Unmanned inspection and automatic operation and maintenance system for charging pile operation

Abstract

The invention relates to the technical field of charging pile operation and maintenance systems, and particularly discloses a charging pile operation unmanned inspection and automatic operation and maintenance system which comprises a data acquisition module, a state initial judgment module, a fault diagnosis module and an automatic operation and maintenance module. An unmanned three-dimensional inspection network is constructed through an unmanned aerial vehicle and an Internet of Things sensor, accurate fault diagnosis is realized in combination with multi-modal feature fusion, a dynamic operation and maintenance priority is generated and a differential strategy is formulated based on a machine learning model, and a complete operation and maintenance closed loop of acquisition, diagnosis, execution, feedback and optimization is formed at the same time. According to the method, the inspection real-time performance and the fault recognition accuracy are greatly improved, the operation and maintenance resource configuration efficiency is optimized, the operation and maintenance cost is reduced, the equipment downtime can be shortened, the charging experience of a user is improved, the operation and maintenance capability can be continuously iteratively optimized through data accumulation, and powerful support is provided for stable and reliable operation of a charging pile network.

Description

Technical Field

[0001] This invention relates to the field of charging pile operation and maintenance system technology, specifically to an unmanned inspection and automatic operation and maintenance system for charging piles. Background Technology

[0002] With the rapid development of the new energy vehicle industry, charging piles, as a core supporting infrastructure, have seen their deployment scale and coverage expand continuously, and are now widely distributed in diverse scenarios such as urban public areas, residential communities, and highway service areas. However, the current operation and maintenance field of charging piles still faces many technical bottlenecks, making it difficult to meet the efficient management needs under large-scale operation. The main problems are reflected in the following aspects:

[0003] The traditional inspection model is inefficient and lacks coverage: Relying on manual on-site inspections, traditional charging pile inspections are not only costly in terms of manpower and time, but also suffer from long inspection cycles and limited coverage. For charging piles scattered in remote areas, manual inspections cannot achieve high-frequency monitoring, leading to delays in detecting abnormal equipment conditions and potentially missing the optimal time for fault handling. Furthermore, manual inspections are heavily influenced by subjective experience, resulting in low accuracy in identifying subtle deformations in the charging pile's outer casing or hidden internal temperature rises, making it impossible to comprehensively understand the equipment's health status.

[0004] Fault diagnosis lacks systematicity and accuracy: In the existing operation and maintenance system, fault diagnosis often relies on single-parameter monitoring (such as judging electrical faults solely through voltage and current data), ignoring the synergistic relationship between the charging pile's structural condition (such as shell cracks or deformation), thermal distribution characteristics (such as localized hot spots or abnormal temperature gradients), and electrical parameters. This single-dimensional diagnostic approach easily leads to misdiagnosis or missed diagnosis. For example, judging equipment faults solely based on voltage fluctuations may overlook potential risks caused by shell deformation compressing internal circuitry, thus failing to accurately locate the fault type and root cause, increasing the cost of trial and error in operation and maintenance.

[0005] Operational and maintenance (O&M) decisions lack data support and dynamic adaptability: Current O&M strategies mostly adopt a "one-size-fits-all" fixed-cycle maintenance model, failing to develop differentiated solutions based on the actual health status and fault risk level of the charging piles. Some devices in good health are over-maintained due to fixed cycles, resulting in wasted resources; while devices with potential fault risks are not given priority attention, which may lead to serious faults, resulting in prolonged equipment downtime and affecting the user charging experience. At the same time, O&M resource scheduling lacks scientific basis; when multiple devices alarm simultaneously, it is difficult to quickly determine O&M priorities, easily causing delays in emergency fault handling.

[0006] The lack of closed-loop management in operations and maintenance (O&M) and the underutilization of data value are significant issues. The existing system fails to establish a complete closed loop encompassing data collection, fault diagnosis, O&M execution, and feedback. O&M operation records are not effectively linked to equipment operating data, making it impossible to optimize subsequent decisions based on historical O&M experience. For example, the failure to continuously track equipment recovery and conduct data review after fault handling leads to recurring similar faults and hinders the ability to improve the accuracy of fault diagnosis and O&M strategies through data accumulation and iteration.

[0007] In summary, the traditional charging pile operation and maintenance model can no longer meet the needs of large-scale and highly complex operations. It is necessary to build a system that integrates unmanned inspection, multi-dimensional fault diagnosis, and intelligent operation and maintenance decision-making to improve operation and maintenance efficiency, reduce fault losses, and ensure the stable and reliable operation of the charging pile network. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an unmanned inspection and automated operation and maintenance system for charging piles, which solves the problems mentioned in the background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an unmanned inspection and automatic operation and maintenance system for charging piles, comprising a data acquisition module, a status preliminary judgment module, a fault diagnosis module, and an automatic operation and maintenance module;

[0010] The data acquisition module is used to collect the operating data of the charging pile through drones and IoT sensors, and after data preprocessing, to identify the abnormal state of the charging pile and obtain the multimodal feature set of the charging pile based on the abnormal state.

[0011] The initial state assessment module is used to perform preliminary analysis of the charging pile's health status based on the multimodal feature set to obtain a health index. Based on health index The numerical diagnostic tool checks whether there is a fault risk in the current charging station; if so, it issues a diagnostic command.

[0012] The fault diagnosis module is used to extract relevant distinguishing features from the multimodal feature set after receiving a diagnosis command, and then diagnose the fault type of the current charging pile again based on the relevant distinguishing features to obtain a fault risk coefficient. ;

[0013] The automated operation and maintenance module is used to build an operation and maintenance decision model using machine learning technology, and to determine the fault risk coefficient. and health index The data is input into the operation and maintenance decision-making model, and after dimensionless processing, the operation and maintenance priority index is fitted and output. And it is presented in a visual format.

[0014] Preferably, the data acquisition module includes a UAV inspection unit, a sensor data acquisition unit, and a data preprocessing unit;

[0015] The drone inspection unit is used to conduct regular patrol scans of the charging pile using a visible light camera and an infrared thermal imager mounted on the drone. The patrol frequency is set to once every 2 hours, and each scan covers the charging pile itself and a 3-meter radius around it, collecting visual images and thermal imaging data of the charging pile.

[0016] The sensor data acquisition unit is used to collect electrical parameters of the charging pile in real time through IoT sensors, including voltage, current, temperature and power consumption, and record time series data according to timestamps;

[0017] The data preprocessing unit is used to process the acquired visual images, thermal imaging data, and electrical parameters to generate a multimodal feature set. Specific processing steps include:

[0018] Step 1.1: Denoising process. Gaussian filtering is used to eliminate salt-and-pepper noise in visual images, median filtering is used to smooth temperature fluctuations in thermal imaging data, and Kalman filtering is used to remove abnormal jump values ​​in electrical parameters.

[0019] Step 1.2: Normalization processing, using Z-score standardization for electrical parameters: ,in These are the original electrical parameter values. This is the average value of this parameter in the historical normal operation database. The reference standard deviation is used for thermal imaging data. Normalization: ,in This is the original temperature value. These represent the highest and lowest temperatures in a single frame of thermal imaging;

[0020] Step 1.3: Feature extraction to generate a multimodal feature set, which includes the structural, electrical, and thermal features of the charging pile; wherein:

[0021] Structural features: Shell deformation Average deformation Surface crack density ;

[0022] Electrical characteristics: average voltage Average current Rated voltage Actual voltage Maximum temperature Average power consumption ;

[0023] Thermal characteristics: hot spot area Temperature gradient .

[0024] Preferably, the initial status judgment module includes a health status analysis unit and an initial judgment unit;

[0025] The health status analysis unit is used to analyze the electrical and structural integrity of the charging pile based on the multimodal feature set to obtain a health index. The health index Obtain it using the following formula:

[0026] ;

[0027] in, The average voltage reflects the stability of the charging pile's power supply. The average current reflects the load balance. This is the highest temperature, reflecting the equipment's heating status; Average power consumption, reflecting the level of energy loss; Electrical-thermal characteristic weighting factor; The structural feature correction constant; the health index The value ranges from 0 to 100, with higher values ​​indicating better health of the charging station.

