Charging pile remote operation and maintenance system based on Internet of Things

Abstract

The invention discloses a charging pile remote operation and maintenance system based on the Internet of Things, and the system comprises a data collection and edge calculation module which is used for collecting multi-mode operation data of a charging pile, and generating standardized operation data; the health degree dynamic evaluation module is used for executing multi-modal feature coupling analysis on the standardized operation data; the fault root cause analysis module is used for starting a time-space coupled fault root cause analysis process to obtain a root cause confidence score and a high-risk component; the self-healing strategy execution module is used for generating a self-healing strategy of the charging pile; the resource dynamic scheduling module is used for realizing dynamic optimal distribution of operation and maintenance resources according to the health degree score and the geographic position; the intelligent operation and maintenance management module is used for updating the weight of the dynamic pruning random forest model and generating a hierarchical maintenance report; and the data full-process control module is used for completing automatic closed-loop management from data acquisition to operation and maintenance decision. According to the invention, remote intelligent diagnosis and self-healing operation and maintenance of the charging pile can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things and energy management, and in particular to a charging pile remote operation and maintenance system based on the Internet of Things. Background Art

[0002] With the increasing popularity of new energy vehicles, charging piles, as critical infrastructure, are attracting widespread attention for their operational stability and maintenance efficiency. Existing charging pile operation and maintenance systems often rely on IoT technology, using sensors to collect basic operational data and upload it to cloud platforms for remote monitoring, alerting, and management. Some systems use data visualization to assist in operational decision-making, and some platforms are also initially employing machine learning methods to analyze the health status of charging piles, aiming to enhance intelligent operation and maintenance.

[0003] Existing technologies still have significant deficiencies in multimodal perception, intelligent diagnosis, and resource scheduling. Most systems only collect a single or small amount of data, such as electrical parameters, and lack the fusion analysis of multi-source data such as mechanical, thermal, and environmental data. This makes it difficult to fully reflect the operating status of the charging pile, affecting the accuracy of fault diagnosis and operation and maintenance efficiency. Even though some solutions have introduced new sensors such as infrared and vibration, multimodal data often lack high-precision synchronization and effective fusion, resulting in limited fault identification capabilities and prominent false alarm and missed alarm problems. At the same time, although some systems have built analysis models, they generally have insufficient adaptive capabilities and are unable to cope with environmental changes and dynamic changes in abnormal patterns during the long-term operation of charging piles, resulting in delayed or inaccurate health status assessment results. Common algorithms are mostly traditional machine learning methods such as static decision trees and support vector machines. Model updates usually rely on manual intervention and lack adaptive evolution capabilities.

[0004] In terms of fault handling and strategy optimization, the current remote operation and maintenance of charging piles mainly relies on manual dispatching and empirical judgment, lacking a spatiotemporal coupled root cause analysis mechanism and a self-healing strategy generation and closed-loop optimization process. At the same time, due to the wide distribution of charging infrastructure and the complex operating environment, traditional operation and maintenance resource scheduling is often based on static partitioning or the principle of "nearest priority", which makes it difficult to respond to changes in on-site demand and achieve efficient resource matching in a timely manner, resulting in delayed maintenance responses and high operation and maintenance costs. In addition, existing systems generally have the problem of fragmented data processes on multiple platforms. There is a lack of automated closed-loop coordination in the collection, analysis, decision-making and operation and maintenance execution links, and data redundancy and insufficient traceability, which affect the value mining of operation and maintenance data and the improvement of equipment reliability. Some solutions do not pay enough attention to data tiered storage and security protection, making it difficult to meet the needs of data governance, process traceability and intelligent operation and maintenance in large-scale IoT scenarios.

[0005] Therefore, how to propose a remote operation and maintenance system for charging piles based on the Internet of Things is an urgent problem that those skilled in the art need to solve. Summary of the Invention

[0006] One objective of the present invention is to propose an IoT-based remote operation and maintenance system for charging piles. By comprehensively leveraging hardware topology awareness protocols, multimodal feature coupling analysis, a dynamically pruned random forest model, digital twin simulation, and intelligent resource scheduling, this system systematically implements dynamic assessment of charging pile health, intelligent root cause analysis, automatic generation of self-healing strategies, and efficient and optimized allocation of operation and maintenance resources. Compared to existing technologies, this system can dynamically adapt to equipment health management in complex operating environments, offering advantages such as high intelligence, rapid response, and low operation and maintenance costs.

[0007] The remote operation and maintenance system for charging piles based on the Internet of Things according to an embodiment of the present invention includes:

[0008] The data acquisition and edge computing module is used to collect multimodal operating data of charging piles, align the timestamps of the multimodal operating data, and eliminate invalid data to generate standardized operating data;

[0009] A dynamic health assessment module, which performs multimodal feature coupling analysis on standardized operating data and generates a health score for charging piles using an improved dynamic pruning random forest model;

[0010] The fault root cause analysis module is used to initiate the spatiotemporal coupled fault root cause analysis process and obtain the root cause confidence score and high-risk components;

[0011] The self-healing strategy execution module is used to generate self-healing strategies for charging piles. When a strategy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making security.

[0012] A dynamic resource scheduling module is used to dynamically optimize the allocation of O&M resources based on health scores and geographic location. This includes a geo-fence-driven scheduling algorithm that matches mobile energy storage vehicles or backup charging stations, and calculates the cost and benefit weights of resource scheduling in real time.

[0013] Intelligent operation and maintenance management module, used to update the weights of the dynamically pruned random forest model and generate hierarchical maintenance reports;

[0014] The data full-process control module is used to complete automated closed-loop management from data collection to operation and maintenance decision-making.

