Charging pile real-time state monitoring and early warning system based on edge calculation

Abstract

The invention discloses a charging pile real-time state monitoring and early warning system based on edge calculation. Comprising an edge sensing module, an edge calculation module, a fault prediction module, an AR maintenance module, a security control module, a resource scheduling module and a communication transmission module. Electrical-mechanical-environment multi-modal data are collected through a multi-element heterogeneous sensor array, data cleaning and feature decoupling are completed through an edge calculation module, and fault prediction and four-level grading early warning are achieved through an attention mechanism enhanced LSTM-FFT hybrid algorithm and time-space consistency verification. According to the system and the method, AR visual maintenance and a block chain + AES-256 security protection system are combined, edge-cloud dynamic resource scheduling is matched, real-time response, accurate early warning and remote cooperative operation and maintenance are realized, real-time monitoring, accurate early warning, secure transmission and efficient operation and maintenance of the state of the charging pile are realized, the operation reliability and the operation and maintenance efficiency of equipment are improved, and the system and the method are suitable for popularization and application. And application requirements in different scenes can be met.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for charging piles, and specifically discloses a real-time status monitoring and early warning system for charging piles based on edge computing. Background Technology

[0002] With the explosive growth of electric vehicle ownership, charging piles, as core energy replenishment infrastructure, directly impact user experience and public safety through their operational stability, safety, and maintenance efficiency. Existing charging pile monitoring systems mostly adopt the traditional architecture of "terminal data collection - centralized cloud processing," which has the following key technical shortcomings: Poor real-time performance: Raw data (voltage, current, temperature, etc.) are uploaded to the cloud in full. Network transmission delay (usually ≥200ms) causes the fault warning response to be delayed, making it impossible to deal with emergency hazards such as short circuits and overheating in a timely manner. Severe bandwidth consumption: A single pile generates more than 10GB of raw data per day. When deploying a large-scale cluster, the cloud access bandwidth consumption exceeds 70%, which can easily lead to network congestion and data loss. Low fault prediction accuracy: It relies only on single-dimensional data (such as electrical parameters) and simple threshold judgment, lacks in-depth fusion analysis of multi-source data, has a high false alarm rate and a prominent risk of missed alarms; Low maintenance efficiency: Fault location can only be accurate to the pile level, and cannot locate specific faulty components (such as charging modules and contactors). In addition, there is a lack of remote visual guidance, and the average fault repair time exceeds 4 hours. Weak security protection: There is no comprehensive protection system for encryption mechanisms, device authentication and malicious data attacks for communication between edge computing nodes and the cloud, which poses a risk of data leakage and illegal manipulation.

[0003] Therefore, it is necessary to invent a real-time status monitoring and early warning system for charging piles based on edge computing to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides a real-time status monitoring and early warning system for charging piles based on edge computing. It collects multimodal data on electrical, mechanical, and environmental aspects through a multi-element heterogeneous sensor array, performs data cleaning and feature decoupling via an edge computing module, and employs an attention-enhanced LSTM-FFT hybrid algorithm and spatiotemporal consistency verification to achieve high-precision fault prediction and four-level hierarchical early warning. Combined with AR visualization maintenance and a blockchain + AES-256 security protection system, along with dynamic resource scheduling between the edge and cloud, it effectively solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time status monitoring and early warning system for charging piles based on edge computing, specifically including an edge sensing module, an edge computing module, a cloud analysis module, an AR maintenance module, a security management module, and a resource scheduling module: Edge sensing module: A multi-element heterogeneous sensor array deployed locally on the charging pile, used to collect raw multi-dimensional data in real time. The raw multi-dimensional data specifically includes electrical operating parameters, mechanical status parameters and environmental parameters. Edge computing module: An edge computing node set up locally on the charging pile or near the aggregation gateway, used to perform data cleaning, feature extraction and real-time anomaly detection; Cloud analytics module: A big data analytics engine deployed on a cloud computing platform, responsible for global model training and optimization, cluster status monitoring and control, and historical data tracing and analysis; Fault prediction module: Employs an attention-enhanced LSTM-FFT hybrid algorithm to achieve fault prediction and graded early warning based on multi-dimensional data fusion; AR Maintenance Module: Based on computer vision and augmented reality technologies, it provides equipment model recognition, fault visualization annotation, and remote expert collaborative maintenance support. Security management module: integrates blockchain identity authentication, AES-256 encrypted transmission and xDeepCIN (eXtensible Deep Cross-Iteration Network) deep learning model to achieve device authentication, data encryption and attack detection; Resource scheduling module: Based on Docker containerization technology and gradient descent optimization of the minimum rate of change algorithm, it realizes dynamic allocation and load balancing of edge-cloud computing resources; Communication transmission module: Responsible for secure communication of cloud-edge collaborative data transmission and command interaction.

