A charging pile fault cloud edge collaborative diagnosis and early warning system and method

Abstract

The application discloses a charging pile fault cloud edge collaborative diagnosis and early warning system and method, and relates to the technical field of charging pile fault diagnosis and operation and maintenance. The edge side node comprises a data acquisition unit, an edge diagnosis unit and a task scheduling unit, the cloud platform comprises a cloud control unit, a global diagnosis unit, a result fusion unit and a diagnosis output unit; the data acquisition unit acquires multi-modal data to obtain edge feature data; the edge diagnosis unit instantaneously diagnoses and outputs a first diagnosis result, a fault emergency degree and a first confidence degree based on a lightweight diagnosis model, and triggers a local protection action in an emergency; the task scheduling unit decides a local closed loop or uploads data according to relevant data; the cloud control unit transmits data to the global diagnosis unit for secondary diagnosis to obtain a second diagnosis result and a confidence degree, and obtains a common cause fault and a fault area according to multi-pile feature correlation analysis; the result fusion unit fuses according to the confidence degree to obtain a single pile diagnosis result, and the diagnosis output unit determines a final diagnosis result accordingly.

Description

Technical Field

[0001] This invention relates to the field of charging pile fault diagnosis and maintenance technology, and in particular to a cloud-edge collaborative diagnosis and early warning system and method for charging pile faults. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the number of charging piles continues to grow. These piles operate under complex environments such as high loads and wide temperature ranges, making key components prone to electrical aging and mechanical wear. The causes of failure are complex, ranging from minor interruptions to electrical fires. Therefore, rapid fault diagnosis and early warning are crucial.

[0003] The current stage of charging pile fault diagnosis technology has generally gone through three stages of development, and the solutions and existing problems of each stage are as follows.

[0004] The first stage is local monitoring and diagnosis of a single pile based on fixed thresholds: data such as current, voltage, and temperature are compared with preset thresholds to determine the type of abnormality and fault. Its limitation is that the fixed thresholds cannot be adaptively adjusted according to equipment aging and environmental conditions, false alarms and missed alarms coexist, and it does not have the global analysis capability across piles and stations.

[0005] The second stage is intelligent diagnosis of a single pile based on multimodal data and deep learning: multimodal data such as electrical, temperature, and sound are input into a deep network model to output the failure probability. Its limitations are that the model is deployed at the pile end, and it needs to be lightweight due to computing power constraints, and its feature learning ability is limited; moreover, the training only relies on local samples of a single pile, and the model becomes inaccurate over time due to aging.

[0006] The third phase involves remote diagnostics via edge computing gateways: the edge gateway performs preprocessing and feature extraction, makes local decisions for emergency faults, and uploads feature data to the cloud for collaborative analysis by a rule engine and machine learning model. This initial solution establishes a division of labor between "edge preprocessing + cloud diagnostics," but the cloud-edge collaboration is unidirectional and static: First, data flows unidirectionally to the cloud, and cloud capacity growth cannot flow back to the edge; the edge model becomes fixed once deployed. Second, task assignments are pre-defined and do not dynamically adjust with network quality, computing power, and confidence levels, leading to link breakage during network outages. Third, multiple anomalies caused by power grid disturbances are still diagnosed independently, easily generating a large number of invalid maintenance work orders. Fourth, it primarily focuses on post-incident handling and is not linked to lifespan prediction and impact assessment.

[0007] In summary, existing technologies still have shortcomings in balancing local response and global in-depth diagnosis, bidirectional evolution of cloud-edge models, continuity of network outage links, and identification of common-cause faults across charging piles. There is an urgent need in this field for a cloud-edge collaborative diagnosis and early warning scheme for electric vehicle charging piles, which can complete emergency fault diagnosis and protection in a timely manner at the edge, and dynamically schedule tasks according to the degree of urgency, confidence level, computing power and network quality, so as to realize bidirectional evolution of cloud-edge models, autonomous reconciliation during network outages, and identification of common-cause faults across charging piles. Summary of the Invention