[0028] Preferably, the initial judgment unit is used to diagnose whether there is a fault risk in the current charging pile, and the specific diagnosis process includes:

[0029] Step 2.1: Health Threshold The setting is to extract 1000 normally operating charging piles from the historical database. The K-means clustering algorithm is used to generate a critical value between the normal and abnormal intervals, and this critical value is taken as the health threshold. ;

[0030] Step 2.2: Risk comparison, assessing the current health index of the charging station. and Compare:

[0031] like Upon diagnosing a potential fault risk in the current charging station, a diagnostic command is immediately sent to the fault diagnosis module, and a preliminary risk assessment report is simultaneously pushed to the operation and maintenance management terminal. The report includes... Numerical values ​​and features below the threshold;

[0032] like The system diagnoses that there is no current risk of failure in the charging pile, stores the current operating data in the historical database, updates the HI value every 30 minutes, and enters the next monitoring cycle.

[0033] Preferably, the fault diagnosis module includes a feature analysis unit and a fault identification unit;

[0034] The feature analysis unit is used to analyze the fault mode of the charging pile after receiving diagnostic commands. The specific steps include:

[0035] Step 3.1: Feature selection, calculate the Pearson correlation coefficient between each feature in the multimodal feature set and the occurrence of the fault. ,reserve The salient features;

[0036] Step 3.2: Anomaly coefficient calculation. Based on the screened features, calculate the structural anomaly coefficient SA and the electrical anomaly coefficient EA respectively:

[0037] The structural anomaly coefficient SA is obtained using the following formula:

[0038] ;

[0039] in, For the maximum deformation, The average deformation is the ratio of the two values, which reflects the degree of deformation concentration. Surface crack density, The baseline crack density; This is the deformation weighting factor, with a value ranging from 0.4 to 0.6; This is the crack weighting factor, with a value ranging from 0.4 to 0.6, and... Complementary, that is ; The value ranges from 0 to 1, with a larger value indicating a more severe structural anomaly.

[0040] Electrical abnormality coefficient Obtain it using the following formula:

[0041] ;

[0042] in, This is the actual voltage. The voltage is the rated voltage, and the relative value of the difference between the two reflects the degree of voltage deviation; For temperature gradient, As a reference temperature gradient; This is the voltage deviation weighting factor, with a value ranging from 0.5 to 0.7; This is the temperature gradient weighting factor, with a value ranging from 0.3 to 0.5, and... Complementary, that is ; The value ranges from 0 to 1, and the larger the value, the more serious the electrical abnormality.

[0043] Preferably, the fault identification unit is used to obtain the fault risk coefficient FR and diagnose the fault type, and the specific process includes:

[0044] Step 4.1: Dimensionless processing, converting structural anomaly coefficients... and electrical anomaly coefficient Mapping to a unified dimension: ,in These represent the minimum and maximum values ​​of historical structural anomalies, respectively. Similarly;

[0045] Step 4.2: Calculate the failure risk factor FR using the following formula:

[0046] ;

[0047] in, This represents the structural anomaly weight value, ranging from 0.3 to 0.5. This is the electrical anomaly weighting value, ranging from 0.5 to 0.7, and... Complementary, that is ; This is a correction constant; The value ranges from 0 to 1, with a larger value indicating a higher risk of failure.

[0048] Step 4.3: Weight calibration, recalculated quarterly based on newly added fault data. , ,make sure The calculation accuracy.

[0049] Preferably, the fault identification unit uses a fault threshold. The specific details of the diagnosed fault type are as follows:

[0050] Step 5.1: Fault Threshold The goal is to collect 500 different types of charging station malfunctions. The values ​​were analyzed using ROC curves, and those with sensitivity ≥95% and specificity ≥90% were selected. Value as fault threshold ;

[0051] Step 5.2: Fault type determination:

[0052] like The current charging station is diagnosed as having a serious fault. Typical fault scenarios include: For the shell crack, For internal short circuit, If the voltage deviation exceeds 10%, a Level 1 fault record is triggered.

[0053] like The current charging station's fault type is diagnosed as a minor fault, and the corresponding typical fault scenarios include: For slight deformation of the outer shell, For local temperature rise, A voltage deviation of 5%-10% triggers a level two fault record.

[0054] Step 5.3: Fault record storage, including fault type, Values ​​and abnormal features are stored in the fault database for subsequent training and optimization of the operation and maintenance decision model.

[0055] Preferably, the automated operation and maintenance module includes a modeling unit and an operation and maintenance execution unit;

[0056] The modeling unit is used to construct an operation and maintenance decision model, and the specific steps include:

[0057] Step 6.1: Dataset construction. Using the multimodal feature set as input features and the actual equipment recovery rate after operation and maintenance as the label, collect operation and maintenance data of 2000 charging piles and divide them into training set and test set in a 7:3 ratio.

[0058] Step 6.2: Model training. The initial model is built using the XGBoost algorithm. The parameters are set as follows: learning rate 0.05, tree depth 6, number of iterations 100. The mean squared error is used as the loss function. The test set accuracy is calculated every 10 epochs. Training is stopped when the test set accuracy improves by less than 0.5% for 5 consecutive epochs.

[0059] Step 6.3: Model validation and optimization. Use 5-fold cross-validation to evaluate model performance. The validation set accuracy should be ≥92% and the recall should be ≥90%. If the requirements are not met, add training data and retrain. Use the optimized model as the final operation and maintenance decision model.

[0060] The operation and maintenance execution unit is used to generate operation and maintenance strategies based on the operation and maintenance decision model. The strategies are divided into three categories: emergency operation and maintenance, planned operation and maintenance, and remote operation and maintenance.

[0061] Preferably, the operation and maintenance execution unit is used to fit and output the operation and maintenance priority index OP, and the specific process includes:

[0062] Step 7.1: Parameter Definition and Calculation:

[0063] Operations priority index (OP) is obtained using the following formula:

[0064] ;

[0065] in, This is a dynamic adjustment factor for the failure risk coefficient, used to balance... and The impact on operation and maintenance priority is calculated using the following formula: ,in It is the base of the natural logarithm; when hour, , Focusing on health status; when hour, , It focuses on basic health levels; For health index, Both are mapped to the 0-100 range to ensure dimensional consistency, serving as quantification values ​​for fault risk. This is a correction constant; The value ranges from 0 to 100, with a larger value indicating a higher priority for operation and maintenance.

[0066] Step 7.2: Validation of dynamic adjustment factors, using 500 cases of operation and maintenance data. The rationality requires The correlation coefficient with the actual urgency of maintenance should be ≥0.85; otherwise, adjustments should be made. The formula form was then re-verified.

[0067] Preferably, the operation and maintenance execution unit uses operation and maintenance thresholds. Assess the urgency of the maintenance and perform the necessary maintenance actions, as detailed below:

[0068] Step 8.1: Operation and Maintenance Threshold Based on the settings, and considering the operational resource costs and failure loss rate, a cost-based approach is adopted. Analysis and determination :when In such cases, the benefits of emergency maintenance typically outweigh the costs. ;

[0069] Step 8.2: Assessment and Execution of Operational Urgency:

[0070] like If the current charging station requires immediate maintenance, perform the following operations:

[0071] Location: Obtain the device location through the built-in GPS of the charging pile and generate an operation and maintenance route;

[0072] Dispatch: If the fault is remotely repairable, automatically send a remote restart command to the charging pile; if the fault is hardware fault, dispatch the maintenance robot or notify the maintenance personnel, and push a fault details report.

[0073] Closed-loop confirmation: Recalculate within 1 hour after maintenance is completed. and ,like and Mark as completed operation and maintenance; otherwise, trigger a second operation and maintenance.

[0074] like If it is determined that the current charging station does not require immediate maintenance, perform the following operations:

[0075] Regular monitoring: shorten the monitoring cycle to 1 hour / time, focusing on tracking the changing trends of abnormal characteristics;

[0076] Planned maintenance: based on The value determines the maintenance cycle. Maintenance is performed every 3 months. Maintenance should be performed once a month, and the maintenance content should be recorded.