[0006] Preferably, the edge sensing module includes a temperature sensor, a vibration sensor, an energy metering chip, a door magnetic switch, and a temperature and humidity sensor, and adopts a configurable sampling frequency of 1Hz-100Hz, supporting adaptive adjustment of the sampling frequency according to the charging pile's operating status (standby / charging / fault).

[0007] Preferably, the edge computing module uses a stacked autoencoder to realize nonlinear feature mapping, converts the original multidimensional data into low-dimensional feature vectors, and decomposes them into electrical sub-vectors (voltage, current, power) and mechanical sub-vectors (vibration frequency, switching state) through a feature decoupling algorithm, and performs time-domain-frequency domain joint analysis respectively.

[0008] Preferably, the fault prediction module captures long-term dependencies in time series data through an attention-enhanced LSTM network, extracts frequency domain features of vibration and current signals using FFT, integrates the output results of the two models using a weighted voting fusion algorithm, establishes a dynamic threshold mechanism, and constructs a four-level fault classification system corresponding to different response strategies.

[0009] Preferably, the system adopts a spatiotemporal consistency verification mechanism. In the time dimension, the deviation between the current data and the historical data of the same period is compared through a sliding window. In the spatial dimension, the data of the same type of charging pile in the region are cross-validated. Combined with a confidence screening algorithm, the authenticity of abnormal states is identified.

[0010] Preferably, the AR maintenance module identifies the charging pile model and faulty components using the YOLOv8 target detection algorithm, and overlays and displays the fault location, abnormal parameter values, and standardized maintenance steps based on the equipment 3D model reconstruction technology; it also supports interactive annotation guidance from remote experts through real-time audio and video connections, realizing collaborative operation and maintenance of "on-site personnel + remote experts".

[0011] Preferably, the security management module uses blockchain technology to achieve distributed identity authentication between edge computing nodes and the cloud, and uses the AES-256 algorithm to encrypt transmitted data end-to-end; it uses Markov chain Monte Carlo technology to estimate network parameters, and combines the xDeepCIN deep learning model to analyze network traffic and data patterns, accurately detecting aggregation attacks, data tampering and unauthorized access behaviors.

[0012] Preferably, the resource scheduling module uses Docker containers to isolate and deploy edge computing tasks, predicts the container's CPU and memory resource requirements through a minimum rate of change algorithm optimized by gradient descent, and pre-allocates resources in advance; it adopts a multi-timescale control strategy: millisecond-level response to security-related tasks, second-level processing of status analysis tasks, and hourly execution trend prediction and model update tasks.

[0013] Preferably, the system further includes: Cloud analytics module: A big data analytics engine deployed on a cloud computing platform, responsible for global model training and optimization, cluster status monitoring and control, and historical data tracing and analysis; Communication transmission module: responsible for secure communication of cloud-edge collaborative data transmission and command interaction; The communication transmission module supports adaptive switching between multiple protocols such as MQTT, HTTP / 2, and Modbus-TCP. When network bandwidth is sufficient, HTTP / 2 is used to transmit batch historical data, while in bandwidth-limited scenarios, the MQTT protocol is switched to transmit critical abnormal data. The communication link status is monitored in real time through a heartbeat mechanism, and data is automatically cached when the connection is lost, and the transmission is resumed after reconnection.

[0014] Preferably, the system provides a standardized API interface to support integration with third-party charging pile operation platforms, urban smart transportation systems, and fire early warning systems; the edge computing nodes adopt a modular design to support sensor expansion and algorithm iteration upgrades, adapting to the personalized monitoring needs of different types of charging piles.

[0015] The technical effects and advantages of this invention are as follows: 1. Significantly improved real-time response capability: Localized processing at the edge enables anomaly detection and early warning response in ≤30ms, which is more than 85% faster than the response speed of traditional cloud systems (≥200ms), and can quickly handle critical faults such as short circuits and overheating; 2. High accuracy in fault prediction: The multimodal fusion algorithm combined with spatiotemporal consistency verification achieves a fault detection accuracy of ≥95% and a false alarm rate of less than 10%, which is an improvement in accuracy and a reduction in false alarm rate compared to the traditional single threshold method. 3. Optimized resource utilization efficiency: In a test environment with a cluster of 1,000 charging piles (each pile generates an average of 10GB of raw data per day), the dynamic resource scheduling algorithm achieves edge-cloud load balancing, improving infrastructure utilization by 35% (average CPU / memory utilization increased from 55% to 74%), reducing network bandwidth utilization from 70% to 28% (a reduction of 60%), and reducing the annual operating cost of a single cluster by approximately RMB 120,000. 4. Comprehensive Security Upgrade: Blockchain identity authentication + end-to-end encryption + deep learning attack detection, anomaly detection F1 score is 32.3% higher than traditional methods, effectively resisting security threats such as data leakage and malicious attacks. Attached Figure Description

[0016] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0018] Figure 2 This is a flowchart of the fault prediction and graded early warning process of the present invention.