[0008] In view of this, the present invention provides a cloud-edge collaborative diagnosis and early warning system and method for charging pile faults. Through a two-way closed-loop collaborative architecture between the edge node network and the cloud platform, it realizes dynamic scheduling of diagnostic tasks, continuous two-way evolution of diagnostic models between the cloud and the edge, autonomous network disconnection and recovery reconciliation of diagnostic links, and fusion output of single-pile diagnostic results and cross-pile common-cause faults and fault areas, comprehensively improving the real-time performance, accuracy and operational foresight of charging pile fault diagnosis.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A cloud-edge collaborative diagnosis and early warning system for charging pile faults includes: an edge-side node network and a cloud platform. The edge-side node network includes multiple edge-side nodes, each deployed in different charging stations. Each edge-side node includes a data acquisition unit, an edge diagnosis unit, and a task scheduling unit connected in sequence. The cloud platform includes a cloud control unit, a global diagnosis unit, a result fusion unit, and a diagnosis output unit connected in sequence. The cloud control unit is connected to the task scheduling unit of each edge-side node and also communicates with each edge-side node. The data acquisition unit is used to collect multimodal operation data of each charging pile in the corresponding charging station, and obtains edge feature data after preprocessing and feature extraction. The edge diagnostic unit is used to perform local real-time diagnosis of each charging pile based on a lightweight fault diagnosis model and edge feature data to obtain the first diagnostic result, the fault urgency level and the first confidence level, and to trigger local protection action when the first diagnostic result meets the emergency protection conditions. The task scheduling unit is used to obtain the computing power margin and communication quality of the edge computing node. Based on the computing power margin, communication quality, first confidence level and fault urgency, it makes dynamic decisions in combination with preset diagnostic conditions. If the preset diagnostic conditions are not met, local closed-loop diagnosis is performed; if they are met, edge feature data is uploaded. The cloud control unit is used to control the start and stop actions of each edge node and transmit edge feature data to the global diagnostic unit; The global diagnostic unit is used to perform secondary diagnosis based on the global fault diagnosis model and edge feature data, and output the second diagnostic result and its second confidence level; it is also used to perform correlation analysis based on the edge feature data of each charging pile when a fault occurs in the same or multiple charging piles, and obtain information on common cause faults and fault areas. The result fusion unit is used to fuse the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level to obtain the single-pile diagnostic result; wherein, when a local closed-loop diagnosis is performed but no second diagnostic result is generated, the first diagnostic result is used as the single-pile diagnostic result; The diagnostic output unit is used to determine the final diagnostic result based on the single-pile diagnostic results, common-cause faults, and fault area information.

[0010] The aforementioned system may optionally include multimodal operating data such as electrical data, temperature data, acoustic vibration data, insulation monitoring data, and the operating logs and fault codes of the charging pile controller.

[0011] The above system, optionally, uses an artificial intelligence model compression technique based on a teacher-student model, which takes the global fault diagnosis model as the teacher model, performs knowledge distillation and model compression on it to obtain a lightweight fault diagnosis model as the student model; the global fault diagnosis model includes a parallel rule matching module and a neural network learning module. The rule matching module is used to match known fault types to edge feature data based on a preset fault branch tree and expert rules, and output a fault determination path. The neural network learning module is used to train a deep neural network model using historical charging pile fault samples. Edge feature data is input into the trained deep neural network model to obtain charging pile fault identification results.

[0012] The above-mentioned system may optionally include the following preset diagnostic conditions: the first confidence level is lower than the preset confidence threshold and the communication quality is not lower than the preset quality threshold; the first confidence level is within the preset fuzzy range; the fault urgency is lower than the preset highest urgency level and the first confidence level is lower than the preset confidence threshold; the computing power margin is lower than the preset computing power threshold and the communication quality is not lower than the preset quality threshold. The task scheduling unit is used to upload edge feature data to the global diagnostic unit when any of the preset diagnostic conditions are met.

[0013] The aforementioned system, optionally, allows the global diagnostic unit to perform correlation analysis based on the edge feature data of each charging pile to obtain information on common cause faults and fault areas, specifically including: The edge feature data of each charging pile is extracted according to the preset time window, the correlation coefficient and feature similarity between each pair of charging piles are calculated, and the three elements of the charging pile fault are determined. The three elements of the fault include the charging pile identity information and the time and location of the fault. When the correlation coefficient exceeds the preset limit and the number of charging piles involved is no less than two, a common cause fault is determined to exist. Based on the similarity of characteristics of each charging pile and the three elements of fault, the faults within the pile, common faults on the site side, and disturbances on the power grid side are distinguished, and the fault area information is determined accordingly. The fault area information includes the charging site, transformer area, or power supply branch where the faulty charging pile is located.

[0014] In the aforementioned system, optionally, the result fusion unit performs confidence fusion on the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level, specifically including: The first and second diagnostic results are weighted and fused using the first and second confidence levels as weights, or the basic probability distribution of the first and second diagnostic results is synthesized using DS evidence theory, and the single-case diagnostic result and the overall confidence level are output.

[0015] Optionally, the cloud platform of the above system also includes a model evolution unit, which is bidirectionally connected to the global diagnostic unit and connected to the result fusion unit and the edge diagnostic unit respectively. The model evolution unit is used to identify single-pile diagnostic results with a comprehensive confidence score less than a preset comprehensive confidence threshold as difficult examples, incrementally train the global fault diagnosis model, and generate a differentiated lightweight model update package adapted to each edge node by knowledge distillation and quantization compression of the incrementally trained global fault diagnosis model, and then distribute it to the corresponding edge node.

[0016] Optionally, any edge node in the above system may also include an edge caching unit, which is connected to the task scheduling unit and the global diagnostic unit respectively. This unit is used to cache edge feature data when the network is interrupted or the communication quality is not lower than a preset quality threshold, and to resume transmission after the network is restored.