[0077] Warning upgrade: If three consecutive monitoring tests are positive... If it continues to rise, then... The level was temporarily lowered by 10 to reassess the urgency of the maintenance.

[0078] Step 8.3: Update operation and maintenance data, including operation and maintenance records, and data before and after the operation and maintenance. Values ​​and equipment recovery status are stored in the operation and maintenance database for quarterly optimization of operation and maintenance decision model parameters.

[0079] This invention provides an unmanned inspection and automated operation and maintenance system for charging piles, which has the following beneficial effects:

[0080] 1. Achieve unmanned and efficient inspection, enhancing the comprehensiveness and real-time nature of monitoring: The system relies on drone inspection and IoT sensor technology to construct a three-dimensional data acquisition network covering the air and the ground. Drones can automatically complete the acquisition of visual and thermal images of charging piles at preset frequencies, and combine this with IoT sensors to obtain electrical parameters in real time, enabling high-frequency and wide-coverage equipment monitoring without human intervention. This unmanned inspection mode not only significantly reduces labor costs but also quickly captures multi-dimensional status information such as the structure, heat distribution, and electrical properties of charging piles, avoiding the subjective limitations and lag of manual inspections and ensuring timely detection of equipment anomalies.

[0081] 2. Multimodal Fusion Diagnosis: Enhancing Fault Identification Accuracy and Root Cause Location: The system overcomes the limitations of traditional single-parameter diagnosis by integrating structural, electrical, and thermal features of the charging pile through a data preprocessing module, forming a multimodal feature set. During fault diagnosis, significant features strongly correlated with the fault are first screened using Pearson correlation coefficients. Then, the structural anomaly coefficient and electrical anomaly coefficient are calculated collaboratively to accurately determine the fault type. This multi-dimensional fusion diagnostic approach effectively correlates abnormal signals from different dimensions of the equipment. For example, combining shell deformation with internal temperature gradient anomalies accurately locates short-circuit faults caused by structural compression, avoiding misjudgments and missed diagnoses, and providing accurate root cause information for subsequent maintenance.

[0082] 3. Intelligent O&M Decision-Making and Dynamic Priority Scheduling for Optimized Resource Allocation Efficiency: The system's O&M decision-making model, built upon machine learning, dynamically generates an O&M priority index (OP) by combining the charging pile health index (HI) and fault risk coefficient (FR), and formulates differentiated O&M strategies based on the OP value. For high-risk, high-urgency faults, emergency O&M processes are automatically triggered, dispatching O&M robots or personnel for priority handling; for low-risk equipment, planned maintenance is arranged to avoid over-maintenance. This data-driven dynamic decision-making model not only ensures rapid response to emergency faults but also maximizes the utilization of O&M resources, reduces unnecessary O&M costs, and improves overall O&M efficiency.

[0083] 4. Constructing a complete operation and maintenance closed loop for data-driven continuous optimization: The system forms a complete closed loop of "data collection - fault diagnosis - operation and maintenance execution - effect feedback - model iteration" through the automatic operation and maintenance module. After the operation and maintenance is completed, the system will conduct a secondary assessment of the equipment health status and fault risk, and store the operation and maintenance results in association with historical data for periodic optimization of operation and maintenance decision model parameters and fault diagnosis algorithms. Through this data review and model iteration mechanism, the system can continuously accumulate operation and maintenance experience, improve the accuracy of fault diagnosis and the adaptability of operation and maintenance strategies, realize the continuous evolution of operation and maintenance capabilities, gradually reduce equipment failure rate and downtime, and ensure the long-term stable operation of the charging pile network.

[0084] 5. Enhanced User Experience and Operational Economic Efficiency: Through unmanned inspection and rapid fault handling, the system significantly shortens the time for detecting and repairing equipment anomalies, reduces charging interruptions caused by charging pile malfunctions, and significantly improves the user charging experience. Simultaneously, precise fault diagnosis and differentiated operation and maintenance strategies reduce maintenance manpower and material costs, as well as excessive equipment maintenance losses, extending the lifespan of charging piles. This, in turn, enhances the operator's economic efficiency and market competitiveness, providing strong support for the large-scale, high-quality operation of new energy vehicle charging infrastructure. Attached Figure Description

[0085] Figure 1 This is a block diagram illustrating the principle of the unmanned inspection and automatic maintenance system for charging piles described in this invention.

[0086] Figure 2 This is a schematic diagram illustrating the principle of the unmanned inspection and automatic operation and maintenance system for charging piles described in this invention. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] like Figures 1-2 As shown, this invention provides a technical solution: an unmanned inspection and automatic operation and maintenance system for charging piles, including a data acquisition module, a preliminary status judgment module, a fault diagnosis module, and an automatic operation and maintenance module. The data acquisition module collects operational data from charging piles via drones and IoT sensors, and performs data preprocessing to identify abnormal states of the charging piles, obtaining a multimodal feature set based on these abnormal states. The preliminary status judgment module performs a preliminary analysis of the charging pile's health status based on the multimodal feature set to obtain a health index. Based on health index The numerical diagnostic tool determines whether the current charging station has a fault risk; if so, it issues a diagnostic command. Upon receiving the diagnostic command, the fault diagnosis module extracts relevant distinguishing features from the multimodal feature set and, based on these features, re-diagnoses the fault type of the current charging station to obtain a fault risk coefficient. The automated operation and maintenance module is used to construct an operation and maintenance decision model using machine learning technology, and to determine the fault risk coefficient. and health index The data is input into the operation and maintenance decision-making model, and after dimensionless processing, the operation and maintenance priority index is fitted and output. And it is presented in a visual format.

[0089] More specifically, the data acquisition module includes a drone inspection unit, a sensor data acquisition unit, and a data preprocessing unit;

[0090] The drone inspection unit is used to periodically patrol and scan the charging pile using a visible light camera (resolution not less than 4K) and an infrared thermal imager (temperature measurement range -20℃-150℃) mounted on the drone. The patrol frequency is set to once every 2 hours. Each scan covers the charging pile itself and a 3-meter radius around it, and collects visual images (frame rate 25fps) and thermal imaging data (sampling interval 1s) of the charging pile.

[0091] The sensor data acquisition unit is used to collect electrical parameters of the charging pile in real time through IoT sensors (sampling frequency 50Hz), including voltage (measurement range 0-1000V, accuracy ±0.5%), current (measurement range 0-50A, accuracy ±0.3%), temperature (measurement range -40℃-125℃, accuracy ±0.2℃) and power consumption (measurement range 0-20kW, accuracy ±1%), and record time series data according to timestamps (accurate to milliseconds);

[0092] The data preprocessing unit is used to process the acquired visual images, thermal imaging data, and electrical parameters to generate a multimodal feature set. Specific processing steps include:

[0093] Step 1.1: Denoising process, apply Gaussian filtering to the visual image (kernel size 3×3, standard deviation...) To eliminate salt-and-pepper noise, median filtering (window size 5×5) was used on thermal imaging data to smooth temperature fluctuations, and Kalman filtering (process noise covariance Q=0.01, observation noise covariance R=0.1) was used on electrical parameters to remove anomalous jump values.

[0094] Step 1.2: Normalization processing, using Z-score standardization for electrical parameters: ,in These are the original electrical parameter values. This is the average value of this parameter in the historical normal operation database (such as voltage). Current ), For reference standard deviation (such as voltage) Current ); Use thermal imaging data Normalization: ,in This is the original temperature value. These represent the highest and lowest temperatures in a single frame of thermal imaging;

[0095] Step 1.3: Feature extraction to generate a multimodal feature set, which includes the structural, electrical, and thermal features of the charging pile; wherein:

[0096] Structural features: Shell deformation (Maximum shell offset calculated using a visual image edge detection algorithm, unit: mm), average deformation. (Calculate the arithmetic mean of the offsets of all sampling points on the outer shell, in mm), surface crack density (The ratio of total crack length to shell surface area, in units) Crack recognition employs Canny edge detection combined with morphological operations.