[0015] Optionally, modules can be connected using the following methods:

[0016] S1. Collect multimodal operating data of charging piles, align the timestamps of the multimodal operating data using the hardware topology awareness protocol, remove invalid data, and generate standardized operating data;

[0017] S2. Perform multimodal feature coupling analysis on standardized operating data and generate a health score for the charging pile using an improved dynamic pruning random forest model;

[0018] S3: When the health score of the charging pile is lower than the set dynamic threshold, the spatiotemporal coupled fault root cause analysis process is automatically started to obtain the root cause confidence score and high-risk components;

[0019] S4. For high-risk components, a hybrid decision-making framework is established. Under normal operating conditions, a spatiotemporal layered dual-Q network is used to generate self-healing strategies. When the charging pile health score is lower than 30, the system switches to digital twin simulation verification mode. When a policy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making safety.

[0020] S5. Based on the health score of the charging pile and the geo-fence information, a resource scheduling optimization algorithm is applied to achieve dynamic allocation of mobile energy storage vehicles or backup charging piles;

[0021] S6. Using incremental learning to update the weights of the dynamic pruning random forest model, and generating a hierarchical maintenance report through an expert network;

[0022] S7. Manage operation and maintenance data through structuring and indexing, and implement a tiered data storage strategy.

[0023] Optionally, step S1 specifically includes:

[0024] S11. Collect electrical operating parameters of the charging pile using a current / voltage harmonic sensor. The electrical operating parameters include current, voltage, current harmonic distortion rate, voltage harmonic distortion rate, and instantaneous power fluctuation variance. Perform time domain analysis on the obtained current data to obtain the number of current mutations within a given sampling period. Use a narrowband Internet of Things communication module to upload the electrical operating parameters to a cloud platform in real time.

[0025] S12. Regularly collect temperature distribution data of the power module using an infrared thermal imaging module. Use a hardware topology-based thermal field alignment algorithm to map the temperature distribution data to the corresponding locations on the charging pile's physical structure. Upload the complete temperature field distribution every 30 seconds via the 5G network. Further, calculate the temperature gradient based on the temperature differences at each monitoring point on the physical structure.

[0026] S13. Collect mechanical vibration signals using a vibration accelerometer, extract mechanical wear energy characteristics using a 0.5-2 kHz bandpass filter, generate a vibration energy index, and aggregate and transmit the vibration energy indexes of all sampling points through a long-distance wide-area gateway;

[0027] S14, collecting temperature and humidity data through environmental sensors, and synchronizing electrical operating parameters with temperature and humidity data using the IEEE1588 precision time protocol;

[0028] S15. Use a hardware topology awareness protocol to align the timestamps of the multimodal operation data, and eliminate invalid data based on a preset threshold to form standardized operation data.

[0029] Optionally, step S2 specifically includes:

[0030] S21. Calculate the dynamic correlation coefficient between the current harmonic distortion rate and the temperature gradient using the Pearson correlation coefficient;

[0031] S22, fusing the vibration energy index and the number of current mutations by weighted linear superposition to generate a mechanical wear index;

[0032] S23. Inputting the dynamic correlation coefficient, mechanical wear index, and standardized operating data into an improved dynamic pruning random forest model to obtain a health score for the charging pile. The improved dynamic pruning random forest model introduces a three-level pruning mechanism to automatically adjust the maximum depth of each decision tree in the model based on the CPU utilization of the edge node.

[0033] The three-level pruning mechanism is as follows: when the CPU utilization rate is higher than 80%, the emergency mode is activated, the maximum depth of the decision tree is compressed to 5 layers, and the feature selector is enabled to retain only the top-10 features; when the CPU utilization rate is between 50% and 80%, the balanced mode is entered, the depth is limited to 10 layers, and the dynamic top-20 features are used; when the CPU utilization rate is lower than 50%, the full mode is run, the full depth of 15 layers is maintained, and all 42-dimensional features are activated;

[0034] S24. Apply time decay correction to the health score of the charging pile, and increase the score weight of abnormal events that occur in the most recent time window by 20%.

[0035] Optionally, step S3 specifically includes:

[0036] S31. Based on the standardized operation data, calculate the spatial similarity between the electrical operation parameter characteristics of the charging pile and the electrical operation parameter characteristics in the historical operation and maintenance cases. At the same time, calculate the temporal similarity between the current temperature gradient data and the historical temperature gradient data, and obtain the multimodal similarity score S by weighting. sim :

[0037]

[0038] Among them, V elec and V hist are the current and historical electrical characteristic vectors, T grad and T hist is the current and historical temperature gradient distribution, JSD is the Jensen-Shannon divergence;

[0039] S32. If the multimodal similarity score S sim When the preset threshold is reached, the suspected root cause of the corresponding component and the corresponding score result are directly output based on historical cases;

[0040] S33. If the comprehensive similarity score is lower than the threshold, a component-level relationship graph of the charging pile is constructed based on the current multi-source operation data. The propagation links of the abnormal signal between the components are automatically analyzed through a three-layer heterogeneous graph convolutional network to infer the potential root cause.

[0041] S34. Assign a root cause confidence score to each component based on the multimodal similarity score or the inference output of the heterogeneous graph convolutional network, and mark components with a confidence score greater than or equal to 80% as high-risk components.

[0042] Optionally, step S4 specifically includes:

[0043] S41. For high-risk components, a hybrid decision-making framework is established. The hybrid decision-making framework includes: using a spatiotemporal layered dual-Q network to generate a self-healing strategy under normal operating conditions; automatically switching to a digital twin simulation verification mode when the charging pile health score is below 30; and activating an expert rule-based fuse mechanism to ensure decision safety when a policy conflict is detected;

[0044] S42. Construct a spatiotemporal layered dual-Q network structure, specifically: the spatial layer includes an upper-layer global state network and a lower-layer local feature network, and the temporal layer includes a fast response network and a steady-state optimization network;

[0045] The upper global state network adopts a multi-layer perceptron structure, inputs the global operation characteristics of the charging station level, and outputs the global self-healing action Q value. The lower local feature network adopts a three-layer convolutional neural network structure, inputs the multimodal local characteristics of the charging pile, and outputs the corresponding local self-healing action Q value.