[0017] Optionally, the cloud platform of the above system also includes a lifespan prediction and early warning unit, which is connected to the cloud historical data storage and historical operation and maintenance record library and the diagnostic output unit, respectively. The lifespan prediction and early warning unit is used to obtain the degradation trend characteristics and historical operation and maintenance records of each charging pile from the cloud historical data storage and historical operation and maintenance record library, and predict the remaining lifespan of the charging pile using the degradation model according to the preset cycle.

[0018] A method for cloud-edge collaborative diagnosis and early warning of charging pile faults, applied to any of the above-mentioned cloud-edge collaborative diagnosis and early warning systems for charging pile faults, specifically includes the following steps: S1 Data Acquisition and Processing Steps: Multimodal operation data of each charging pile in each charging station is collected by edge side nodes deployed in each charging station, and edge feature data is obtained after preprocessing and feature extraction. S2 Diagnostic Protection Steps: The edge-side node uses a lightweight fault diagnosis model to perform local real-time diagnosis of each charging pile using edge feature data, obtains the first diagnosis result, the fault urgency level and the first confidence level, and triggers local protection action when the first diagnosis result meets the emergency protection conditions. S3 Dynamic Decision-Making Steps: Edge nodes acquire current computing power reserves and communication quality, and make dynamic decisions based on computing power reserves, communication quality, first confidence level, and fault urgency, combined with preset diagnostic conditions. S4 condition judgment steps: When the preset diagnostic conditions are not met, the edge-side node performs local closed-loop diagnosis; when the preset diagnostic conditions are met, the edge-side node uploads the edge feature data to the cloud platform. S5 Secondary Diagnosis Steps: The cloud platform performs secondary diagnosis based on the global fault diagnosis model and the received edge feature data, outputting the second diagnosis result and its second confidence level; and when a charging pile in the same or multiple charging stations fails, it performs correlation analysis based on the edge feature data of each charging pile to obtain common cause fault and fault area information. S6 Confidence Fusion Steps: The cloud platform fuses the first and second diagnostic results based on the first and second confidence levels to obtain the single-pile diagnostic result; where no second diagnostic result is generated during local closed-loop diagnosis, the first diagnostic result is used as the single-pile diagnostic result. S7 Result Output Steps: The cloud platform determines and outputs the final diagnosis result based on the single-pile diagnosis result, common-cause faults, and fault area information.

[0019] As can be seen from the above technical solution, compared with the prior art, the present invention provides a cloud-edge collaborative diagnosis and early warning system and method for charging pile faults, which has the following beneficial effects: (1) Real-time response: Emergency faults are diagnosed locally and triggered for protection actions by the edge lightweight model in real time, without relying on network backhaul, and meet the millisecond-level security response. (2) Diagnostic accuracy: Difficult faults are dynamically offloaded to the cloud for secondary diagnosis based on confidence level, urgency, computing power and network quality. Cloud and edge conclusions are fused based on confidence level, taking into account both timeliness and accuracy. (3) Link continuity: When the network is weak or disconnected, the edge autonomy is automatically switched and data is cached. The subsequent transmission and reconciliation are restored, the diagnosis is uninterrupted and the conclusion is traceable; (4) Common causes can be identified: Data correlation analysis of multiple piles at the same time can distinguish between pile faults, common causes of the station and power grid disturbances and locate the fault area, thereby reducing misjudgments and invalid work orders; (5) Model self-evolution: The model continuously absorbs knowledge from the entire network without becoming outdated. The model is trained incrementally by aggregating difficult example samples in the cloud and then distilled and distributed. Edge grayscale updates and anomaly rollbacks are performed. (6) Predictive maintenance: Combining remaining lifespan and risk level to output graded early warning and impact assessment, promoting maintenance from post-event repair to pre-event intervention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the structure of a cloud-edge collaborative diagnosis and early warning system for charging pile faults disclosed in this invention; Figure 2 This is a schematic diagram showing the connection between the edge node n and the main and bypass paths of the cloud platform, as disclosed in an embodiment of the present invention. Detailed Implementation

[0022] 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.