[0097] Electrical characteristics: average voltage (Arithmetic mean of voltage data over 10 minutes), average current (Arithmetic mean of current data over 10 minutes), rated voltage (Charging pile factory rated voltage, such as 220V / 380V), actual voltage (Instantaneous voltage value collected in real time), highest temperature (Highest temperature value collected by the sensor within 10 minutes), average power consumption (Arithmetic mean of power consumption data over 10 minutes);

[0098] Thermal characteristics: hot spot area (Temperature exceeds threshold in thermal imaging) (e.g., area of ​​a region at 60℃, unit: cm²) and temperature gradient. (Temperature difference between the hot spot area and the surrounding normal area, in °C).

[0099] The drone inspection unit, equipped with a visible light camera with a resolution of at least 4K and an infrared thermal imager with a temperature measurement range of -20℃ to 150℃, scans the charging pile and its surrounding 3-meter range every 2 hours. It acquires visual images at a frame rate of 25fps and thermal imaging data at 1-second intervals, efficiently capturing structural anomalies such as deformation of the charging pile shell and surface cracks, as well as thermal distribution anomalies such as local hot spots. This solves the problems of limited coverage and insufficient identification of hidden defects in traditional manual inspections. The sensor data acquisition unit collects electrical parameters such as voltage (0~1000V, ±0.5% accuracy), current (0~50A, ±0.3% accuracy), temperature (-40℃~125℃, ±0.2℃ accuracy), and power consumption (0~20kW, ±1% accuracy) in real time through IoT sensors at a sampling frequency of 50Hz. It also records time-series data with millisecond-level timestamps to ensure the real-time and accurate acquisition of electrical parameters, providing a data foundation for equipment electrical status analysis. The data preprocessing unit generates a multimodal feature set through three steps.

[0100] The first step uses a 3×3 filter core. Gaussian filtering eliminates salt-and-pepper noise in visual images, and 5×5 windowed mid-range filtering smooths temperature fluctuations in thermal imaging data. Kalman filtering is used to remove abnormal fluctuations in electrical parameters, ensuring data validity.

[0101] The second step is to use Z-score standardization for electrical parameters based on the mean and standard deviation of historical normal data, and to use min-max normalization for thermal imaging data based on the highest and lowest temperatures of a single frame image, so as to achieve uniformity of the dimensions of different types of data.

[0102] The third step extracts structural features (Dmax, Davg, Cd), electrical features (Vavg, Iavg, Vnom, Vact, Tmax, Pavg), and thermal features (Ha, Tg) to form a multimodal feature set. This transforms the data from raw acquisition to feature-based and standardized processing, providing structured data for subsequent modules to analyze equipment status through multi-dimensional feature fusion. This improves the overall comprehensiveness, accuracy, and usability of the system's data acquisition, laying a reliable data foundation for subsequent fault diagnosis and maintenance decisions.

[0103] More specifically, the initial status judgment module includes a health status analysis unit and an initial judgment unit;

[0104] The health status analysis unit is used to analyze the electrical and structural integrity of the charging pile based on the multimodal feature set to obtain a health index. The health index Obtain it using the following formula:

[0105] ;

[0106] in, The average voltage reflects the stability of the charging pile's power supply. The average current reflects the load balance. This is the highest temperature, reflecting the equipment's heating status; Average power consumption, reflecting the level of energy loss; The electrical-thermal characteristic weighting factor (value range 0.6-0.8, calibrated using AHP combined with historical fault data from 1000 charging piles, with a larger value taken when electrical faults account for a high proportion). A structural feature correction constant (value range 10-20, 20 for intact structure, 15 for slight deformation, and 10 for severe deformation); the health index The value ranges from 0 to 100, with higher values ​​indicating better health of the charging station.

[0107] The health status analysis unit takes the multimodal feature set generated by the data acquisition module as input and focuses on key electrical and structural features to construct a health index. The calculation model, through formulas Average voltage (Reflecting power supply stability), average current (Reflects load balance), highest temperature (Reflecting heating status), average power consumption The correlation between electrical-thermal characteristics (reflecting energy loss) and structural characteristics is quantified, among which... (0.6-0.8) Calibration was performed using the Analytic Hierarchy Process (AHP) combined with historical fault data from 1000 charging piles (the larger value was taken when the proportion of electrical faults was high). (10-20) Dynamically adjust based on the structural integrity (20 for structural integrity, 15 for slight deformation, and 10 for severe deformation), ultimately outputting a value in the range of 0-100. The value (higher values ​​indicate better health) transforms multi-dimensional features into a single health indicator, intuitively quantifying the overall health level of the equipment. This design avoids the limitations of traditional single-parameter assessments and adapts to different fault scenarios through dynamic calibration of weights and correction constants, ensuring... The value can accurately reflect the electrical stability and structural integrity of the equipment, providing a reliable quantitative basis for the initial judgment unit to quickly distinguish the "normal / potential fault" status of the equipment, effectively reducing the initiation of meaningless fault diagnosis processes and improving the overall operating efficiency of the system.

[0108] More specifically, the initial judgment unit is used to diagnose whether there is a fault risk in the current charging pile, and the specific diagnostic process includes:

[0109] Step 2.1: Health Threshold The setup involves extracting 1000 normally functioning charging piles (with no fault records in the past 6 months) from the historical database. The value is used to generate a critical value between the normal and abnormal intervals using the K-means clustering algorithm (number of clusters = 2), and this critical value is taken as the health threshold. (generally ,Right now (When identified as a potential anomaly)

[0110] Step 2.2: Risk comparison, assessing the current health index of the charging station. and Compare:

[0111] like Upon diagnosing a potential fault risk in the current charging station, a diagnostic command is immediately sent to the fault diagnosis module, and a preliminary risk assessment report is simultaneously pushed to the operation and maintenance management terminal. The report includes... Numerical values, features below the threshold (such as...) Exceeding the standard (Too large)

[0112] like The current charging station is diagnosed as not having any fault risk. The current operating data ( The HI value (including key electrical parameters) is stored in the historical database and updated every 30 minutes before entering the next monitoring cycle.

[0113] The initial assessment unit, as the execution link for rapid fault risk screening, achieves efficient initial screening of charging pile fault risks through standardized threshold settings and risk comparison processes. This ensures that potential faults are promptly triggered for in-depth diagnosis, while avoiding redundant maintenance processes for healthy equipment. The results are as follows:

[0114] The initial assessment unit uses the health index output by the health status analysis unit. Based on this as the core principle, we first construct scientific health thresholds through step 2.1. Extract 1,000 normal charging piles with no fault records in the past 6 months from the historical database. The value is determined using the K-means clustering algorithm with a cluster size of 2, which automatically generates the critical values ​​for the normal and abnormal intervals. (generally =60), this threshold setting method is based on large-scale real normal equipment data, avoiding the subjectivity of manually setting thresholds and ensuring the accuracy of the threshold in distinguishing between "normal" and "abnormal" states; then, risk comparison is performed through step 2.2: if the current < Immediately send a diagnostic command to the fault diagnosis module and push information including HI values ​​and abnormal feature items (such as...). Exceeding the standard (Excessive risk reports) ensure potential faults quickly proceed to in-depth diagnostics; if ≥ The runtime data is stored in a historical database and updated every 30 minutes. The value then proceeds to the next round of monitoring, preventing healthy devices from consuming fault diagnosis resources. The overall process achieves "precise identification and tiered response" of fault risks, improving the efficiency of potential fault detection and optimizing system resource allocation, laying the foundation for the accurate activation of subsequent fault diagnosis modules.