[0046] The fast response network processes short-term features in real time through a lightweight feedforward neural network. The steady-state optimization network uses a recurrent neural network and combines historical operation data of charging piles to generate self-healing action recommendations for long-term health optimization and output the optimal self-healing strategy through multi-scale fusion.

[0047] S43. Design a time-varying compound reward function to dynamically weight and integrate health improvement rewards, maintenance cost penalties, and safety constraints according to time:

[0048] R t =μ1(t)·R health +μ2(t)·R cost +μ3(t)·R safety ;

[0049] where R health To predict the improvement of health, Rcost is the maintenance cost of the charging pile, R safety is the safety constraint term, μ1(t), μ2(t), μ3(t) are the reward weights that change over time;

[0050] S44. The improved Huber loss function is used to optimize the weights of the spatiotemporal layered dual-Q network. The loss threshold δ(n) decreases linearly with the number of training rounds. The loss function formula is:

[0051]

[0052] Where Q represents the state-action value output by the Q learning model, y represents the target value calculated based on the reward and the next state, and δ(n) decreases with each training round;

[0053] S45. Based on the trained and optimized spatiotemporal layered dual-Q network, generate the optimal self-healing strategy a under each charging pile operation state s. * , that is, output the self-healing action that maximizes the Q value. The formula is:

[0054]

[0055] Where s represents the current state, a represents an optional action, and Q represents the cumulative reward expected to be obtained by selecting action a in the current state s;

[0056] S46. When the health score of the charging pile is lower than 30, it automatically switches to the digital twin simulation verification mode, pre-simulates the generated self-healing strategy through the digital twin sandbox, and adopts the ∈-greedy decision method to select the self-healing action, that is, the currently known optimal self-healing action is selected with probability 1-∈, and any executable self-healing action is randomly selected with probability ∈. The formula is:

[0057]

[0058] Among them,∈ represents the exploration probability;

[0059] S47. After passing the digital twin sandbox verification, the system further detects whether the self-healing strategy conflicts with existing operation and maintenance actions. If a policy conflict is detected, the expert rule-based circuit breaker mechanism is activated to suspend the issuance and execution of the self-healing strategy. When the self-healing strategy is verified by the expert rules to have no conflict, it is automatically converted into a specific operation and maintenance instruction and issued to the device for execution;

[0060] S48. During the implementation of the self-healing strategy, the health score and operating data of the charging pile are continuously monitored in real time. If the health score does not increase by 20% within 30 minutes, the self-healing operation will be automatically withdrawn immediately and restored to the state before the strategy execution.

[0061] The expert rule-based circuit breaker mechanism is as follows: before the self-healing strategy is issued, a multi-dimensional detection is performed on the self-healing strategy to be executed based on the preset expert knowledge base and operation and maintenance safety rules. If any conflict, risk or unreasonable situation is found in the operation, the execution of the strategy is automatically terminated.

[0062] Optionally, step S5 specifically includes:

[0063] S51. Based on the IoT platform, a 5km radius geo-fence is dynamically constructed with the target charging pile as the center. The geo-fence then screens and matches the available mobile energy storage vehicles or backup charging pile operation and maintenance resources within the fence in real time.

[0064] S52. Calculate the service interruption impact index for the selected operation and maintenance resources based on the current service demand, and optimize the resource allocation priority based on the service interruption impact index;

[0065] S53. Based on the scheduling results, the augmented reality maintenance instructions are pushed to the on-site operation and maintenance personnel terminal, and the faulty components are visualized through the digital twin model.

[0066] S54. Automatically summarize the full-process analysis and execution data, generate a visual maintenance report containing the optimal maintenance time window and fault root cause analysis, and push it to the operation and maintenance terminal and user APP in real time.

[0067] Optionally, step S6 specifically includes:

[0068] S61. Collect and analyze charging pile operation data after executing the self-healing strategy, and use incremental learning to update the weights of the dynamic pruning random forest model;

[0069] S62. Build an expert network that integrates multi-class data features and continuously optimizes health status assessment and diagnosis models by learning different types of operating samples;

[0070] S63. Generate a hierarchical maintenance report to display the health status of the charging pile to the user end, and provide maintenance information including spare parts replacement priority to the operation and maintenance end.

[0071] Optionally, step S7 specifically includes:

[0072] S71. Use structured data encapsulation to manage operation and maintenance process metadata in a JSON-based linked data format, and establish time series indexes and geo-hash-coded spatial indexes for all event data.

[0073] S72. Set a directed acyclic graph scheduling rule to support parallel execution of data collection and health assessment tasks;

[0074] S73. Implement a tiered data storage strategy, storing hot data in the Redis database, warm data for normal operation and maintenance queries in the Influx time series database, and moving cold data to MinIO object storage.

[0075] S74. Regularly generate Merkle tree hashes for logs, indicators, and decision data.

