechanism. When any information entropy is greater than a preset information entropy threshold, the time-series data segment is determined to be an abnormal risk data segment. Using a lightweight anomaly diagnosis model, local spatial features and temporal dependencies are extracted from anomaly risk data fragments, and anomaly type data and diagnostic confidence are output. When the diagnostic confidence level is greater than or equal to the preset confidence threshold, the abnormal data type is determined to be valid abnormal data, so as to complete the diagnosis of abnormal data; Based on valid abnormal data, a Lagrange optimization repair algorithm is used to construct an objective function, and physical hard constraints and communication quality soft constraints are incorporated into the optimization objective function. The optimal repair value that satisfies the physical hard constraints and communication quality soft constraints is iteratively solved to complete the repair of abnormal data. Establish a mapping relationship between valid abnormal data and gateway device identity information, and based on the mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.

2. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 1, characterized in that, The process of collecting and preprocessing multi-source operational data from various devices within the charging station to obtain standardized data includes: Using intelligent data acquisition terminals, multi-source operation data of photovoltaic, energy storage, charging piles, and V2G piles are collected within the charging station; The multi-source operation data includes: photovoltaic output voltage and current data, charging pile charging voltage and current data, V2G pile charging and discharging voltage and current data, energy storage charging and discharging voltage, current, and SOC data, photovoltaic working status, energy storage working status, charging pile working status, and V2G pile working status data. The multi-source operational data is normalized to eliminate the influence of dimensions and remove invalid noise signals, resulting in standardized data.

3. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 1, characterized in that, The process of segmenting standardized data into continuous time-series data segments and calculating the information entropy of each time-series data segment using an information entropy discrimination mechanism, wherein a time-series data segment is determined to be an abnormal risk data segment when any information entropy exceeds a preset information entropy threshold, includes: Using the sliding window histogram method, standardized data is segmented according to a preset sliding window to form continuous time-series data segments; The probability of each time series data segment appearing within a preset value range is statistically analyzed, and the information entropy of each time series data segment is calculated using the Shannon entropy formula. Acquire multi-source operation data of the charging station for a preset period, calculate the average information entropy of the multi-source operation data for the preset period, and use the average information entropy as the preset information entropy threshold. The information entropy of each time series data segment is compared with the preset information entropy threshold one by one; When the information entropy is less than or equal to the preset information entropy threshold, it is determined that the data distribution of the time series data segment is concentrated, the state is ordered, and the operation is normal. When the information entropy is greater than the preset information entropy threshold, the time series data segment is determined to be an abnormal risk data segment.

4. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 1, characterized in that, The lightweight anomaly diagnosis model extracts local spatial features and temporal dependencies from anomaly risk data fragments, and outputs anomaly type data and diagnostic confidence, including: An anomaly diagnosis model is constructed, which includes a convolutional layer, a pooling layer, a long short-term memory layer, and an output layer; A feature database was constructed by selecting historical normal operation data of charging stations within a preset time period, and a labeled dataset containing mutation anomalies, constant value anomalies, missing value anomalies and deviation anomalies was constructed through simulation. The feature database and labeled dataset are used as training samples to iteratively train the anomaly diagnosis model until the loss function converges, thus obtaining the trained anomaly diagnosis model. The trained anomaly diagnosis model is lightweighted to obtain a lightweight anomaly diagnosis model; wherein, the lightweighting process includes pruning neurons whose absolute weight value is lower than a preset threshold and quantizing the weights from floating-point to fixed-point. Anomaly risk data segments are input into a lightweight anomaly diagnosis model. The convolutional layer extracts the local spatial features between the operating data of each device at the same time in the anomaly risk data segments. After feature dimensionality reduction by the pooling layer, the features are input into the long short-term memory layer to capture temporal dependencies. The output layer uses the softmax activation function to output the classification probabilities of the anomaly risk data segments corresponding to mutation anomalies, constant value anomalies, missing value anomalies, and deviation anomalies, thus obtaining the anomaly type and diagnostic confidence of each anomaly risk data segment.

5. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 1, characterized in that, The process of repairing abnormal data involves constructing an objective function based on valid anomaly data using a Lagrange optimization algorithm, incorporating both physical hard constraints and communication quality soft constraints into the objective function, and iteratively solving for the optimal repair value that satisfies both the physical hard constraints and the communication quality soft constraints. The objective function is constructed by minimizing the deviation between the unrepaired value of valid abnormal data and historical similar working condition data, and minimizing the deviation from the Gaussian weighted mean of the same type of equipment in the same region. Constructing physical hard constraints and communication quality soft constraints; The Lagrange multiplier method is used to fuse the objective function, physical hard constraints, and communication quality soft constraints to construct a Lagrange optimization function; The intermediate repair value is obtained by taking the partial derivative of the Lagrange optimization function and setting the partial derivative to zero. Substitute the intermediate repair values ​​into the physical hard constraints and communication quality soft constraints for verification, and determine one by one whether the repaired voltage, current, SOC and power meet the physical hard constraints and communication quality soft constraints. If the intermediate repair value satisfies both the physical hard constraint and the communication quality soft constraint, then the intermediate repair value shall be taken as the optimal repair value. If the intermediate repair value does not meet the physical hard constraints or the communication quality soft constraints, the penalty factor for not meeting the physical hard constraints or the communication quality soft constraints is iteratively updated, and the updated penalty factor is substituted back into the Lagrange optimization function to solve until the obtained intermediate repair value meets the physical hard constraints and the communication quality soft constraints, and then the optimal repair value is output.

6. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 5, characterized in that, The construction of physical hard constraints and communication quality soft constraints includes: The physical hard constraints are that the voltage repair value, current repair value, energy storage SOC repair value, and power repair value of the valid abnormal data satisfy their respective preset physical operating boundaries; The communication quality soft constraint is an adaptive constraint based on the deviation between the value to be repaired and the physical median value of the device, with a higher constraint strength for lower communication quality.

7. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 5, characterized in that, The formula for constructing the Lagrange optimization function is as follows: ; In the formula, For Lagrange optimization functions; These are the values ​​to be repaired for valid outlier data; and These are the weights for the time dimension and the weights for the space dimension, respectively. Data based on similar historical operating conditions; It is the Gaussian weighted mean; As a penalty factor; This is the penalty factor for the lower voltage limit constraint; This is the minimum operating voltage for the equipment; This is the penalty factor for the upper limit of voltage constraint; The rated maximum voltage of the equipment; This is the penalty factor for the lower limit of current constraint; This is the minimum operating current of the device; This is the penalty factor for the upper limit constraint on current. The rated maximum current of the equipment; This is the physical median value of the equipment; This serves as a penalty factor for communication quality constraints. This is the communication quality coefficient.

8. The method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 7, characterized in that, The partial derivative formula for the Lagrange optimization function is as follows: ; In the formula, This is an intermediate repair value.

9. A method for diagnosing and repairing abnormal data in a low-communication environment of a charging station according to claim 1, characterized in that, The process of establishing a mapping between valid abnormal data and gateway device identity information, and extracting source feature sets from the valid abnormal data based on the mapping relationship to determine the faulty device and fault type includes: A device information index library is constructed, which contains the identity information of each device in the charging station and the identity information of its respective gateway. Each piece of multi-source operational data collected is attached with a source tag; wherein, the source tag contains identity information and the identity information of the gateway to which it belongs; Once the lightweight anomaly diagnosis model identifies valid anomaly data, it extracts the anomaly features of each valid anomaly data, and establishes an association mapping between valid anomaly data and gateway device identity information using the anomaly features and timestamp as index keys and the device identity information and its associated gateway identity information as mapping values. Based on the association mapping, the gateway device identity information and abnormal characteristics of each valid abnormal data are extracted. When all devices under the same gateway simultaneously experience missing anomalies, the information entropy is greater than 1.5 times the preset information entropy threshold, and the communication quality coefficient is less than the preset value, it is determined to be a communication link failure. When only a single device exhibits a constant value anomaly, it is determined that the device's interface is faulty. By comparing the timestamps of the abnormal features with the device logs, the faulty device and the type of fault can be located.

10. A system for diagnosing and repairing abnormal data in a low-communication environment at a charging station, used to implement the method for diagnosing and repairing abnormal data in a low-communication environment at a charging station as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition and preprocessing module is used to collect multi-source operating data from various devices in the charging station and preprocess it to obtain standardized data. The orderliness discrimination module is used to segment standardized data into continuous time-series data segments, and calculate the information entropy of each time-series data segment through the information entropy discrimination mechanism. When any information entropy is greater than the preset information entropy threshold, the time-series data segment is judged to be an abnormal risk data segment. The anomaly data diagnosis module is used to extract local spatial features and temporal dependencies from anomaly risk data fragments using a lightweight anomaly diagnosis model, and output anomaly type data and diagnostic confidence. The valid data discrimination module is used to determine the abnormal data as valid abnormal data when the diagnostic confidence is greater than or equal to the preset confidence threshold, so as to complete the diagnosis of abnormal data. The abnormal data repair module is used to construct an objective function based on valid abnormal data using a Lagrange optimization repair algorithm, and incorporates physical hard constraints and communication quality soft constraints into the optimization objective function. It iteratively solves for the optimal repair value that satisfies the physical hard constraints and communication quality soft constraints to complete the repair of abnormal data. The fault location module is used to establish a mapping between valid abnormal data and gateway device identity information, and based on the mapping relationship, extract the source feature set from the valid abnormal data to determine the faulty device and fault type.