[0023] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0024] See Figure 1As shown, this invention discloses a cloud-edge collaborative diagnosis and early warning system for charging pile faults, comprising: an edge-side node network and a cloud platform. The edge-side node network includes multiple edge-side nodes, each deployed in different charging stations. Each edge-side node includes a data acquisition unit, an edge diagnosis unit, and a task scheduling unit connected in sequence. The cloud platform includes a cloud control unit, a global diagnosis unit, a result fusion unit, and a diagnosis output unit connected in sequence. The cloud control unit is connected to the task scheduling unit of each edge-side node and also communicates with each edge-side node. The data acquisition unit is used to collect multimodal operation data of each charging pile in the corresponding charging station, and obtains edge feature data after preprocessing and feature extraction. The edge diagnostic unit is used to perform local real-time diagnosis of each charging pile based on a lightweight fault diagnosis model and edge feature data to obtain the first diagnostic result, the fault urgency level and the first confidence level, and to trigger local protection action when the first diagnostic result meets the emergency protection conditions. The task scheduling unit is used to obtain the computing power margin and communication quality of the edge computing node. Based on the computing power margin, communication quality, first confidence level and fault urgency, it makes dynamic decisions in combination with preset diagnostic conditions. If the preset diagnostic conditions are not met, local closed-loop diagnosis is performed; if they are met, edge feature data is uploaded. The cloud control unit is used to control the start and stop actions of each edge node and transmit edge feature data to the global diagnostic unit; The global diagnostic unit is used to perform secondary diagnosis based on the global fault diagnosis model and edge feature data, and output the second diagnostic result and its second confidence level; it is also used to perform correlation analysis based on the edge feature data of each charging pile when a fault occurs in the same or multiple charging piles, and obtain information on common cause faults and fault areas. The result fusion unit is used to fuse the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level to obtain the single-pile diagnostic result; wherein, when a local closed-loop diagnosis is performed but no second diagnostic result is generated, the first diagnostic result is used as the single-pile diagnostic result; The diagnostic output unit is used to determine the final diagnostic result based on the single-pile diagnostic results, common-cause faults, and fault area information.

[0025] Combination Figure 1As shown in the embodiments disclosed in this invention, the charging pile fault cloud-edge collaborative diagnosis and early warning system mainly consists of an edge-side node network and a cloud platform. The edge-side node network includes N edge-side nodes, where N is a natural number not less than 2. Each edge-side node is deployed in a charging station. Each edge-side node in the network communicates with the controllers of all charging piles in the corresponding station via CAN bus, RS485, or industrial Ethernet, and establishes a communication link with the cloud control unit in the cloud platform via 4G / 5G, fiber optic, or power wireless private network. Figure 1 As shown, the dashed lines connecting edge node 1, edge node 2, ..., edge node N with the cloud control unit represent communication connections.

[0026] Each edge node is an edge computing gateway or edge server deployed within the charging station cabinet, configured with a multi-core ARM or x86 processor, at least 4GB of memory, and an optional NPU acceleration module. After startup, the node sequentially deploys three functional modules: a data acquisition unit, an edge diagnostic unit, and a task scheduling unit, transmitting data through a local transmission channel. The cloud platform is deployed on a public or private cloud, including a globally connected diagnostic unit, a result fusion unit, and a diagnostic output unit, and can be further configured with a model evolution unit and a lifetime prediction and early warning unit. The task scheduling units of each edge node can directly connect to the globally diagnostic unit via a message bus, meaning that the task scheduling units of the edge nodes have a channel to directly unload diagnostic tasks to the cloud, bypassing intermediate links, thus ensuring the shortest unloading path and lowest latency. The following is a detailed description of the charging pile fault cloud-edge collaborative diagnosis and early warning system disclosed in this invention.

[0027] Optionally, multimodal operating data includes electrical data, temperature data, acoustic vibration data, insulation monitoring data, and the operating logs and fault codes of the charging pile controller.

[0028] More specifically, the multimodal operating data collected by the data acquisition unit includes electrical data such as charging voltage and charging current; temperature data such as the temperature of at least one critical heat-generating component in the power module, transformer, and charging interface; and acoustic vibration data such as the operating sound signals of the relays and contactors, as well as the vibration signals of the power module. Furthermore, in the data acquisition unit's preprocessing and feature extraction operations on multimodal operating data, the preprocessing specifically includes filtering and denoising, timestamp alignment, and missing value imputation; the feature extraction specifically includes the extraction of time-domain features, frequency-domain features, and operating mode features.

[0029] In a specific implementation, as an optional embodiment, the data acquisition unit collects multimodal operating data of each charging pile at differentiated sampling frequencies: the sampling frequency of electrical data is not less than 1kHz, including charging voltage, charging current, leakage current, and power factor; the sampling frequency of temperature data is not less than 1Hz, and the collection points cover three key heat-generating parts: the heat sink of the power module, the transformer winding, and the charging interface terminals; the sampling frequency of acoustic vibration data is 8kHz to 16kHz, collecting the operating sound signals of the relays and contactors, as well as the vibration signals of the power module; the insulation monitoring data includes the insulation resistance monitoring value and the trend of change within the insulation monitoring period; and the operation log and fault codes are read through the controller interface.

[0030] Optionally, an AI model compression technique based on a teacher-student model is used, where the global fault diagnosis model is used as the teacher model, and after knowledge distillation and model compression, a lightweight fault diagnosis model is obtained as the student model; the global fault diagnosis model includes a parallel rule matching module and a neural network learning module. The rule matching module is used to match known fault types to edge feature data based on a preset fault branch tree and expert rules, and output a fault determination path. The neural network learning module is used to train a deep neural network model using historical charging pile fault samples. Edge feature data is input into the trained deep neural network model to obtain charging pile fault identification results.