[0115] More specifically, the fault diagnosis module includes a feature analysis unit and a fault identification unit;

[0116] The feature analysis unit is used to analyze the fault mode of the charging pile after receiving diagnostic commands. The specific steps include:

[0117] Step 3.1: Feature selection, calculate the Pearson correlation coefficient between each feature in the multimodal feature set and the occurrence of the fault. ,reserve Significant features (such as) With the shell broken , With internal short circuit );

[0118] Step 3.2: Anomaly coefficient calculation. Based on the screened features, calculate the structural anomaly coefficient SA and the electrical anomaly coefficient EA respectively:

[0119] The structural anomaly coefficient SA is obtained using the following formula:

[0120] ;

[0121] in, This represents the maximum deformation (in mm). The average deformation (in mm) is the average deformation amount, and the ratio of the two reflects the degree of deformation concentration. Surface crack density (unit: mm / mm²). The benchmark crack density (the factory standard value for the same model of charging pile, usually...) ); The deformation weighting factor has a value of 0.4-0.6 (determined based on historical structural failure data; 0.6 is used when the proportion of failures caused by deformation is high). This is the crack weighting factor, with a value ranging from 0.4 to 0.6, and... Complementary, that is ; The value ranges from 0 to 1, with a larger value indicating a more severe structural anomaly.

[0122] Electrical abnormality coefficient Obtain it using the following formula:

[0123] ;

[0124] in, This is the actual voltage. The voltage is the rated voltage, and the relative value of the difference between the two reflects the degree of voltage deviation; For temperature gradient, The reference temperature gradient (the normal temperature difference under rated operating conditions, typically...) ); This is the voltage deviation weighting factor, with a value of 0.5-0.7 (0.7 is used when the proportion of faults caused by voltage abnormalities is high). This is the temperature gradient weighting factor, with a value ranging from 0.3 to 0.5, and... Complementary, that is ; The value ranges from 0 to 1, and the larger the value, the more serious the electrical abnormality.

[0125] As the core component for accurately locating fault types, the fault diagnosis module uses a feature analysis unit to filter fault-related features and quantify the degree of anomaly. This provides a focused and standardized analytical basis for subsequent fault identification, effectively solving the misjudgment problems caused by feature redundancy and single-dimensionality in traditional fault diagnosis. The effects are as follows:

[0126] The feature analysis unit uses the diagnostic command sent by the initial state assessment module as a trigger signal. It first performs feature filtering in step 3.1: calculating the Pearson correlation coefficient r between each feature in the multimodal feature set and the occurrence of the fault, and retaining only... Significant characteristics ≥0.3 (such as maximum deformation of the shell) With the shell broken =0.82, temperature gradient With internal short circuit =0.75), this step can eliminate redundant features with weak correlation to the fault, focus on key abnormal signals, reduce subsequent calculations and improve diagnostic accuracy; then, the abnormality coefficient is calculated dimensionally in step 3.2: for the structural state, based on the filtered , , Features, through formula Calculate the structural anomaly coefficient (Values ​​range from 0 to 1), where Reflects the degree of deformation concentration ( The crack density (typically 0.001 mm / mm²) is the factory-standard crack density for the same model, reflecting the degree of crack deviation from the standard. and (0.4-0.6, complementary) Dynamically adjusted based on historical structural failure data (when deformation failures account for a high proportion). (Taking 0.6), to quantify the degree of structural anomaly; for electrical conditions, based on , , Features, through formula Calculate electrical anomaly coefficient (Values ​​range from 0 to 1), where Reflects relative voltage deviation, ( The reference temperature gradient (typically 5°C) under rated operating conditions reflects the degree of temperature gradient anomaly. and (0.5-0.7 / 0.3-0.5, complementary) dynamically adjusted according to the proportion of electrical faults (when the proportion of voltage abnormality faults is high). (Take 0.7) to quantify the degree of electrical anomaly.

[0127] The overall design transforms multimodal features into structured anomaly coefficients with values ​​ranging from 0 to 1 through "feature screening and multidimensional anomaly quantification". This enables independent evaluation of structural and electrical anomalies and adapts to different fault scenarios through weighting factors, laying a reliable foundation for the fault identification unit to accurately determine the fault type.

[0128] More specifically, the fault identification unit is used to obtain the fault risk coefficient FR and diagnose the fault type, and the specific process includes:

[0129] Step 4.1: Dimensionless processing, converting structural anomaly coefficients... and electrical anomaly coefficient Mapping to a unified dimension: ,in These are the minimum and maximum values ​​of historical structural anomalies (usually...). ), Similarly;

[0130] Step 4.2: Calculate the failure risk factor FR using the following formula:

[0131] ;

[0132] in, This is the structural anomaly weight value, ranging from 0.3 to 0.5 (determined based on the probability of structural and electrical faults occurring; for example, if structural faults account for 30%). ); This is the electrical anomaly weighting value, ranging from 0.5 to 0.7, and... Complementary, that is ; A correction constant (valued between 0.01 and 0.05, used to compensate for errors in the feature selection process, calibrated using 500 failure cases); The value ranges from 0 to 1, with a larger value indicating a higher risk of failure.

[0133] Step 4.3: Weight calibration, recalculated quarterly based on newly added fault data (no less than 50 cases). , ,make sure The calculation accuracy is ≥90%.

[0134] As the core execution link for accurately quantifying fault risks, the fault identification unit achieves collaborative assessment of structural and electrical anomalies and accurate quantification of fault risks through a standardized process of dimensionless processing, risk coefficient calculation, and dynamic weight calibration. This provides a scientific basis for subsequent operation and maintenance decisions, with the following results:

[0135] The fault identification unit outputs the structural anomaly coefficients from the feature analysis unit. Electrical abnormality coefficient For input, first proceed through step 4.1 to... , Mapping to a unified dimension: based on the minimum (usually 0) and maximum (usually 1) values ​​of historical structural / electrical anomaly data, through... Dimensionless processing is performed to eliminate dimensional differences between different coefficients, ensuring the validity of subsequent calculations; then, the failure risk coefficient is calculated in step 4.2. : with complementary weights (0.3-0.5, set according to the probability of structural failure, such as when the proportion is 30%) =0.3), (0.5-0.7) Weighted fusion and The correction constants were superimposed and calibrated using 500 failure cases. (0.01-0.05) Compensate for feature selection error, and finally output a value between 0 and 1. (The higher the value, the higher the risk), thus quantifying the failure risk; finally, in step 4.3, the risk is recalibrated quarterly using no fewer than 50 new failure data points. , ,make sure Prediction accuracy is ≥90%. The overall process not only solves the problem of inconsistency between structural and electrical anomalies, but also improves accuracy through dynamic weighting and error compensation. The calculation accuracy provides precise and reliable risk quantification indicators for subsequent fault type determination and maintenance priority allocation.

[0136] More specifically, the fault identification unit uses a fault threshold. The specific details of the diagnosed fault type are as follows:

[0137] Step 5.1: Fault Threshold The goal is to collect 500 different types of charging pile malfunctions (200 serious malfunctions and 300 minor malfunctions). The values ​​were analyzed using ROC curves (labeled by the actual severity of the fault), and those with sensitivity ≥95% and specificity ≥90% were selected. Value as fault threshold (generally );

[0138] Step 5.2: Fault type determination:

[0139] like The current charging station is diagnosed as having a serious fault. Typical fault scenarios include: For the shell crack, For internal short circuit, If the voltage deviation exceeds 10%, a Level 1 fault record is triggered (marked as "emergency handling").

[0140] like The current charging station's fault type is diagnosed as a minor fault, and the corresponding typical fault scenarios include: For slight deformation of the outer shell, For local temperature rise, When the voltage deviation is 5%-10%, a level two fault record is triggered (marked as "planned handling").

[0141] Step 5.3: Fault record storage, including fault type, Values ​​and abnormal features are stored in the fault database for subsequent training and optimization of the operation and maintenance decision model.

[0142] The fault identification unit achieves accurate classification and data accumulation of the severity of charging pile faults through standardized fault threshold settings, type determination, and record storage processes, providing key support for the formulation of differentiated operation and maintenance strategies and continuous system optimization.