[0076] The beneficial effects of the present invention are:

[0077] First, through multi-source sensor and data fusion, this invention enables real-time sensing of multimodal operating data from charging piles, encompassing electrical, mechanical, thermal, and environmental data. This significantly overcomes the shortcomings of existing technologies, which rely on a single or limited number of parameters and struggle to accurately reflect the actual status of the equipment. Second, the system utilizes a random forest algorithm based on dynamic pruning and multimodal feature coupling to effectively improve the accuracy, timeliness, and adaptability of health status assessment and fault diagnosis, enabling precise localization and dynamic response to charging pile faults under complex operating conditions. Furthermore, the proposed spatiotemporal layered dual-Q network, digital twin sandbox simulation, expert rule-based fusing, and full-process feedback mechanism comprehensively ensure the safety and effectiveness of the self-healing strategy, significantly enhancing the intelligence and automation of self-healing operations. By integrating geo-fencing, resource optimization and scheduling, and hierarchical maintenance reporting, the system achieves efficient coordination of operation and maintenance resources and closed-loop management of the entire process. In summary, this invention significantly enhances the operational reliability, remote intelligent operation and maintenance automation, and energy management efficiency of charging piles, providing solid technical support for the safe, economical, and low-carbon operation of large-scale charging infrastructure. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0079] Figure 1 This is a flow chart of the remote operation and maintenance system for charging piles based on the Internet of Things proposed by the present invention;

[0080] Figure 2 This is a flow chart of charging pile operation data standardization in the present invention;

[0081] Figure 3 This is a flowchart of the remote intelligent operation and maintenance and self-healing decision-making of the charging pile in the present invention. DETAILED DESCRIPTION

[0082] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0083] refer to Figure 1-3, the remote operation and maintenance system of charging piles based on the Internet of Things includes:

[0084] The data acquisition and edge computing module is used to collect multimodal operating data of charging piles, align the timestamps of the multimodal operating data, and eliminate invalid data to generate standardized operating data;

[0085] A dynamic health assessment module, which performs multimodal feature coupling analysis on standardized operating data and generates a health score for charging piles using an improved dynamic pruning random forest model;

[0086] The fault root cause analysis module is used to initiate the spatiotemporal coupled fault root cause analysis process and obtain the root cause confidence score and high-risk components;

[0087] The self-healing strategy execution module is used to generate self-healing strategies for charging piles. When a strategy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making security.

[0088] A dynamic resource scheduling module is used to dynamically optimize the allocation of O&M resources based on health scores and geographic location. This includes a geo-fence-driven scheduling algorithm that matches mobile energy storage vehicles or backup charging stations, and calculates the cost and benefit weights of resource scheduling in real time.

[0089] Intelligent operation and maintenance management module, used to update the weights of the dynamically pruned random forest model and generate hierarchical maintenance reports;

[0090] The data full-process control module is used to complete automated closed-loop management from data collection to operation and maintenance decision-making.

[0091] In this embodiment, the modules are connected through the following methods:

[0092] S1. Collect multimodal operating data of charging piles, align the timestamps of the multimodal operating data using the hardware topology awareness protocol, remove invalid data, and generate standardized operating data;

[0093] S2. Perform multimodal feature coupling analysis on standardized operating data and generate a health score for the charging pile using an improved dynamic pruning random forest model;

[0094] S3: When the health score of the charging pile is lower than the set dynamic threshold, the spatiotemporal coupled fault root cause analysis process is automatically started to obtain the root cause confidence score and high-risk components;

[0095] S4. For high-risk components, a hybrid decision-making framework is established. Under normal operating conditions, a spatiotemporal layered dual-Q network is used to generate self-healing strategies. When the charging pile health score is lower than 30, the system switches to digital twin simulation verification mode. When a policy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making safety.

[0096] S5. Based on the health score of the charging pile and the geo-fence information, a resource scheduling optimization algorithm is applied to achieve dynamic allocation of mobile energy storage vehicles or backup charging piles;

[0097] S6. Using incremental learning to update the weights of the dynamic pruning random forest model, and generating a hierarchical maintenance report through an expert network;

[0098] S7. Manage operation and maintenance data through structuring and indexing, and implement a tiered data storage strategy.

[0099] In this embodiment, step S1 specifically includes:

[0100] S11. Collect electrical operating parameters of the charging pile using a current / voltage harmonic sensor. The electrical operating parameters include current, voltage, current harmonic distortion rate, voltage harmonic distortion rate, and instantaneous power fluctuation variance. Perform time domain analysis on the obtained current data to obtain the number of current mutations within a given sampling period. Use a narrowband Internet of Things communication module to upload the electrical operating parameters to a cloud platform in real time.

[0101] S12. Regularly collect temperature distribution data of the power module using an infrared thermal imaging module. Use a hardware topology-based thermal field alignment algorithm to map the temperature distribution data to the corresponding locations on the charging pile's physical structure. Upload the complete temperature field distribution every 30 seconds via the 5G network. Further, calculate the temperature gradient based on the temperature differences at each monitoring point on the physical structure.

[0102] S13. Collect mechanical vibration signals using a vibration accelerometer, extract mechanical wear energy characteristics using a 0.5-2 kHz bandpass filter, generate a vibration energy index, and aggregate and transmit the vibration energy indexes of all sampling points through a long-distance wide-area gateway;

[0103] S14, collecting temperature and humidity data through environmental sensors, and synchronizing electrical operating parameters with temperature and humidity data using the IEEE1588 precision time protocol;

[0104] S15. Use a hardware topology awareness protocol to align the timestamps of the multimodal operation data, and eliminate invalid data based on a preset threshold to form standardized operation data.

[0105] In this embodiment, step S2 specifically includes:

[0106] S21. Calculate the dynamic correlation coefficient between the current harmonic distortion rate and the temperature gradient using the Pearson correlation coefficient;

[0107] S22, fusing the vibration energy index and the number of current mutations by weighted linear superposition to generate a mechanical wear index;

[0108] S23. Inputting the dynamic correlation coefficient, mechanical wear index, and standardized operating data into an improved dynamic pruning random forest model to obtain a health score for the charging pile. The improved dynamic pruning random forest model introduces a three-level pruning mechanism to automatically adjust the maximum depth of each decision tree in the model based on the CPU utilization of the edge node.

[0109] The three-level pruning mechanism is as follows: when the CPU utilization rate is higher than 80%, the emergency mode is activated, the maximum depth of the decision tree is compressed to 5 layers, and the feature selector is enabled to retain only the top-10 features; when the CPU utilization rate is between 50% and 80%, the balanced mode is entered, the depth is limited to 10 layers, and the dynamic top-20 features are used; when the CPU utilization rate is lower than 50%, the full mode is run, the full depth of 15 layers is maintained, and all 42-dimensional features are activated;

[0110] S24. Apply time decay correction to the health score of the charging pile, and increase the score weight of abnormal events that occur in the most recent time window by 20%.