[0031] The lightweight fault diagnosis model built into the edge diagnostic unit is obtained by knowledge distillation and INT8 quantization compression of the global fault diagnosis model in the cloud platform. The global fault diagnosis model is used as the teacher model, and the simplified student network is used as the model to be trained. The distillation temperature parameter is set to T=4. The total loss is the weighted sum of soft label distillation loss and hard label cross-entropy loss, with weights of 0.7 and 0.3 respectively. After training, quantization perception training and INT8 quantization are performed. The model size is compressed to less than one-tenth of the original global model. The inference time of a single diagnostic cycle does not exceed 50ms, and it can run in real time on the embedded processor of the edge node.

[0032] The edge diagnostic unit inputs edge feature data into the lightweight fault diagnosis model and outputs the first diagnostic result, the fault urgency level, and the first confidence level. The fault urgency level is divided into three levels according to the response time limit requirements: normal level, urgent level, and highest urgency level. The highest urgency level corresponds to fault types that directly endanger personal and equipment safety, such as overcurrent, short circuit precursors, insulation failure, and severe overtemperature, which are also the emergency protection conditions defined in this embodiment.

[0033] Emergency protection conditions are pre-set in the edge diagnostic unit, such as leakage current exceeding 30mA, power module temperature exceeding 105℃, and DC bus voltage exceeding 115% of the rated value. When the first diagnostic result meets the emergency protection conditions, the edge diagnostic unit directly triggers local protection actions without the task scheduling unit's decision-making or waiting for cloud response. Local protection actions include at least one of the following: cutting off the charging circuit, reducing power operation, stopping charging, and locking the charging head. The entire process is completed locally at the edge, with end-to-end latency controlled within 100ms.

[0034] Optionally, the preset diagnostic conditions include: the first confidence level is lower than the preset confidence threshold and the communication quality is not lower than the preset quality threshold; the first confidence level is in the preset fuzzy range; the fault urgency is lower than the preset highest urgency level and the first confidence level is lower than the preset confidence threshold; the computing power margin is lower than the preset computing power threshold and the communication quality is not lower than the preset quality threshold. The task scheduling unit is used to upload edge feature data to the global diagnostic unit when any of the preset diagnostic conditions are met.

[0035] The task scheduling unit performs dynamic decision-making after the edge diagnostic unit completes local real-time diagnosis and outputs the first confidence level. The task scheduling unit calculates the current computing power reserve of the edge node by reading processor utilization, memory usage, and NPU usage. It obtains the current communication quality by periodically probing the uplink, with relevant parameters including round-trip latency, packet loss rate, and available uplink bandwidth. When any of the preset diagnostic conditions is met, the task scheduling unit uploads the edge feature data to the global diagnostic unit.

[0036] It should be noted that the four factors of computing power margin, communication quality, first confidence level, and fault urgency are the basis for the task scheduling unit to make a decision on local closed-loop or cloud offloading, rather than the data content offloaded to the cloud platform. The data used for secondary diagnosis by the cloud platform is edge feature data.

[0037] Optionally, the global diagnostic unit performs correlation analysis based on the edge feature data of each charging pile to obtain common cause fault and fault area information, specifically including: The edge feature data of each charging pile is extracted according to the preset time window, the correlation coefficient and feature similarity between each pair of charging piles are calculated, and the three elements of the charging pile fault are determined. The three elements of the fault include the charging pile identity information and the time and location of the fault. When the correlation coefficient exceeds the preset limit and the number of charging piles involved is no less than two, a common cause fault is determined to exist. Based on the similarity of characteristics of each charging pile and the three elements of fault, the faults within the pile, common faults on the site side, and disturbances on the power grid side are distinguished, and the fault area information is determined accordingly. The fault area information includes the charging site, transformer area, or power supply branch where the faulty charging pile is located.

[0038] Specifically, the time window is set to 5 seconds to 5 minutes, preferably 30 seconds. The calculation involves extracting changes within the same time window, specifically capturing edge feature data of charging piles that malfunctioned within the same time period, extracting changes in electrical characteristics, and calculating the correlation coefficient and feature similarity between any two malfunctioning charging piles. The correlation coefficient is either the Pearson correlation coefficient or the dynamic time-warped distance. After completing the calculation for all charging piles, if the correlation coefficient exceeds a preset limit and involves at least two charging piles, a common cause fault is identified. In this embodiment, the preset limit is 0.8. Finally, based on the identity information of each affected charging pile and the time and location of the fault, the specific judgment criteria are as follows: (1) If only a single charging pile is abnormal and the correlation coefficient is lower than the preset coefficient limit, it is determined to be a fault in the pile, and the fault area is located to the charging pile body and its replaceable parts. (2) If the correlation coefficient of multiple charging piles (no less than two) in the same station exceeds the preset coefficient limit at the same time, and no large-scale failure occurs in other charging pile stations, it is determined to be a common cause failure on the station side, and the fault area is located to the incoming switch, busbar or shared power distribution equipment of the station. (3) If multiple adjacent charging piles in the same time window simultaneously exhibit abnormalities, the correlation coefficient exceeds the preset coefficient limit, and the edge nodes of the fault exhibit connected or dense regional characteristics, it is determined to be a grid-side disturbance. The fault area is located to the corresponding transformer area or power supply branch, and can be further confirmed by combining the grid-side voltage sag record.