[0019] Figure 3 This is a flowchart of the AR maintenance and collaborative operation and maintenance process of the present invention.

[0020] Figure 4 This is the core process of the resource scheduling module of the present invention. Detailed Implementation

[0021] 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.

[0022] This invention provides, for example Figure 1 The edge computing-based real-time status monitoring and early warning system for charging piles shown includes an edge sensing module, an edge computing module, a cloud analysis module, a fault prediction module, an AR maintenance module, a safety management and control module, a resource scheduling module, and a communication transmission module. Edge sensing module: A multi-element heterogeneous sensor array deployed locally on the charging pile, used to collect raw multi-dimensional data in real time. The raw multi-dimensional data specifically includes electrical operating parameters, mechanical status parameters and environmental parameters. Furthermore, in the above technical solution, the edge sensing module includes a temperature sensor, a vibration sensor, an energy metering chip, a door magnetic switch, and a temperature and humidity sensor. It adopts a configurable sampling frequency of 1Hz-100Hz and supports adaptive adjustment of the sampling frequency according to the charging pile's operating status (standby / charging / fault).

[0023] It should be further noted that all sensors are connected to the edge computing nodes via industrial-grade interfaces such as RS-485, I2C, and SPI. Electrical operating parameters: A high-precision power metering chip with an error of ≤0.2% is integrated into the current / voltage sampling circuit of the charging pile main control board to collect parameters such as voltage, current, power, and power factor in real time. Mechanical condition parameters: The vibration sensor is a piezoelectric vibration sensor with a sampling rate of 100Hz, used to detect the internal mechanical structure of the charging pile. It is fixed to the part of the internal structure of the charging pile that vibrates significantly and detects abnormal vibration of the fan and relay. The door magnetic switch is installed on the inside of the charging pile cabinet door to detect illegal opening or abnormal status. Environmental parameters: The temperature sensor is a digital temperature sensor with an accuracy of ±0.1℃, which is placed in heat-sensitive areas such as the power module heat sink and wiring terminals to monitor the temperature changes of key parts such as the charging module, contactor, and cable joint in real time. The temperature and humidity sensor is placed in the ventilation area inside the cabinet to monitor the temperature and humidity of the charging pile and prevent failures caused by environmental factors such as condensation and overheating.

[0024] Furthermore, the system supports configurable sampling frequencies from 1Hz to 100Hz; The data collection frequency is dynamically adjusted based on the charging pile's operating status (standby / charging / fault). Standby mode (no charging request from the charging station for more than 5 minutes): Low-frequency sampling, such as 1Hz, is used to reduce power consumption; Charging status (detected current ≥0.5A): Increase sampling frequency to medium to high frequency, such as 10Hz–50Hz, and focus on monitoring electrical and mechanical parameters; Fault status (abnormal parameters detected): Switch to the highest sampling frequency (100Hz) to achieve high-frequency data capture, which facilitates accurate location of the fault source.

[0025] Edge computing module: An edge computing node set up locally on the charging pile or near the aggregation gateway, used to perform data cleaning, feature extraction and real-time anomaly detection; Furthermore, in the above technical solution, the edge computing module uses a stacked autoencoder to realize nonlinear feature mapping, converts the original multidimensional data into low-dimensional feature vectors, and decomposes them into electrical sub-vectors (voltage, current, power) and mechanical sub-vectors (vibration frequency, switching state) through a feature decoupling algorithm, and performs time-domain-frequency domain joint analysis respectively.

[0026] A1. Data Preprocessing Workflow The edge computing module receives raw multidimensional data from the edge sensing module in real time; it initiates a sliding window processing mechanism, setting the window length to 5 seconds and the step size to 1 second, and performs moving average smoothing on the raw data. It uses the 3σ criterion to remove outliers, calculates the mean μ and standard deviation σ for the historical data (most recent 1000 sampling points) of each sensor, and determines and removes outliers when the current sampling data exceeds the range of [μ-3σ, μ+3σ]. For sensor drift, linear interpolation is used for compensation; for missing data caused by transmission interruption, cubic spline interpolation is used to complete the missing data based on the previous and next 5 data points.