[0111] In this embodiment, step S3 specifically includes:

[0112] S31. Based on the standardized operation data, calculate the spatial similarity between the electrical operation parameter characteristics of the charging pile and the electrical operation parameter characteristics in the historical operation and maintenance cases. At the same time, calculate the temporal similarity between the current temperature gradient data and the historical temperature gradient data, and obtain the multimodal similarity score S by weighting. sim :

[0113]

[0114] Among them, V elec and V hist are the current and historical electrical characteristic vectors, T grad and T hist is the current and historical temperature gradient distribution, JSD is the Jensen-Shannon divergence;

[0115] S32. If the multimodal similarity score S sim When the preset threshold is reached, the suspected root cause of the corresponding component and the corresponding score result are directly output based on historical cases;

[0116] S33. If the comprehensive similarity score is lower than the threshold, a component-level relationship graph of the charging pile is constructed based on the current multi-source operation data. The propagation links of the abnormal signal between the components are automatically analyzed through a three-layer heterogeneous graph convolutional network to infer the potential root cause.

[0117] S34. Assign a root cause confidence score to each component based on the multimodal similarity score or the inference output of the heterogeneous graph convolutional network, and mark components with a confidence score greater than or equal to 80% as high-risk components.

[0118] In this embodiment, step S4 specifically includes:

[0119] S41. For high-risk components, a hybrid decision-making framework is established. The hybrid decision-making framework includes: using a spatiotemporal layered dual-Q network to generate a self-healing strategy under normal operating conditions; automatically switching to a digital twin simulation verification mode when the charging pile health score is below 30; and activating an expert rule-based fuse mechanism to ensure decision safety when a policy conflict is detected;

[0120] S42. Construct a spatiotemporal layered dual-Q network structure, specifically: the spatial layer includes an upper-layer global state network and a lower-layer local feature network, and the temporal layer includes a fast response network and a steady-state optimization network;

[0121] The upper global state network adopts a multi-layer perceptron structure, inputs the global operation characteristics of the charging station level, and outputs the global self-healing action Q value. The lower local feature network adopts a three-layer convolutional neural network structure, inputs the multimodal local characteristics of the charging pile, and outputs the corresponding local self-healing action Q value.

[0122] The fast response network processes short-term features in real time through a lightweight feedforward neural network. The steady-state optimization network uses a recurrent neural network and combines historical operation data of charging piles to generate self-healing action recommendations for long-term health optimization and output the optimal self-healing strategy through multi-scale fusion.

[0123] S43. Design a time-varying compound reward function to dynamically weight and integrate health improvement rewards, maintenance cost penalties, and safety constraints according to time:

[0124] R t =μ1(t)·R health +μ2(t)·R cost +μ3(t)·R safety ;

[0125] where R health To predict the improvement of health, R cost is the maintenance cost of the charging pile, R safety is the safety constraint term, μ1(t), μ2(t), μ3(t) are the reward weights that change over time;

[0126] S44. The improved Huber loss function is used to optimize the weights of the spatiotemporal layered dual-Q network. The loss threshold δ(n) decreases linearly with the number of training rounds. The loss function formula is:

[0127]

[0128] Where Q represents the state-action value output by the Q learning model, y represents the target value calculated based on the reward and the next state, and δ(n) decreases with each training round;

[0129] S45. Based on the trained and optimized spatiotemporal layered dual-Q network, generate the optimal self-healing strategy a under each charging pile operation state s. * , that is, output the self-healing action that maximizes the Q value. The formula is:

[0130]

[0131] Where s represents the current state, a represents an optional action, and Q represents the cumulative reward expected to be obtained by selecting action a in the current state s;

[0132] S46. When the health score of the charging pile is lower than 30, it automatically switches to the digital twin simulation verification mode, pre-simulates the generated self-healing strategy through the digital twin sandbox, and adopts the ∈-greedy decision method to select the self-healing action, that is, the currently known optimal self-healing action is selected with probability 1-∈, and any executable self-healing action is randomly selected with probability ∈. The formula is:

[0133]

[0134] Among them,∈ represents the exploration probability;

[0135] S47. After passing the digital twin sandbox verification, the system further detects whether the self-healing strategy conflicts with existing operation and maintenance actions. If a policy conflict is detected, the expert rule-based circuit breaker mechanism is activated to suspend the issuance and execution of the self-healing strategy. When the self-healing strategy is verified by the expert rules to have no conflict, it is automatically converted into a specific operation and maintenance instruction and issued to the device for execution;

[0136] S48. During the implementation of the self-healing strategy, the health score and operating data of the charging pile are continuously monitored in real time. If the health score does not increase by 20% within 30 minutes, the self-healing operation will be automatically withdrawn immediately and restored to the state before the strategy execution.

[0137] The expert rule-based circuit breaker mechanism is as follows: before the self-healing strategy is issued, a multi-dimensional detection is performed on the self-healing strategy to be executed based on the preset expert knowledge base and operation and maintenance safety rules. If any conflict, risk or unreasonable situation is found in the operation, the execution of the strategy is automatically terminated.

[0138] In this embodiment, step S5 specifically includes:

[0139] S51. Based on the IoT platform, a 5km radius geo-fence is dynamically constructed with the target charging pile as the center. The geo-fence then screens and matches the available mobile energy storage vehicles or backup charging pile operation and maintenance resources within the fence in real time.

[0140] S52. Calculate the service interruption impact index for the selected operation and maintenance resources based on the current service demand, and optimize the resource allocation priority based on the service interruption impact index;

[0141] S53. Based on the scheduling results, the augmented reality maintenance instructions are pushed to the on-site operation and maintenance personnel terminal, and the faulty components are visualized through the digital twin model.