[0039] Optionally, the result fusion unit performs confidence fusion on the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level, specifically including: The first and second diagnostic results are weighted and fused using the first and second confidence levels as weights, or the basic probability distribution of the first and second diagnostic results is synthesized using DS evidence theory, and the single-case diagnostic result and the overall confidence level are output.

[0040] Optional, see Figure 2 As shown, the cloud platform also includes a model evolution unit and a lifetime prediction and early warning unit; the model evolution unit is bidirectionally connected to the global diagnosis unit, and is also connected to the result fusion unit and the edge diagnosis unit respectively; the lifetime prediction and early warning unit is connected to the cloud historical data storage and historical operation and maintenance record library and the diagnosis output unit respectively.

[0041] The model evolution unit is used to identify single-pile diagnostic results with a comprehensive confidence score less than a preset comprehensive confidence threshold as difficult examples, incrementally train the global fault diagnosis model, and generate a differentiated lightweight model update package adapted to each edge node by knowledge distillation and quantization compression of the incrementally trained global fault diagnosis model, and then distribute it to the corresponding edge node.

[0042] Furthermore, the edge diagnostic unit adopts a gray-scale update strategy: first, the update package is loaded on no more than 5% of the edge-side nodes, and the effect is verified using locally retained historical replay samples. When the accuracy of at least 100 consecutive diagnoses is not lower than the baseline before the update and there is no increase in the false alarm rate, the verification is deemed successful and a full replacement is completed; otherwise, it automatically rolls back to the lightweight fault diagnosis model before the update and reports the rollback event to the cloud platform. This forms a two-way continuous evolution closed loop of edge difficult cases going to the cloud, cloud incremental training, distillation distribution, edge gray-scale update, and effect feedback.

[0043] The lifespan prediction and early warning unit is used to obtain the degradation trend characteristics and historical operation and maintenance records of each charging pile from the cloud historical data storage and historical operation and maintenance record library. Based on the preset cycle, it uses the degradation model to predict the remaining lifespan of the charging pile. The degradation model can adopt the degradation trajectory model based on Wiener process or the survival analysis model based on historical failure data, and output the expected value and confidence interval of the remaining lifespan.

[0044] like Figure 2 As shown, each edge node in the edge-side node network also includes an edge cache unit, which is connected to the data acquisition unit and the cloud control unit respectively. It is used to cache edge feature data when the network is interrupted or the communication quality is not lower than a preset quality threshold, and to resume transmission after the network is restored.

[0045] Optional, such as Figure 2 As shown, the cloud control unit and the global diagnostic unit are bidirectionally connected, and are also connected to the operation and maintenance terminal and the model evolution unit respectively.

[0046] Specifically, when the network is interrupted or the transmission quality is lower than the preset quality threshold, the edge node switches to edge autonomous mode: the edge diagnostic unit continues to independently complete local real-time diagnostics and trigger local protection actions when necessary, the task scheduling unit keeps the diagnostic tasks local at the edge, the edge caching unit continuously caches edge feature data and diagnostic events, and the cache adopts a cyclic overwrite strategy and prioritizes the retention of high-urgency events.

[0047] After the network is restored, the global diagnostic unit re-examines the cached diagnostic events. The cloud control unit reconciles the re-examination conclusions obtained by the global diagnostic unit with the local diagnostic conclusions of the edge nodes during the edge autonomy period, item by item: events with consistent conclusions are archived and closed; events with inconsistent conclusions generate review work orders and push them to the operation and maintenance terminal, and are sent back to the model evolution unit as difficult case samples for incremental training.

[0048] The diagnostic output unit determines the fault type, location, and severity based on the single-pile diagnostic results, and combines this with information on common-cause faults and fault areas to determine the fault's impact range and source, thus forming the final diagnostic result. The final diagnostic result structurally includes both single-pile and cross-pile dimension information: single-pile dimension information includes fault type, location, severity, and overall confidence level; cross-pile dimension information includes whether it is a common-cause fault, the type of common-cause fault (station-side common-cause fault or grid-side disturbance), and the fault area.

[0049] Combination Figure 2 The diagnostic output unit further combines the remaining service life of the charging piles output by the life prediction and early warning unit with the final diagnostic results to calculate the risk level. The risk level includes three levels of early warning: alert, alarm, and emergency. The alert level is pushed via message from the operation and maintenance platform with a response time of 72 hours; the alarm level is pushed via SMS and App with a response time of 24 hours; and the emergency level is pushed via telephone and SMS in parallel with a response time of 2 hours. The graded early warning information includes fault location path, handling suggestions, impact assessment, and operation and maintenance work order. When a common cause fault is determined, the impact assessment provides a list of affected charging piles and a recommended shutdown area.