[0027] A2. Feature Extraction and Decoupling A stacked autoencoder is used to perform nonlinear feature learning on the original multidimensional data; the processed original multidimensional data is mapped into low-dimensional, dense feature vectors, which retains key state information while significantly reducing data dimensionality. It should be further explained that the stacked autoencoder model uses historical normal operation data (100,000+ samples) to train a 3-layer autoencoder. The input layer has 12 dimensions (voltage, current, power, power factor, 3 vibration accelerations, door magnetic switch status, 3 temperatures, and temperature and humidity), the hidden layer has 3 layers (8-dimensional, 6-dimensional, and 4-dimensional), and the output layer is a 4-dimensional low-dimensional feature vector. The training iterations are 1000 rounds, and the learning rate is 0.001.

[0028] The 4-dimensional low-dimensional feature vector is decomposed into electrical sub-vectors (2-dimensional: voltage-power integrated feature, current-power factor integrated feature) and mechanical sub-vectors (2-dimensional: vibration frequency feature, switching state feature) by a feature decoupling algorithm (based on mutual information minimization). A3. Joint Time-Frequency Domain Analysis Electrical vector time-domain analysis: Calculate the voltage sag and current harmonic distortion (THD) within each sampling window, where the voltage sag = (rated voltage - actual voltage) / rated voltage × 100%, and the THD is calculated using fast Fourier transform.

[0029] Mechanical subvector frequency domain analysis: Perform FFT transformation on vibration acceleration data to extract resonant frequency and vibration energy in the 10-1000Hz frequency band. The resonant frequency is the frequency corresponding to the maximum amplitude in the frequency domain, and the vibration energy is the integral of the amplitude in the frequency domain.

[0030] Based on the decoupled electrical and mechanical sub-vectors, anomaly scoring models are constructed respectively: Electrical anomaly score = voltage sag amplitude × 0.6 + THD × 0.4; Mechanical anomaly score = (vibration energy / rated vibration energy) × 0.7 + abnormal switch state duration × 0.3 (abnormal switch state duration refers to the duration of continuous disconnection of the door magnetic switch for more than 5 seconds).

[0031] A lightweight GRU classifier takes anomaly scores and feature vectors as input and outputs state discrimination results, where the state discrimination logic is as follows: Normal: Both sub-scores (electrical anomaly score and mechanical anomaly score) are <0.3; Warning: If any sub-score is 0.3 or less than 0.7, and the other sub-score is less than 0.7; Abnormal: Any sub-score ≥ 0.7.

[0032] The processed raw multidimensional data, extracted feature vectors, and state discrimination results are uploaded to the cloud analysis module in real time. Cloud analytics module: A big data analytics engine deployed on a cloud computing platform, responsible for global model training and optimization, cluster status monitoring and control, and historical data tracing and analysis; Furthermore, the cloud analytics module receives raw data, feature vectors, and status discrimination results uploaded by edge computing nodes and stores them in an HBase database: the raw data is stored in partitions according to "charging pile ID-timestamp" and retained for 90 days; the feature vectors and status discrimination results are stored in Elasticsearch, supporting fast retrieval.

[0033] The big data analytics engine automatically extracts all historical data (normal and fault data of all charging piles) during the daily off-peak charging period (2-4 am) and retrains the attention-enhanced LSTM-FFT hybrid algorithm model to optimize the model parameters (learning rate 0.0001, number of iterations 500).

[0034] Based on cluster status data, adjust the dynamic threshold of the fault prediction model: calculate the statistical distribution of abnormal scores of charging piles of the same type in the region, and set the threshold to the 95th percentile to ensure the adaptability of the threshold.

[0035] The cloud analytics module receives real-time status data from each edge computing node and displays the operating status of the charging pile cluster through a visualization platform, including the real-time status of each charging pile (standby / charging / warning / abnormal), key parameters (voltage, current, temperature), and fault statistics (number of faults and fault types in the past 7 days).

[0036] Historical data traceability and analysis: Supports querying historical data by charging pile ID, time range, and fault type, and generates data trend curves (including temperature change trend and power change trend) to provide data support for operation and maintenance decisions.

[0037] Fault prediction module: Employs an attention-enhanced LSTM-FFT hybrid algorithm to achieve fault prediction and graded early warning based on multi-dimensional data fusion; Furthermore, in the above technical solution, the fault prediction module captures the long-term dependencies of time series data through an attention-enhanced LSTM network, extracts the frequency domain features of vibration and current signals using FFT, integrates the output results of the two models using a weighted voting fusion algorithm, establishes a dynamic threshold mechanism, and constructs a four-level fault classification system corresponding to different response strategies.