[0143] The fault identification unit first constructs a high-reliability fault threshold through step 5.1: collecting 500 different types of charging pile faults, including 200 severe faults and 300 minor faults. Values ​​were used to perform ROC curve analysis, with the actual severity of the fault as the label. Values ​​with sensitivity ≥95% and specificity ≥90% were selected. The value is used as a fault threshold (usually) =0.6), this threshold setting is based on large-scale real fault data, taking into account both the accuracy of fault identification and the risk control of missed and false judgments, ensuring the accuracy of distinguishing the severity of faults; then, step 5.2 is used to perform fault type determination: if > It is determined to be a serious fault (such as...) >5mm shell crack, Internal short circuit, If the voltage deviation exceeds 10% and triggers a Level 1 "Emergency Handling" record, ≤ The fault is classified as minor (e.g., 2mm <). Slight deformation of the outer shell ≤5mm Local temperature rise, The voltage deviation is 5%-10%, triggering a secondary "planned processing" record to achieve precise matching of fault type and processing priority; finally, step 5.3 is used to classify the fault type, Values ​​and abnormal features are stored in the fault database to provide data support for the training and optimization of the subsequent operation and maintenance decision model.

[0144] The overall process not only ensures the accuracy of fault type identification and the targeted nature of operation and maintenance response, but also lays the foundation for system iteration and upgrade through data accumulation, effectively improving fault handling efficiency and overall system operation and maintenance capabilities.

[0145] More specifically, the automated operation and maintenance module includes a modeling unit and an operation and maintenance execution unit;

[0146] The modeling unit is used to construct an operation and maintenance decision model, and the specific steps include:

[0147] Step 6.1: Dataset construction. Using a multimodal feature set (structural features, electrical features, thermal features) as input features and the actual equipment recovery rate after operation and maintenance (recovery rate = number of devices restored to normal operation / total number of devices under operation and maintenance) as labels, collect operation and maintenance data of 2000 charging piles (each device contains at least 10 data records), and divide them into a training set (1400 units) and a test set (600 units) in a 7:3 ratio.

[0148] Step 6.2: Model training. The initial model is built using the XGBoost algorithm. The parameters are set as follows: learning rate 0.05, tree depth 6, number of iterations 100. The mean squared error is used as the loss function. The test set accuracy is calculated every 10 epochs. Training is stopped when the test set accuracy improves by less than 0.5% for 5 consecutive epochs.

[0149] Step 6.3: Model validation and optimization. Use 5-fold cross-validation to evaluate model performance. The validation set accuracy should be ≥92% and the recall should be ≥90%. If the requirements are not met, add training data (supplement with data from no less than 300 devices) and retrain. Use the optimized model as the final operation and maintenance decision model.

[0150] The operation and maintenance execution unit is used to generate operation and maintenance strategies based on the operation and maintenance decision model. The strategies are divided into three categories: emergency operation and maintenance (for serious faults), planned operation and maintenance (for minor faults), and remote operation and maintenance (for faults that can be repaired remotely, such as parameter configuration errors).

[0151] The automated operation and maintenance module constructs a high-precision operation and maintenance decision model through the modeling unit and generates differentiated operation and maintenance strategies through the operation and maintenance execution unit, thereby realizing intelligent operation and maintenance decision-making and precise operation and maintenance execution, effectively solving the problems of resource waste and response delay caused by the "one-size-fits-all" approach of traditional operation and maintenance strategies.

[0152] The modeling unit uses a multimodal feature set (structural, electrical, and thermal features) as input and device recovery rate as the label. It collects operation and maintenance data from 2000 charging piles (≥10 records per pile) and divides it into a training set (1400 piles) and a test set (600 piles) in a 7:3 ratio. The XGBoost algorithm (learning rate 0.05, tree depth 6, 100 iterations) is used to build the initial model. The mean squared error is used as the loss function, and the training process is controlled by the rule of "accuracy is calculated every 10 epochs, and the training stops if the improvement is <0.5% for 5 consecutive epochs". The model is then optimized by 5-fold cross-validation (requiring accuracy ≥92% and recall ≥90%; if not met, ≥300 pieces of equipment data are added for retraining) to ensure that the operation and maintenance decision model has high reliability and adaptability. Based on this model, the operation and maintenance execution unit generates emergency operation and maintenance strategies for serious faults, planned operation and maintenance strategies for minor faults, and remote operation and maintenance strategies for faults that can be remotely repaired (such as parameter configuration errors), achieving accurate matching of "fault type - operation and maintenance strategy". The overall design ensures intelligent operation and maintenance decision-making through scientific data construction and model training, and achieves efficient allocation of operation and maintenance resources through differentiated strategies, which greatly improves the speed of operation and maintenance response and equipment recovery efficiency, providing strong support for the stable operation of the charging pile network.

[0153] More specifically, the operation and maintenance execution unit is used to fit and output the operation and maintenance priority index OP, and the specific process includes:

[0154] Step 7.1: Parameter Definition and Calculation:

[0155] Operations priority index (OP) is obtained using the following formula:

[0156] ;

[0157] in, This is a dynamic adjustment factor for the failure risk coefficient, used to balance... and The impact on operation and maintenance priority is calculated using the following formula: ,in The base of the natural logarithm ( );when In case of (serious malfunction), , Focusing on health status ( The lower the level, the more urgent maintenance is needed. (When there is no risk of failure) , It focuses on basic health levels; Health index (value 0-100). The fault risk quantification value (ranging from 0 to 100) is mapped to the 0-100 range to ensure dimensional consistency. To correct the constant (value 0.02-0.08), calibration was performed using 100 operational cases to ensure... The distinguishability is defined as the difference between operations and maintenance priorities of OPs ≥ 5. The value ranges from 0 to 100, with a larger value indicating a higher priority for operation and maintenance.

[0158] Step 7.2: Validation of dynamic adjustment factors, using 500 cases of operation and maintenance data. The rationality requires The correlation coefficient with the actual urgency of maintenance should be ≥0.85; otherwise, adjustments should be made. Formula form (e.g.) And re-verify.

[0159] The operations and maintenance execution unit uses a standardized operations and maintenance priority index. The calculation process enables precise quantification of the urgency of charging pile operation and maintenance, providing a core basis for the scientific scheduling of operation and maintenance resources and the implementation of differentiated operation and maintenance strategies, and effectively solving the problems of subjective priority judgment and disordered resource scheduling in traditional operation and maintenance.

[0160] The operation and maintenance execution unit is first defined and calculated in step 7.1. Based on formula Introducing dynamic adjustment factors ( Balanced Health Index (0-100, reflecting the basic health level of the equipment) and the quantitative value of failure risk 100× The impact of (0-100, reflecting the degree of fault risk) when =1 (serious fault) =1, Emphasis ( The lower the maintenance cost, the more urgent the situation); when =0 (no risk of failure) ≈0.367, OP focuses on basic health levels, while using a correction constant calibrated through 100 operational cases. (0.02-0.08) Ensure different priorities A difference ≥ 5 is required to achieve the desired result. Effective differentiation (0-100, higher values ​​indicate higher priority); then verify the rationality of F(FR) through step 7.2: using 500 cases of operation and maintenance data for testing. The correlation coefficient with the actual urgency of maintenance should be ≥0.85; otherwise, adjustments should be made. Formula form (e.g.) Re-verification ensures the accuracy of OP's reflection of maintenance priorities. The entire process is achieved through dynamic adjustment factors. and The collaborative weighting, combined with the correction constant and rationality verification, improves The high accuracy of the calculations provides objective and reliable quantitative indicators for subsequent assessment of the urgency of operations and maintenance and resource scheduling, significantly improving the efficiency of operations and maintenance decisions and the rationality of resource allocation.

[0161] More specifically, the operation and maintenance execution unit uses operation and maintenance thresholds. Assess the urgency of the maintenance and perform the necessary maintenance actions, as detailed below:

[0162] Step 8.1: Operation and Maintenance Threshold The settings are based on operation and maintenance resource costs (such as operation and maintenance robot scheduling costs and labor costs) and failure loss rates (such as equipment downtime losses and safety accident losses), and adopt cost-based management. Analysis and determination :when In such cases, the benefits of emergency maintenance (reduced losses due to failure) usually outweigh the costs. ;

[0163] Step 8.2: Assessment and Execution of Operational Urgency:

[0164] like If the current charging station requires immediate maintenance, perform the following operations:

[0165] Location: The device location is obtained through the built-in GPS of the charging pile (accuracy ≤10m), and an operation and maintenance route is generated (the shortest route is selected first, avoiding traffic congestion areas).