[0050] This invention also discloses a cloud-edge collaborative diagnosis and early warning method for charging pile faults, applicable to, for example... Figure 1 The cloud-edge collaborative diagnosis and early warning system for charging pile faults shown includes the following steps: S1 Data Acquisition and Processing Steps: Multimodal operation data of each charging pile in each charging station is collected by edge side nodes deployed in each charging station, and edge feature data is obtained after preprocessing and feature extraction. S2 Diagnostic Protection Steps: The edge-side node uses a lightweight fault diagnosis model to perform local real-time diagnosis of each charging pile using edge feature data, obtains the first diagnosis result, the fault urgency level and the first confidence level, and triggers local protection action when the first diagnosis result meets the emergency protection conditions. S3 Dynamic Decision-Making Steps: Edge nodes acquire current computing power reserves and communication quality, and make dynamic decisions based on computing power reserves, communication quality, first confidence level, and fault urgency, combined with preset diagnostic conditions. S4 condition judgment steps: When the preset diagnostic conditions are not met, the edge-side node performs local closed-loop diagnosis; when the preset diagnostic conditions are met, the edge-side node uploads the edge feature data to the cloud platform. S5 Secondary Diagnosis Steps: The cloud platform performs secondary diagnosis based on the global fault diagnosis model and the received edge feature data, outputting the second diagnosis result and its second confidence level; and when a charging pile in the same or multiple charging stations fails, it performs correlation analysis based on the edge feature data of each charging pile to obtain common cause fault and fault area information. S6 Confidence Fusion Steps: The cloud platform fuses the first and second diagnostic results based on the first and second confidence levels to obtain the single-pile diagnostic result; where no second diagnostic result is generated during local closed-loop diagnosis, the first diagnostic result is used as the single-pile diagnostic result. S7 Result Output Steps: The cloud platform determines and outputs the final diagnosis result based on the single-pile diagnosis result, common-cause faults, and fault area information.

[0051] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cloud-edge collaborative diagnosis and early warning system for charging pile faults, characterized in that, include: The edge-side node network and cloud platform consist of multiple edge-side nodes, each deployed in different charging stations. Each edge-side node includes a data acquisition unit, an edge diagnostic unit, and a task scheduling unit connected in sequence. The cloud platform includes a cloud control unit, a global diagnostic unit, a result fusion unit, and a diagnostic output unit connected in sequence. The cloud control unit is connected to the task scheduling unit of each edge-side node and also communicates with each edge-side node. The data acquisition unit is used to collect multimodal operation data of each charging pile in the corresponding charging station, and obtains edge feature data after preprocessing and feature extraction. The edge diagnostic unit is used to perform local real-time diagnosis of each charging pile based on a lightweight fault diagnosis model and edge feature data to obtain the first diagnostic result, the fault urgency level and the first confidence level, and to trigger local protection action when the first diagnostic result meets the emergency protection conditions. The task scheduling unit is used to obtain the computing power margin and communication quality of the edge computing node. Based on the computing power margin, communication quality, first confidence level and fault urgency, it makes dynamic decisions in combination with preset diagnostic conditions. If the preset diagnostic conditions are not met, local closed-loop diagnosis is performed; if they are met, edge feature data is uploaded. The cloud control unit is used to control the start and stop actions of each edge node and transmit edge feature data to the global diagnostic unit; The global diagnostic unit is used to perform secondary diagnosis based on the global fault diagnosis model and edge feature data, and output the second diagnostic result and its second confidence level; it is also used to perform correlation analysis based on the edge feature data of each charging pile when a fault occurs in the same or multiple charging piles, and obtain information on common cause faults and fault areas. The result fusion unit is used to fuse the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level to obtain the single-pile diagnostic result; wherein, when a local closed-loop diagnosis is performed but no second diagnostic result is generated, the first diagnostic result is used as the single-pile diagnostic result; The diagnostic output unit is used to determine the final diagnostic result based on the single-pile diagnostic results, common-cause faults, and fault area information.

2. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, Multimodal operating data includes electrical data, temperature data, acoustic vibration data, insulation monitoring data, and the operating logs and fault codes of the charging pile controller.

3. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, The AI ​​model compression technology based on the teacher-student model uses the global fault diagnosis model as the teacher model. After knowledge distillation and model compression, a lightweight fault diagnosis model is obtained as the student model. The global fault diagnosis model includes a parallel rule matching module and a neural network learning module. The rule matching module is used to match known fault types to edge feature data based on a preset fault branch tree and expert rules, and output a fault determination path. The neural network learning module is used to train a deep neural network model using historical charging pile fault samples. Edge feature data is input into the trained deep neural network model to obtain charging pile fault identification results.

4. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, The preset diagnostic conditions include: the first confidence level is lower than the preset confidence threshold and the communication quality is not lower than the preset quality threshold; the first confidence level is in the preset fuzzy range; the fault urgency is lower than the preset highest urgency level and the first confidence level is lower than the preset confidence threshold; the computing power margin is lower than the preset computing power threshold and the communication quality is not lower than the preset quality threshold. The task scheduling unit is used to upload edge feature data to the global diagnostic unit when any of the preset diagnostic conditions are met.

5. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, The global diagnostic unit performs correlation analysis based on the edge feature data of each charging pile to obtain information on common cause faults and fault areas, specifically including: The edge feature data of each charging pile is extracted according to the preset time window, the correlation coefficient and feature similarity between each pair of charging piles are calculated, and the three elements of the charging pile fault are determined. The three elements of the fault include the charging pile identity information and the time and location of the fault. When the correlation coefficient exceeds the preset limit and the number of charging piles involved is no less than two, a common cause fault is determined to exist. Based on the similarity of characteristics of each charging pile and the three elements of fault, the faults within the pile, common faults on the site side, and disturbances on the power grid side are distinguished, and the fault area information is determined accordingly. The fault area information includes the charging site, transformer area, or power supply branch where the faulty charging pile is located.

6. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 3, characterized in that, The result fusion unit performs confidence fusion of the first diagnostic result and the second diagnostic result based on the first confidence level and the second confidence level, specifically including: The first and second diagnostic results are weighted and fused using the first and second confidence levels as weights, or the basic probability distribution of the first and second diagnostic results is synthesized using DS evidence theory, and the single-case diagnostic result and the overall confidence level are output.

7. A charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 6, characterized in that, The cloud platform also includes a model evolution unit, which is bidirectionally connected to the global diagnostic unit and connected to the result fusion unit and the edge diagnostic unit, respectively. The model evolution unit is used to identify single-pile diagnostic results with a comprehensive confidence score less than a preset comprehensive confidence threshold as difficult examples, incrementally train the global fault diagnosis model, and generate a differentiated lightweight model update package adapted to each edge node by knowledge distillation and quantization compression of the incrementally trained global fault diagnosis model, and then distribute it to the corresponding edge node.

8. The charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, Each edge node also includes an edge caching unit, which is connected to the task scheduling unit and the cloud control unit respectively. It is used to cache edge feature data when the network is interrupted or the communication quality is not lower than a preset quality threshold, and to resume transmission after the network is restored.

9. A charging pile fault cloud-edge collaborative diagnosis and early warning system according to claim 1, characterized in that, The cloud platform also includes a lifespan prediction and early warning unit, which is connected to the cloud historical data storage and historical operation and maintenance record library and the diagnostic output unit, respectively. The lifespan prediction and early warning unit is used to obtain the degradation trend characteristics and historical operation and maintenance records of each charging pile from the cloud historical data storage and historical operation and maintenance record library, and predict the remaining lifespan of the charging pile using the degradation model according to the preset cycle.

10. A method for cloud-edge collaborative diagnosis and early warning of charging pile faults, characterized in that, The charging pile fault cloud-edge collaborative diagnosis and early warning system according to any one of claims 1-9 specifically includes the following steps: S1 Data Acquisition and Processing Steps: Multimodal operation data of each charging pile in each charging station is collected by edge side nodes deployed in each charging station, and edge feature data is obtained after preprocessing and feature extraction. S2 Diagnostic Protection Steps: The edge-side node uses a lightweight fault diagnosis model to perform local real-time diagnosis of each charging pile using edge feature data, obtains the first diagnosis result, the fault urgency level and the first confidence level, and triggers local protection action when the first diagnosis result meets the emergency protection conditions. S3 Dynamic Decision-Making Steps: Edge nodes acquire current computing power reserves and communication quality, and make dynamic decisions based on computing power reserves, communication quality, first confidence level, and fault urgency, combined with preset diagnostic conditions. S4 condition judgment steps: When the preset diagnostic conditions are not met, the edge-side node performs local closed-loop diagnosis; when the preset diagnostic conditions are met, the edge-side node uploads the edge feature data to the cloud platform. S5 Secondary Diagnosis Steps: The cloud platform performs secondary diagnosis based on the global fault diagnosis model and the received edge feature data, outputting the second diagnosis result and its second confidence level; and when a charging pile in the same or multiple charging stations fails, it performs correlation analysis based on the edge feature data of each charging pile to obtain common cause fault and fault area information. S6 Confidence Fusion Steps: The cloud platform fuses the first and second diagnostic results based on the first and second confidence levels to obtain the single-pile diagnostic result; where no second diagnostic result is generated during local closed-loop diagnosis, the first diagnostic result is used as the single-pile diagnostic result. S7 Result Output Steps: The cloud platform determines and outputs the final diagnosis result based on the single-pile diagnosis result, common-cause faults, and fault area information.

Images