[0038] It should be further explained that the fault prediction and graded early warning process is as follows: Figure 2 As shown, where: An attention-enhanced LSTM network: Input is time-series data of electrical and mechanical subvectors (the most recent 60 sampling windows), the network has 3 layers (64, 32, and 16 neurons in the hidden layer), the attention mechanism weights are optimized through training to capture long-term data dependencies, and the output is a fault probability prediction value P1.

[0039] FFT frequency domain feature extraction: Perform FFT transformation on vibration acceleration and current signals to extract 5 key frequency domain features (fundamental frequency, second harmonic, third harmonic, resonant frequency, and frequency band energy), input them into a fully connected neural network, and output the fault probability prediction value P2.

[0040] Weighted voting fusion: Set weights ω1=0.6, ω2=0.4, and the failure probability after fusion is P=ω1×P1+ω2×P2.

[0041] Dynamic threshold calculation: Based on the cloud-optimized threshold, the threshold T is adjusted in combination with the current charging pile operating status (charging power, running time). When the charging power is ≥50kW, T is reduced by 10%; when the running time exceeds 24 hours of continuous operation, T is reduced by 15%.

[0042] Four-level fault classification system: Level 1 fault (emergency fault): P≥0.9, corresponding to critical faults such as short circuit and over-temperature (temperature≥85℃); Level 2 fault (serious fault): 0.7≤P<0.9, corresponding to faults affecting the charging function such as charging module failure and relay sticking; Level 3 fault (general fault): 0.5≤P<0.7, corresponding to fan failure, cabinet door not closed, etc., which do not affect the current charging but need to be dealt with in time; Level 4 fault (minor fault): 0.3≤P<0.5, corresponding to potential hazards such as sensor drift and slight parameter overshoot.

[0043] Tiered early warning response Level 1 fault: Immediately send a shutdown command (cut off the charging circuit) to the main control board of the charging pile, and notify the maintenance personnel via SMS and APP push (respond within 15 minutes). Level 2 fault: Send a pause charging command to the charging station and push a notification to maintenance personnel (response within 1 hour); Level 3 fault: Continuously monitor data and push notifications to maintenance personnel (respond within 24 hours); Level 4 fault: Record potential risks, focus on checking during the next maintenance, and conduct regular (7-day) reassessments. If the P value rises above 0.5, upgrade the warning level.

[0044] AR Maintenance Module: Based on computer vision and augmented reality technologies, it provides equipment model recognition, fault visualization annotation, and remote expert collaborative maintenance support. Furthermore, in the above technical solution, the AR maintenance module identifies the charging pile model and faulty components through the YOLOv8 target detection algorithm, and overlays and displays the fault location, abnormal parameter values, and standardized maintenance steps based on the equipment 3D model reconstruction technology; it also supports remote experts to provide interactive annotation guidance through real-time audio and video connections, realizing collaborative operation and maintenance of "on-site personnel + remote experts".

[0045] In a preferred embodiment of the present invention, the AR maintenance and collaborative operation process is as follows: Figure 3As shown, on-site maintenance personnel can identify the charging pile model and faulty parts in real time by taking pictures of the charging pile as a whole and the faulty area, and by using the YOLOv8 object detection model (the pre-trained dataset contains 100 mainstream charging pile models and 50 common faulty parts). If the identification fails, the charging pile model and faulty area can be manually selected, and the cloud database can be called to supplement the model identification data to improve the subsequent identification accuracy.

[0046] Based on the equipment 3D model reconstruction technology (downloading the 3D model of the corresponding charging pile model from the cloud), the fault location (highlighted in red), abnormal parameter values, and standardized maintenance procedures are overlaid on the real-time camera image.

[0047] On-site personnel can initiate a remote expert collaboration request, select the corresponding fault type and professional maintenance expert, and establish a real-time audio and video connection. The remote expert can view the real-time on-site video and use interactive annotation tools (such as arrows, circles, and text) to mark key repair points. The annotation content is synchronized to the on-site personnel in real time. The expert can provide voice guidance on the repair steps, and the on-site personnel can provide real-time feedback on the repair progress.

[0048] After the repair is completed, on-site personnel upload repair photos and repair step records, and the system automatically compares the equipment status data (such as temperature and vibration parameters) before and after the repair to confirm whether the fault has been eliminated.

[0049] Maintenance records are stored in a cloud database, forming a closed loop of operation and maintenance, providing data support for subsequent fault analysis and model optimization.