[0142] S54. Automatically summarize the full-process analysis and execution data, generate a visual maintenance report containing the optimal maintenance time window and fault root cause analysis, and push it to the operation and maintenance terminal and user APP in real time.

[0143] In this embodiment, step S6 specifically includes:

[0144] S61. Collect and analyze charging pile operation data after executing the self-healing strategy, and use incremental learning to update the weights of the dynamic pruning random forest model;

[0145] S62. Build an expert network that integrates multi-class data features and continuously optimizes health status assessment and diagnosis models by learning different types of operating samples;

[0146] S63. Generate a hierarchical maintenance report to display the health status of the charging pile to the user end, and provide maintenance information including spare parts replacement priority to the operation and maintenance end.

[0147] In this embodiment, step S7 specifically includes:

[0148] S71. Use structured data encapsulation to manage operation and maintenance process metadata in a JSON-based linked data format, and establish time series indexes and geo-hash-coded spatial indexes for all event data.

[0149] S72. Set a directed acyclic graph scheduling rule to support parallel execution of data collection and health assessment tasks;

[0150] S73. Implement a tiered data storage strategy, storing hot data in the Redis database, warm data for normal operation and maintenance queries in the Influx time series database, and moving cold data to MinIO object storage.

[0151] S74. Regularly generate Merkle tree hashes for logs, indicators, and decision data.

[0152] Example 1:

[0153] In order to verify the feasibility of the present invention in implementation, the present invention was applied to the remote intelligent operation and maintenance scenario of a group of electric vehicle charging stations in a large urban public transportation hub. The charging infrastructure in this area is dense, with an average of more than 800 electric vehicles served daily. The charging piles are of various models and the operating environment is changeable. They are easily affected by high loads and multi-source interference, which poses a considerable challenge to the traditional operation and maintenance model. Previously, the site mainly relied on manual inspections and traditional remote alarm systems. When faced with problems such as equipment aging, sudden failures, and frequent alarms from some charging piles, there were widespread problems such as response delays, high incidence of false alarms, low efficiency in the use of manual resources, and unreasonable allocation of maintenance response resources, which significantly affected the overall energy utilization efficiency and user charging experience.

[0154] The operator selected 60 charging piles to deploy the remote intelligent operation and maintenance system based on the Internet of Things of the present invention, including 18 fast charging piles and 42 slow charging piles. At the same time, another 60 charging piles were selected for comparative testing according to the original traditional operation and maintenance scheme. The new system collects various operating data of the charging piles in real time through multi-modal sensors such as multi-source electrical parameters, infrared thermal imaging, mechanical vibration, and ambient temperature and humidity, and uses hardware topology awareness protocol in conjunction with the edge gateway to accurately align and clean the data stream. After the data is uploaded to the cloud, the system intelligently classifies the physical state of the charging pile and dynamically evaluates its health based on the dynamic pruning random forest and spatiotemporal layered dual-Q network algorithm. When the health score is lower than the threshold or an obvious anomaly is detected, the system automatically starts the digital twin simulation sandbox, performs a full-process virtual test of the self-healing strategy, and introduces an expert rule fuse mechanism to prevent the occurrence of abnormal policy conflicts in real time. For self-healing operation instructions that have passed the simulation evaluation and have no abnormalities after the fuse review, such as dynamic power allocation, forced air cooling, and interface locking, the system will automatically issue them for execution. The entire process achieves closed-loop management from data collection, intelligent analysis, strategy generation, simulation verification, automatic execution to effect backtracking.

[0155] In actual application, during a one-month pilot phase, the system of the present invention achieved significant improvements in charging pile fault response speed, health recovery, fault warning accuracy, operation and maintenance resource allocation efficiency, and user satisfaction. Detailed data are shown in Table 1.

[0156] Table 1 Performance comparison data of the system of the present invention and the traditional system in the operation and maintenance scenario of a large charging station group

[0157]

[0158]

[0159] According to the data in Table 1, it can be seen that the system of the present invention is significantly better than the traditional system in all key performance indicators. First, in terms of equipment fault response efficiency, after adopting the system of the present invention, the average single fault response time of the charging pile is greatly shortened to only 12 minutes, compared with 65 minutes of the traditional system, which is about 83% faster, greatly reducing the impact of equipment failures on user services. Secondly, relying on multimodal data fusion and intelligent analysis algorithms, the false alarm rate and missed alarm rate of faults are reduced to 2.5% and 1.3% respectively, which are much lower than the 16.6% and 7.9% of the traditional system, reflecting that the system of the present invention has higher accuracy in anomaly detection and risk identification, can effectively reduce unnecessary maintenance operations, and significantly reduce the risk of missed faults.

[0160] In addition, the compliance rate of the self-healing strategy generated by the system of the present invention has increased to 91%, which is significantly higher than the 70% of the control group; after the self-healing operation, the average improvement in the health of the pile body within 30 minutes is 26.7%, which is about 2.3 times that of the traditional system, which fully demonstrates that the present invention can restore and optimize the operating status of the equipment more quickly and efficiently. In terms of operation and maintenance resource allocation, the frequency of operation and maintenance personnel participation has dropped from 218 to 129, and the monthly operation and maintenance cost of a single device has dropped from 115 yuan to 77 yuan, a decrease of 40.8% and 33% respectively, which greatly saves labor and financial investment. The repair rate of critical faults has increased from 61% to 92%, indicating that the new system can effectively ensure the continuous and stable operation of charging piles.

[0161] Thanks to the expert-rule circuit-breaker mechanism, the system experienced no secondary equipment failures or major safety incidents during the trial, while the traditional system experienced two secondary failures and one safety incident. During the pilot period, the system executed 11 policy circuit breakers, effectively preventing potential risks. User satisfaction ratings increased from 85% to 96%, demonstrating the system's significant improvement in charging experience and service quality.