[0166] Dispatch: For remotely repairable faults (such as parameter drift or software lag), automatically send a remote restart command to the charging station (monitor the HI value within 5 minutes after restarting; if...) If the maintenance is completed, then the maintenance is complete. If it is a hardware failure (such as a cracked casing or damaged components), dispatch the maintenance robot (response time ≤ 30 minutes) or notify maintenance personnel (arrive on site within 1 hour) and push a fault details report (including fault type and model of the component to be replaced).

[0167] Closed-loop confirmation: Recalculate within 1 hour after maintenance is completed. and ,like and Mark as completed operation and maintenance; otherwise, trigger a second operation and maintenance.

[0168] like If it is determined that the current charging station does not require immediate maintenance, perform the following operations:

[0169] Regular monitoring: Shorten the monitoring cycle to once every hour, focusing on tracking abnormal characteristics (such as...) , The changing trend of );

[0170] Planned maintenance: based on The value determines the maintenance cycle. Maintenance is performed every 3 months. Perform maintenance once a month and record the maintenance content (such as cleaning, tightening parts, parameter calibration).

[0171] Warning upgrade: If three consecutive monitoring tests are positive... If it continues to rise (each rise ≥5), then... Temporarily lowered by 10 (i.e.) ), reassess the urgency of operations and maintenance;

[0172] Step 8.3: Update operation and maintenance data. Store operation and maintenance records, HI / FR values ​​before and after operation and maintenance, and equipment recovery status into the operation and maintenance database for quarterly optimization of operation and maintenance decision model parameters (such as learning rate and tree depth).

[0173] The operation and maintenance execution unit achieves accurate assessment of the urgency of operation and maintenance and efficient scheduling of operation and maintenance resources through scientific setting of operation and maintenance thresholds, hierarchical operation and maintenance execution and data update process. At the same time, it forms an operation and maintenance closed loop to continuously optimize system performance and effectively solves the problems of resource waste, response delay and insufficient exploitation of data value in traditional operation and maintenance.

[0174] The operation and maintenance execution unit first sets the operation and maintenance threshold through step 8.1, based on the cost analysis of operation and maintenance resource costs (such as robot scheduling and labor costs) and failure loss rates (such as downtime and security incident losses). , ensure when > (generally When the threshold is set to 70%, the benefits of emergency maintenance outweigh the costs, ensuring the economic efficiency and rationality of the threshold setting; then, maintenance operations are performed in stages as per step 8.2: if... > It generates the optimal path through GPS positioning (accuracy ≤10m) and automatically sends a restart command for remotely repairable faults (verified within 5 minutes). ≥ The system will either complete the task, dispatch a hardware failure dispatch robot (response time ≤ 30 minutes) or notify personnel (arrive within 1 hour), and conduct a re-evaluation after 1 hour. / (If both standards are met, the process is marked as complete; otherwise, secondary maintenance is required), enabling rapid closure of emergency faults; if ≤ Shorten the monitoring cycle to 1 hour / time to track abnormal characteristics, and determine the maintenance cycle according to the HI value. ≥80 once every 3 months, 60≤ <80 once a month), and Temporary reduction when three consecutive increases of ≥5. Reassess up to 60 to prevent the fault from worsening; finally, proceed with step 8.3 to review the maintenance records. / Changes and equipment recovery status are stored in the database for quarterly optimization of model parameters (such as learning rate and tree depth).

[0175] The overall process not only achieves precise matching and efficient utilization of operation and maintenance resources, but also promotes continuous system iteration through closed-loop management and data accumulation, ensuring the stable operation of the charging pile network and reducing the overall operation and maintenance costs.

[0176] In summary, this invention constructs an unmanned, three-dimensional inspection network using drones and IoT sensors, achieves accurate fault diagnosis through multimodal feature fusion, generates dynamic maintenance priorities and formulates differentiated strategies based on machine learning models, and forms a complete closed loop of operation and maintenance, encompassing data collection, diagnosis, execution, feedback, and optimization. This not only significantly improves the real-time performance of inspections and the accuracy of fault identification, optimizes the efficiency of operation and maintenance resource allocation, and reduces operation and maintenance costs, but also reduces equipment downtime, improves the user charging experience, and allows for continuous iteration and optimization of operation and maintenance capabilities through data accumulation, providing strong support for the stable and reliable operation of the charging pile network.

[0177] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A charging pile operation unmanned inspection and automatic maintenance system, characterized in that, It includes a data acquisition module, a preliminary status assessment module, a fault diagnosis module, and an automated operation and maintenance module; The data acquisition module is used to collect the operating data of the charging pile through drones and IoT sensors, and after data preprocessing, to identify the abnormal state of the charging pile and obtain the multimodal feature set of the charging pile based on the abnormal state. The initial state assessment module is used to perform preliminary analysis of the charging pile's health status based on the multimodal feature set to obtain a health index. Based on health index The numerical diagnostic tool checks whether there is a fault risk in the current charging station; if so, it issues a diagnostic command. The fault diagnosis module is used to extract relevant distinguishing features from the multimodal feature set after receiving a diagnosis command, and then diagnose the fault type of the current charging pile again based on the relevant distinguishing features to obtain a fault risk coefficient. ; The automated operation and maintenance module is used to build an operation and maintenance decision model using machine learning technology, and to determine the fault risk coefficient. and health index The data is input into the operation and maintenance decision-making model, and after dimensionless processing, the operation and maintenance priority index is fitted and output. And it is presented in a visual format.

2. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 1, characterized in that, The data acquisition module includes a drone inspection unit, a sensor data acquisition unit, and a data preprocessing unit. The drone inspection unit is used to conduct regular patrol scans of the charging pile using a visible light camera and an infrared thermal imager mounted on the drone. The patrol frequency is set to once every 2 hours, and each scan covers the charging pile itself and a 3-meter radius around it, collecting visual images and thermal imaging data of the charging pile. The sensor data acquisition unit is used to collect electrical parameters of the charging pile in real time through IoT sensors, including voltage, current, temperature and power consumption, and record time series data according to timestamps; The data preprocessing unit is used to process the acquired visual images, thermal imaging data, and electrical parameters to generate a multimodal feature set. Specific processing steps include: Step 1.1: Denoising process. Gaussian filtering is used to eliminate salt-and-pepper noise in visual images, median filtering is used to smooth temperature fluctuations in thermal imaging data, and Kalman filtering is used to remove abnormal jump values ​​in electrical parameters. Step 1.2: Normalization processing, using Z-score standardization for electrical parameters: ,in These are the original electrical parameter values. This is the average value of this parameter in the historical normal operation database. The reference standard deviation is used for thermal imaging data. Normalization: ,in This is the original temperature value. These represent the highest and lowest temperatures in a single frame of thermal imaging; Step 1.3: Feature extraction to generate a multimodal feature set, which includes the structural, electrical, and thermal features of the charging pile; wherein: Structural features: Shell deformation Average deformation Surface crack density ; Electrical characteristics: average voltage Average current Rated voltage Actual voltage Maximum temperature Average power consumption ; Thermal characteristics: hot spot area Temperature gradient .

3. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 2, characterized in that, The initial status assessment module includes a health status analysis unit and an initial assessment unit; The health status analysis unit is used to analyze the electrical and structural integrity of the charging pile based on the multimodal feature set to obtain a health index. The health index Obtain it using the following formula: ； in, The average voltage reflects the stability of the charging pile's power supply. The average current reflects the load balance. This is the highest temperature, reflecting the equipment's heating status; This represents average power consumption, reflecting the level of energy loss. Electrical-thermal characteristic weighting factor; The structural feature correction constant; the health index The value ranges from 0 to 100, with higher values ​​indicating better health of the charging station.

4. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 3, characterized in that, The initial assessment unit is used to diagnose whether the current charging pile has a fault risk. The specific diagnostic process includes: Step 2.1: Health Threshold The setting is to extract 1000 normally operating charging piles from the historical database. The K-means clustering algorithm is used to generate a critical value between the normal and abnormal intervals, and this critical value is taken as the health threshold. ; Step 2.2: Risk comparison, assessing the current health index of the charging station. and Compare: like Upon diagnosing a potential fault risk in the current charging station, a diagnostic command is immediately sent to the fault diagnosis module, and a preliminary risk assessment report is simultaneously pushed to the operation and maintenance management terminal. The report includes... Numerical values ​​and features below the threshold; like The system diagnoses that there is no current risk of failure in the charging pile, stores the current operating data in the historical database, updates the HI value every 30 minutes, and enters the next monitoring cycle.

5. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 4, characterized in that, The fault diagnosis module includes a feature analysis unit and a fault identification unit; The feature analysis unit is used to analyze the fault mode of the charging pile after receiving diagnostic commands. The specific steps include: Step 3.1: Feature selection, calculate the Pearson correlation coefficient between each feature in the multimodal feature set and the occurrence of the fault. ,reserve The salient features; Step 3.2: Anomaly coefficient calculation. Based on the screened features, calculate the structural anomaly coefficient SA and the electrical anomaly coefficient EA respectively: The structural anomaly coefficient SA is obtained using the following formula: ； in, For the maximum deformation, The average deformation is the ratio of the two values, which reflects the degree of deformation concentration. Surface crack density, The baseline crack density; This is the deformation weighting factor, with a value ranging from 0.4 to 0.6; This is the crack weighting factor, with a value ranging from 0.4 to 0.6, and... Complementary, that is ; The value ranges from 0 to 1, with a larger value indicating a more severe structural anomaly. Electrical abnormality coefficient Obtain it using the following formula: ； in, This is the actual voltage. The voltage is the rated voltage, and the relative value of the difference between the two reflects the degree of voltage deviation; For temperature gradient, As a reference temperature gradient; This is the voltage deviation weighting factor, with a value ranging from 0.5 to 0.7; This is the temperature gradient weighting factor, with a value ranging from 0.3 to 0.5, and... Complementary, that is ; The value ranges from 0 to 1, and the larger the value, the more serious the electrical abnormality.

6. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 5, characterized in that, The fault identification unit is used to obtain the fault risk coefficient FR and diagnose the fault type. The specific process includes: Step 4.1: Dimensionless processing, converting structural anomaly coefficients... and electrical anomaly coefficient Mapping to a unified dimension: ,in These represent the minimum and maximum values ​​of historical structural anomalies, respectively. Similarly; Step 4.2: Calculate the failure risk factor FR using the following formula: ； in, This represents the structural anomaly weight value, ranging from 0.3 to 0.

5. This is the electrical anomaly weighting value, ranging from 0.5 to 0.7, and... Complementary, that is ; This is a correction constant; The value ranges from 0 to 1, with a larger value indicating a higher risk of failure. Step 4.3: Weight calibration, recalculated quarterly based on newly added fault data. , ,make sure The calculation accuracy.

7. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 6, characterized in that, The fault identification unit uses a fault threshold. The specific details of the diagnosed fault type are as follows: Step 5.1: Fault Threshold The goal is to collect 500 different types of charging station malfunctions. The values ​​were analyzed using ROC curves, and those with sensitivity ≥95% and specificity ≥90% were selected. Value as fault threshold ; Step 5.2: Fault type determination: like The current charging station is diagnosed as having a serious fault. Typical fault scenarios include: For the shell crack, For internal short circuit, If the voltage deviation exceeds 10%, a Level 1 fault record is triggered. like The current charging station's fault type is diagnosed as a minor fault, and the corresponding typical fault scenarios include: For slight deformation of the outer shell, For local temperature rise, A voltage deviation of 5%-10% triggers a level two fault record. Step 5.3: Fault record storage, including fault type, Values ​​and abnormal features are stored in the fault database for subsequent training and optimization of the operation and maintenance decision model.

8. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 7, characterized in that, The automated operation and maintenance module includes a modeling unit and an operation and maintenance execution unit; The modeling unit is used to construct an operation and maintenance decision model, and the specific steps include: Step 6.1: Dataset construction. Using the multimodal feature set as input features and the actual equipment recovery rate after operation and maintenance as the label, collect operation and maintenance data of 2000 charging piles and divide them into training set and test set in a 7:3 ratio. Step 6.2: Model training. The initial model is built using the XGBoost algorithm. The parameters are set as follows: learning rate 0.05, tree depth 6, number of iterations 100. The mean squared error is used as the loss function. The test set accuracy is calculated every 10 epochs. Training is stopped when the test set accuracy improves by less than 0.5% for 5 consecutive epochs. Step 6.3: Model validation and optimization. Use 5-fold cross-validation to evaluate model performance. The validation set accuracy should be ≥92% and the recall should be ≥90%. If the requirements are not met, add training data and retrain. Use the optimized model as the final operation and maintenance decision model. The operation and maintenance execution unit is used to generate operation and maintenance strategies based on the operation and maintenance decision model. The strategies are divided into three categories: emergency operation and maintenance, planned operation and maintenance, and remote operation and maintenance.

9. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 8, characterized in that, The operation and maintenance execution unit is used to fit and output the operation and maintenance priority index OP. The specific process includes: Step 7.1: Parameter Definition and Calculation: Operations priority index (OP) is obtained using the following formula: ； in, This is a dynamic adjustment factor for the failure risk coefficient, used to balance... and The impact on operation and maintenance priority is calculated using the following formula: ,in It is the base of the natural logarithm; when hour, , Focusing on health status; when hour, , It focuses on basic health levels; For health index, Both are mapped to the 0-100 range to ensure dimensional consistency, serving as quantification values ​​for fault risk. This is a correction constant; The value ranges from 0 to 100, with a larger value indicating a higher priority for operation and maintenance. Step 7.2: Validation of dynamic adjustment factors, using 500 cases of operation and maintenance data. The rationality requires The correlation coefficient with the actual urgency of maintenance should be ≥0.85; otherwise, adjustments should be made. The formula form was then re-verified.

10. The unmanned inspection and automatic operation and maintenance system for charging piles according to claim 9, characterized in that, The operation and maintenance execution unit uses operation and maintenance thresholds. Assess the urgency of the maintenance and perform the necessary maintenance actions, as detailed below: Step 8.1: Operation and Maintenance Threshold Based on the settings, and considering the operational resource costs and failure loss rate, a cost-based approach is adopted. Analysis and determination :when In such cases, the benefits of emergency maintenance typically outweigh the costs. ; Step 8.2: Assessment and Execution of Operational Urgency: like If the current charging station requires immediate maintenance, perform the following operations: Location: Obtain the device location through the built-in GPS of the charging pile and generate an operation and maintenance route; Dispatch: If the fault is remotely repairable, automatically send a remote restart command to the charging pile; if the fault is hardware fault, dispatch the maintenance robot or notify the maintenance personnel, and push a fault details report. Closed-loop confirmation: Recalculate within 1 hour after maintenance is completed. and ,like and Mark as completed operation and maintenance; otherwise, trigger a second operation and maintenance. like If it is determined that the current charging station does not require immediate maintenance, perform the following operations: Regular monitoring: shorten the monitoring cycle to 1 hour / time, focusing on tracking the changing trends of abnormal characteristics; Planned maintenance: based on The value determines the maintenance cycle. Maintenance is performed every 3 months. Maintenance should be performed once a month, and the maintenance content should be recorded. Warning upgrade: If three consecutive monitoring tests are positive... If it continues to rise, then... The level was temporarily lowered by 10 to reassess the urgency of the maintenance. Step 8.3: Update operation and maintenance data, including operation and maintenance records, and data before and after the operation and maintenance. Values ​​and equipment recovery status are stored in the operation and maintenance database for quarterly optimization of operation and maintenance decision model parameters.

Images