[0050] Security management module: integrates blockchain identity authentication, AES-256 encrypted transmission and xDeepCIN deep learning model to achieve device authentication, data encryption and attack detection; Furthermore, in the above technical solution, the security management module uses blockchain technology to achieve distributed identity authentication between edge computing nodes and the cloud, and uses the AES-256 algorithm to encrypt the transmitted data end-to-end; it uses Markov chain Monte Carlo technology to estimate network parameters, and combines the xDeepCIN deep learning model to analyze network traffic and data patterns, accurately detecting aggregation attacks, data tampering and unauthorized access behaviors.

[0051] It should be further explained that edge computing nodes and the cloud generate public and private key pairs respectively, and the public key is uploaded to the blockchain network (consortium blockchain, the nodes include all edge computing nodes, cloud servers, and operation and maintenance terminals) to form a distributed identity ledger.

[0052] Before an edge computing node communicates with the cloud, it sends its signed identity information (including device ID and public key) to the cloud. After the cloud verifies the validity of the signature, it queries the edge computing node's public key from the blockchain ledger to complete two-way authentication. If authentication fails, communication is refused, and the authentication log is stored on the blockchain and cannot be tampered with.

[0053] The AES-256 algorithm is used to encrypt the transmitted data end-to-end: the edge computing node negotiates and generates a session key with the cloud (based on the ECDH key exchange algorithm), and the session key is updated every 24 hours; before data transmission, the data is encrypted using the session key (encryption mode CBC, initialization vector IV is randomly generated), and only the encrypted data is transmitted during the transmission process to prevent data leakage.

[0054] Network parameter estimation: Markov chain Monte Carlo technique is used to estimate network traffic parameters (data packet size, transmission frequency, source IP address distribution) in real time and establish a normal traffic model.

[0055] xDeepCIN deep learning model: Input network traffic features (100 dimensions, including packet length statistics, protocol type, connection duration, etc.), the model analyzes traffic patterns in real time, and detects aggregation attacks (such as DDoS attacks), data tampering (by verifying data hash values, hash algorithm SHA-256), and illegal access (access requests that have not passed blockchain identity authentication).

[0056] Upon detecting an attack, the system immediately disconnects the abnormal communication link, records the attack log (including attack time, attack type, and source IP address), and sends a security alert to the operations and maintenance personnel. For unauthorized access requests, the system refuses to provide any data services and restricts the access permissions of the IP address.

[0057] Resource scheduling module: Based on Docker containerization technology and gradient descent optimization of the minimum rate of change algorithm, it realizes dynamic allocation and load balancing of edge-cloud computing resources; Furthermore, in the above technical solution, the resource scheduling module uses Docker containers to isolate and deploy edge computing tasks. It predicts container CPU and memory resource requirements using a gradient descent-optimized minimum rate of change algorithm, and pre-allocates resources accordingly. A multi-timescale control strategy is employed: millisecond-level response to security-related tasks, second-level processing of status analysis tasks, and hourly execution trend prediction and model update tasks. Its core process is as follows: Figure 4 As shown.

[0058] It should be further explained that the computing tasks of the edge computing node (data cleaning, feature extraction, and anomaly detection) are encapsulated into three independent Docker containers: data processing container, feature extraction container, and anomaly detection container. Each container is configured with an independent resource quota (initial CPU quota of 20% and memory quota of 256MB).

[0059] The cloud computing tasks (model training, global monitoring, and historical data analysis) are encapsulated into four Docker containers: model training container, monitoring container, data retrieval container, and API service container. Resource quotas are configured according to the task complexity (50% CPU quota and 16GB memory quota for the model training container).

[0060] The minimum rate of change algorithm for gradient descent optimization: Input historical resource usage data (CPU and memory usage over the last 7 days) and predict the resource requirements of each container in the next hour.

[0061] Furthermore, the calculation process of the minimum rate of change algorithm for gradient descent optimization is as follows: Input: Historical resource usage sequence R[t], time window T; Output: Predicted resource demand P[t+T]; Calculate the rate of change of resources ΔR = (R[t] - R[tT]) / R[tT]; The gradient descent method is used to optimize the weights W and minimize the prediction error. Output P[t+T]=W×ΔR+R[t]; Based on the prediction results, adjust the resource quotas of containers in advance. For example, when the resource demand for fault prediction tasks increases, appropriately increase the CPU quota of model training containers; when edge computing node resources are scarce, prioritize the resource supply for security-related tasks.

[0062] Millisecond-level tasks (security-related tasks): including attack detection and emergency fault response, are allocated the highest priority resources, and CPU scheduling uses the FIFO algorithm to ensure response latency ≤10ms.

[0063] Second-level tasks (status analysis tasks): including data cleaning, feature extraction, and real-time anomaly detection. CPU scheduling uses the CFS algorithm, resource quotas are dynamically adjusted, and response latency is ≤1 second.