[0162] In summary, the system of the present invention has achieved remarkable results in operation and maintenance efficiency, response speed, fault prevention and operational safety, while significantly reducing operating costs and labor input, providing a solid guarantee for the large-scale, intelligent and high-reliability operation and maintenance of charging piles, and showing broad practical application prospects and promotion value.

[0163] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A remote operation and maintenance system for charging piles based on the Internet of Things, characterized in that: include: The data acquisition and edge computing module is used to collect multimodal operating data of charging piles, align the timestamps of the multimodal operating data, and eliminate invalid data to generate standardized operating data; A dynamic health assessment module, which performs multimodal feature coupling analysis on standardized operating data and generates a health score for charging piles using an improved dynamic pruning random forest model; The fault root cause analysis module is used to initiate the time-space coupled fault root cause analysis process and obtain the root cause confidence score and high-risk components; The self-healing strategy execution module is used to generate self-healing strategies for charging piles. When a strategy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making security. A dynamic resource scheduling module is used to dynamically optimize the allocation of O&M resources based on health scores and geographic location. This includes a geo-fence-driven scheduling algorithm that matches mobile energy storage vehicles or backup charging stations, and calculates the cost and benefit weights of resource scheduling in real time. Intelligent operation and maintenance management module, used to update the weights of the dynamically pruned random forest model and generate hierarchical maintenance reports; The data full-process control module is used to complete automated closed-loop management from data collection to operation and maintenance decision-making.

2. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 1, characterized in that: The modules are implemented as follows: S1. Collect multimodal operation data of charging piles, align the timestamps of multimodal operation data using hardware topology awareness protocol, remove invalid data, and generate standardized operation data; S2. Perform multimodal feature coupling analysis on standardized operating data and generate a health score for the charging pile using an improved dynamic pruning random forest model; S3: When the health score of the charging pile is lower than the set dynamic threshold, the spatiotemporal coupled fault root cause analysis process is automatically started to obtain the root cause confidence score and high-risk components; S4. For high-risk components, a hybrid decision-making framework is established. Under normal operating conditions, a spatiotemporal layered dual-Q network is used to generate self-healing strategies. When the charging pile health score is lower than 30, the system switches to digital twin simulation verification mode. When a policy conflict occurs, an expert rule-based fuse mechanism is activated to ensure decision-making safety. S5. Based on the health score of the charging pile and the geo-fence information, a resource scheduling optimization algorithm is applied to achieve dynamic allocation of mobile energy storage vehicles or backup charging piles; S6. Using incremental learning to update the weights of the dynamic pruning random forest model, and generating a hierarchical maintenance report through an expert network; S7. Manage operation and maintenance data through structuring and indexing, and implement a tiered data storage strategy.

3. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S1 specifically includes: S11. Collect electrical operating parameters of the charging pile using a current / voltage harmonic sensor. The electrical operating parameters include current, voltage, current harmonic distortion rate, voltage harmonic distortion rate, and instantaneous power fluctuation variance. Perform time domain analysis on the obtained current data to obtain the number of current mutations within a given sampling period. Use a narrowband Internet of Things communication module to upload the electrical operating parameters to a cloud platform in real time. S12. Regularly collect temperature distribution data of the power module using an infrared thermal imaging module. Use a hardware topology-based thermal field alignment algorithm to map the temperature distribution data to the corresponding locations on the charging pile's physical structure. Upload the complete temperature field distribution every 30 seconds via the 5G network. Further, calculate the temperature gradient based on the temperature differences at each monitoring point on the physical structure. S13. Collect mechanical vibration signals using a vibration accelerometer, extract mechanical wear energy characteristics using a 0.5-2 kHz bandpass filter, generate a vibration energy index, and aggregate and transmit the vibration energy indexes of all sampling points through a long-distance wide-area gateway; S14, collecting temperature and humidity data through environmental sensors, and synchronizing electrical operating parameters with temperature and humidity data using the IEEE1588 precision time protocol; S15. Use a hardware topology awareness protocol to align the timestamps of the multimodal operation data, and eliminate invalid data based on a preset threshold to form standardized operation data.

4. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S2 specifically includes: S21. Calculate the dynamic correlation coefficient between the current harmonic distortion rate and the temperature gradient using the Pearson correlation coefficient; S22, fusing the vibration energy index and the number of current mutations by weighted linear superposition to generate a mechanical wear index; S23. Inputting the dynamic correlation coefficient, mechanical wear index, and standardized operating data into an improved dynamic pruning random forest model to obtain a health score for the charging pile. The improved dynamic pruning random forest model introduces a three-level pruning mechanism to automatically adjust the maximum depth of each decision tree in the model based on the CPU utilization of the edge node. The three-level pruning mechanism is as follows: when the CPU utilization rate is higher than 80%, the emergency mode is activated, the maximum depth of the decision tree is compressed to 5 layers, and the feature selector is enabled to retain only the top-10 features; when the CPU utilization rate is between 50% and 80%, the balanced mode is entered, the depth is limited to 10 layers, and the dynamic top-20 features are used; when the CPU utilization rate is lower than 50%, the full mode is run, the full depth of 15 layers is maintained, and all 42-dimensional features are activated; S24. Apply time decay correction to the health score of the charging pile, and increase the score weight of abnormal events that occur in the most recent time window by 20%.

5. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2 is characterized in that: The step S3 specifically includes: S31. Based on the standardized operation data, calculate the spatial similarity between the electrical operation parameter characteristics of the charging pile and the electrical operation parameter characteristics in the historical operation and maintenance cases. At the same time, calculate the temporal similarity between the current temperature gradient data and the historical temperature gradient data, and obtain the multimodal similarity score S by weighting. sim : Among them, V elec and V hist are the current and historical electrical characteristic vectors, T grad and T hist is the current and historical temperature gradient distribution, JSD is the Jensen-Shannon divergence; S32. If the multimodal similarity score S sim When the preset threshold is reached, the suspected root cause of the corresponding component and the corresponding score result are directly output based on historical cases; S33. If the comprehensive similarity score is lower than the threshold, a component-level relationship graph of the charging pile is constructed based on the current multi-source operation data. The propagation links of the abnormal signal between the components are automatically analyzed through a three-layer heterogeneous graph convolutional network to infer the potential root cause. S34. Assign a root cause confidence score to each component based on the multimodal similarity score or the inference output of the heterogeneous graph convolutional network, and mark components with a confidence score greater than or equal to 80% as high-risk components.

6. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S4 specifically includes: S41. For high-risk components, a hybrid decision-making framework is established. The hybrid decision-making framework includes: using a spatiotemporal layered dual-Q network to generate a self-healing strategy under normal operating conditions; automatically switching to a digital twin simulation verification mode when the charging pile health score is below 30; and activating an expert rule-based fuse mechanism to ensure decision safety when a policy conflict is detected; S42. Construct a spatiotemporal layered dual-Q network structure, specifically: the spatial layer includes an upper-layer global state network and a lower-layer local feature network, and the temporal layer includes a fast response network and a steady-state optimization network; The upper global state network adopts a multi-layer perceptron structure, inputs the global operation characteristics of the charging station level, and outputs the global self-healing action Q value. The lower local feature network adopts a three-layer convolutional neural network structure, inputs the multimodal local characteristics of the charging pile, and outputs the corresponding local self-healing action Q value. The fast response network processes short-term features in real time through a lightweight feedforward neural network. The steady-state optimization network uses a recurrent neural network and combines historical operation data of charging piles to generate self-healing action recommendations for long-term health optimization and output the optimal self-healing strategy through multi-scale fusion. S43. Design a time-varying compound reward function to dynamically weight and integrate health improvement rewards, maintenance cost penalties, and safety constraints according to time: R t =μ1(t)·R health +μ2(t)·R cost +μ3(t)·R safety ; where R health To predict the improvement of health, R cost is the maintenance cost of the charging pile, R safety is the safety constraint term, μ1(t), μ2(t), μ3(t) are the reward weights that change over time; S44. The improved Huber loss function is used to optimize the weights of the spatiotemporal layered dual-Q network. The loss threshold δ(n) decreases linearly with the number of training rounds. The loss function formula is: Where Q represents the state-action value output by the Q learning model, y represents the target value calculated based on the reward and the next state, and δ(n) decreases with each training round; S45. Based on the trained and optimized spatiotemporal layered dual-Q network, generate the optimal self-healing strategy a under each charging pile operating state s. * , that is, output the self-healing action that maximizes the Q value. The formula is: Where s represents the current state, a represents an optional action, and Q represents the cumulative reward expected to be obtained by selecting action a in the current state s; S46. When the health score of the charging pile is lower than 30, it automatically switches to the digital twin simulation verification mode, pre-simulates the generated self-healing strategy through the digital twin sandbox, and adopts the ∈-greedy decision method to select the self-healing action, that is, the currently known optimal self-healing action is selected with probability 1-∈, and any executable self-healing action is randomly selected with probability ∈. The formula is: Among them,∈ represents the exploration probability; S47. After passing the digital twin sandbox verification, the system further detects whether the self-healing strategy conflicts with existing operation and maintenance actions. If a policy conflict is detected, the expert rule-based circuit breaker mechanism is activated to suspend the issuance and execution of the self-healing strategy. When the self-healing strategy is verified by the expert rules to have no conflict, it is automatically converted into a specific operation and maintenance instruction and issued to the device for execution; S48. During the implementation of the self-healing strategy, the health score and operating data of the charging pile are continuously monitored in real time. If the health score does not increase by 20% within 30 minutes, the self-healing operation will be automatically withdrawn immediately and restored to the state before the strategy execution.

7. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S5 specifically includes: S51. Based on the IoT platform, a 5km radius geo-fence is dynamically constructed with the target charging pile as the center. The geo-fence then screens and matches the available mobile energy storage vehicles or backup charging pile operation and maintenance resources within the fence in real time. S52. Calculate the service interruption impact index for the selected operation and maintenance resources based on the current service demand, and optimize the resource allocation priority based on the service interruption impact index; S53. Based on the scheduling results, the augmented reality maintenance instructions are pushed to the on-site operation and maintenance personnel terminal, and the faulty components are visualized through the digital twin model. S54. Automatically summarize the full-process analysis and execution data, generate a visual maintenance report containing the optimal maintenance time window and fault root cause analysis, and push it to the operation and maintenance terminal and user APP in real time.

8. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S6 specifically includes: S61. Collect and analyze charging pile operation data after executing the self-healing strategy, and use incremental learning to update the weights of the dynamic pruning random forest model; S62. Build an expert network that integrates multi-class data features and continuously optimizes health status assessment and diagnosis models by learning different types of operating samples; S63. Generate a hierarchical maintenance report to display the health status of the charging pile to the user end, and provide maintenance information including spare parts replacement priority to the operation and maintenance end.

9. The remote operation and maintenance system for charging piles based on the Internet of Things according to claim 2, characterized in that: The step S7 specifically includes: S71. Use structured data encapsulation to manage operation and maintenance process metadata in a JSON-based linked data format, and establish time series indexes and geo-hash-coded spatial indexes for all event data. S72. Set a directed acyclic graph scheduling rule to support parallel execution of data collection and health assessment tasks; S73. Implement a tiered data storage strategy, storing hot data in the Redis database, warm data for normal operation and maintenance queries in the Influx time series database, and moving cold data to MinIO object storage. S74. Regularly generate Merkle tree hashes for logs, indicators, and decision data.

Images