[0064] Hourly tasks (trend prediction and model update): These include global model training and historical data trend analysis. They are scheduled to be executed during off-peak hours and use low-priority scheduling to avoid occupying resources during peak hours.

[0065] The cloud monitors the resource utilization (CPU utilization, memory utilization, network bandwidth utilization) of edge computing nodes in real time. When the resource utilization of a certain edge computing node exceeds 80% for 5 consecutive minutes, some non-real-time tasks (such as historical data uploads) are migrated to nearby idle edge computing nodes. When the cloud resource utilization exceeds 70%, elastic expansion is initiated to add temporary server nodes to share the computing pressure.

[0066] Communication transmission module: Responsible for secure communication of cloud-edge collaborative data transmission and command interaction.

[0067] Furthermore, in the above technical solution, the communication transmission module supports adaptive switching of multiple protocols such as MQTT, HTTP / 2, and Modbus-TCP. When network bandwidth is sufficient, HTTP / 2 is used to transmit batch historical data, and in bandwidth-limited scenarios, the MQTT protocol is switched to transmit critical abnormal data. The communication link status is monitored in real time through a heartbeat mechanism, and data is automatically cached when the connection is lost, and the transmission is resumed after reconnection.

[0068] Furthermore, in the above technical solution, the system adopts a spatiotemporal consistency verification mechanism. In the time dimension, the deviation between the current data and the historical data of the same period is compared through a sliding window. In the spatial dimension, the data of the same type of charging piles in the region are cross-validated. Combined with a confidence screening algorithm, the authenticity of abnormal states is identified.

[0069] Furthermore, in the above technical solution, the system provides a standardized API interface to support integration with third-party charging pile operation platforms, urban smart transportation systems, and fire early warning systems; the edge computing nodes adopt a modular design to support sensor expansion and algorithm iteration upgrades, adapting to the personalized monitoring needs of different types of charging piles.

[0070] It should be further explained that the communication transmission module monitors the network bandwidth status in real time and calculates the network bandwidth (bandwidth = heartbeat packet size / transmission time) through the heartbeat packet mechanism. When the bandwidth is ≥10Mbps, the HTTP / 2 protocol is used to transmit batch historical data (such as daily raw data summaries), which has high transmission efficiency. When the bandwidth is <10Mbps, it automatically switches to the MQTT protocol and only transmits key abnormal data (fault warning information, equipment status change data) to reduce bandwidth consumption.

[0071] Heartbeat mechanism: The edge computing node sends a heartbeat packet (containing device ID, current status, and network parameters) to the cloud every 10 seconds. The cloud receives the packet and returns a response packet. If the edge computing node does not receive a response from the cloud three times in a row, it is determined that the communication is disconnected.

[0072] When the connection is lost, the edge computing node stores the data to be transmitted in the local cache and sorts it by timestamp. After reconnection, the communication module automatically initiates a resume transmission request and continues transmitting data from the position of the last successful transmission to avoid data loss.

[0073] A retransmission mechanism is used to ensure transmission quality. If no acknowledgment response is received after data transmission, it will be automatically retransmitted. The maximum number of retransmissions is 3, with an interval of 1 second between each retransmission. If the retransmission fails 3 times, the data will be stored in the buffer and wait for the next retransmission.

[0074] For large files (such as batch historical data, with a single file ≥100MB), a chunked transmission mechanism is adopted, which divides the file into 10MB chunks and transmits them one by one. After the transmission is completed, the cloud automatically merges them to improve transmission stability.

[0075] Typical application scenario examples: Fast charging station scenario at highway service area It is compatible with 120kW DC fast charging piles. The edge sensing module adopts a 50Hz charging status sampling rate to focus on monitoring voltage fluctuations (threshold ±10% of rated voltage) and power module temperature (threshold ≤85℃). The resource scheduling module prioritizes fault prediction and safety protection tasks, and responds to first-level faults such as short circuits and over-temperature in milliseconds. The AR maintenance module supports remote expert guidance for replacing faulty charging modules, reducing the average fault repair time to 1.5 hours.

[0076] Community slow charging station scenario It is compatible with 7kW AC slow charging piles. The edge sensing module adopts a 10Hz charging status sampling rate to focus on monitoring the current harmonic distortion rate (threshold ≤5%) and cabinet door opening and closing status. The resource scheduling module adopts a low-frequency resource allocation strategy to reduce standby power consumption by 30%. The system integrates with the community smart property platform through a standardized API interface to realize automatic fault information push and maintenance work order generation.

[0077] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time status monitoring and early warning system for charging piles based on edge computing, characterized in that, It includes an edge awareness module, an edge computing module, a fault prediction module, an AR maintenance module, a security management module, and a resource scheduling module. Edge sensing module: A multi-element heterogeneous sensor array deployed locally on the charging pile, used to collect raw multi-dimensional data in real time. The raw multi-dimensional data specifically includes electrical operating parameters, mechanical status parameters and environmental parameters. Edge computing module: An edge computing node set up locally on the charging pile or near the aggregation gateway, used to perform data cleaning, feature extraction and real-time anomaly detection; Fault prediction module: Employs an attention-enhanced LSTM-FFT hybrid algorithm to achieve fault prediction and graded early warning based on multi-dimensional data fusion; AR Maintenance Module: Based on computer vision and augmented reality technologies, it provides equipment model recognition, fault visualization annotation, and remote expert collaborative maintenance support. Security management module: integrates blockchain identity authentication, AES-256 encrypted transmission and xDeepCIN (eXtensible DeepCross-Iteration Network) deep learning model to achieve device authentication, data encryption and attack detection; Resource scheduling module: Based on Docker containerization technology and gradient descent optimization of the minimum rate of change algorithm, it realizes dynamic allocation and load balancing of edge-cloud computing resources.

2. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The edge sensing module includes a temperature sensor, a vibration sensor, an energy metering chip, a door magnetic switch, and a temperature and humidity sensor. It adopts a configurable sampling frequency of 1Hz-100Hz and supports adaptive adjustment of the sampling frequency according to the charging pile's operating status (standby / charging / fault).

3. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The edge computing module uses a stacked autoencoder to implement nonlinear feature mapping, converting the original multidimensional data into low-dimensional feature vectors, and decomposing them into electrical sub-vectors (voltage, current, power) and mechanical sub-vectors (vibration frequency, switching state) through a feature decoupling algorithm, and performing time-domain-frequency domain joint analysis respectively.

4. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The fault prediction module captures long-term dependencies in time-series data through an attention-enhanced LSTM network, extracts frequency domain features of vibration and current signals using FFT, and integrates the output results of the two models using a weighted voting fusion algorithm. It also establishes a dynamic threshold mechanism and constructs a four-level fault classification system corresponding to different response strategies.

5. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The system employs a spatiotemporal consistency verification mechanism. In the time dimension, it compares the deviation between the current data and the historical data of the same period through a sliding window. In the spatial dimension, it cross-verifies the data of the same type of charging piles in the region and combines the confidence filtering algorithm to identify the authenticity of abnormal states.

6. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The AR maintenance module identifies the charging pile model and faulty components using the YOLOv8 target detection algorithm, and overlays the fault location, abnormal parameter values, and standardized maintenance steps based on the equipment 3D model reconstruction technology. It also supports interactive annotation guidance from remote experts through real-time audio and video connections, enabling collaborative operation and maintenance between "on-site personnel and remote experts".

7. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The security management module uses blockchain technology to achieve distributed identity authentication between edge computing nodes and the cloud, and uses the AES-256 algorithm to encrypt transmitted data end-to-end. It uses Markov chain Monte Carlo technology to estimate network parameters and combines the xDeepCIN deep learning model to analyze network traffic and data patterns, accurately detecting aggregation attacks, data tampering and unauthorized access.

8. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The resource scheduling module uses Docker containers to isolate and deploy edge computing tasks. It predicts the CPU and memory resource requirements of containers through a minimum rate of change algorithm optimized by gradient descent and pre-allocates resources in advance. It adopts a multi-timescale control strategy: millisecond-level response to security-related tasks, second-level processing of status analysis tasks, and hourly execution trend prediction and model update tasks.

9. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that, The system also includes: Cloud analytics module: A big data analytics engine deployed on a cloud computing platform, responsible for global model training and optimization, cluster status monitoring and control, and historical data tracing and analysis; Communication transmission module: responsible for secure communication of cloud-edge collaborative data transmission and command interaction; The communication transmission module supports adaptive switching between multiple protocols such as MQTT, HTTP / 2, and Modbus-TCP. When network bandwidth is sufficient, HTTP / 2 is used to transmit batch historical data, while in bandwidth-limited scenarios, the MQTT protocol is switched to transmit critical abnormal data. The communication link status is monitored in real time through a heartbeat mechanism, and data is automatically cached when the connection is lost, and the transmission is resumed after reconnection.

10. The real-time status monitoring and early warning system for charging piles based on edge computing according to claim 1, characterized in that: The system provides a standardized API interface, supporting integration with third-party charging pile operation platforms, urban smart transportation systems, and fire early warning systems; the edge computing nodes adopt a modular design, supporting sensor expansion and algorithm iteration upgrades to adapt to the personalized monitoring needs of different types of charging piles.

Images