Artificial Intelligence-Based Remote Monitoring and Early Warning Method and System for Charging Piles

Abstract

The application discloses a kind of based on artificial intelligence's charging pile remote monitoring early warning method, it is related to artificial intelligence charging relevant technical field, including deployment bionic sensor network and computer vision monitoring unit, real-time monitoring is carried out to charging pile, data pre-processing, analysis is carried out using group intelligence algorithm and division cooperation submodel, judge whether there is fault and determine fault type, fault diagnosis result output and decision, start genetic programming evolution's maintenance strategy module, generate initial maintenance strategy.The application also discloses a kind of based on artificial intelligence's charging pile remote monitoring early warning system, including bionic sensor network, computer vision monitoring unit, data acquisition and transmission module, central processing system, early warning unit, control unit, man-machine interface.The application is monitored to charging pile multidimensionally by deployment bionic sensor network, comprehensively understands the state of charging pile, improves the accuracy of fault detection.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence charging technology, and in particular to a remote monitoring and early warning method and system for charging piles based on artificial intelligence. Background Technology

[0002] With the increasing popularity of electric vehicles, charging stations, as energy replenishment stations for electric vehicles, are being built and used more and more widely. However, various faults may occur during the operation of charging stations, such as overheating, overvoltage, and undervoltage. If these problems are not detected and handled in time, they may lead to damage to the charging station, safety accidents, or even affect the user experience.

[0003] Traditional charging pile monitoring systems rely heavily on manual inspections or simple sensor detection, making it difficult to achieve real-time and comprehensive monitoring. Manual inspections are inefficient and easily affected by human factors, leading to discontinuity and inaccuracy in monitoring. Simple sensor detection usually only provides limited information and is difficult to comprehensively and accurately monitor the operating status of charging piles. Traditional systems often lack the ability to sensitively capture early signs of electrical faults, resulting in delayed fault handling and increased maintenance costs and safety risks.

[0004] Secondly, some AI-based charging pile fault diagnosis methods have achieved certain research results, such as using neural networks to analyze sensor data to classify and predict charging pile faults. However, these methods still have some shortcomings, such as the need to improve diagnostic accuracy, high algorithm complexity, and high hardware performance requirements.

[0005] Finally, existing maintenance strategies are often based on fixed experience or rules, lacking flexibility and adaptability. When faced with complex and ever-changing fault types, these strategies may not provide optimal solutions. The monitoring system and maintenance record database of charging piles contain a large amount of valuable data, but existing technologies often fail to make full use of this data for in-depth analysis and mining, resulting in the data value not being fully realized.

[0006] Therefore, it is necessary to propose a remote monitoring and early warning method and system for charging piles based on artificial intelligence to solve the above problems. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a remote monitoring and early warning method and system for charging piles based on artificial intelligence. This solves the problems of traditional charging pile monitoring systems, which rely heavily on manual inspections or simple sensor detection, making it difficult to achieve real-time and comprehensive monitoring, resulting in poor classification and prediction of charging pile faults, and maintenance strategies that are often based on fixed experience or rules, lacking flexibility and adaptability.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A remote monitoring and early warning method for charging piles based on artificial intelligence includes the following steps:

[0012] Step 1: Sensor Data Acquisition: Deploy a biomimetic sensor network and computer vision monitoring unit to monitor the charging pile in real time and collect data including physicochemical parameters and image information;

[0013] Step 2: Data Preprocessing: Preprocess the acquired sensor data and image data; for sensor data, perform filtering and noise reduction operations; for image data, perform grayscale conversion, cropping, and scaling preprocessing operations.

[0014] Step 3: Fault Diagnosis: Input the collected data into the fault diagnosis model, and use swarm intelligence algorithms and collaborative sub-models to analyze the data, determine whether a fault exists and identify the type of fault.

[0015] Step 4: Fault Diagnosis Result Output and Decision-Making: Based on the fused fault diagnosis results, determine whether the charging pile has a fault and the specific type of fault. If the result indicates the presence of a certain fault, issue an alarm in a timely manner and take corresponding measures. At the same time, record the fault information.

[0016] Step 5: Maintenance Strategy Generation and Optimization: Based on the fault diagnosis results, start the genetic programming evolution maintenance strategy module to generate an initial maintenance strategy and optimize it.

[0017] Optionally, the bionic sensor network includes temperature sensors, liquid pressure sensors, gas sensors, chemical sensors, pressure sensors, and vibration sensors installed on the charging pile. The sensors transmit the collected data to the central processing system via a wireless communication network.

[0018] The chemical sensor uses nanomaterials and conductive polymers as sensing elements; the pressure sensor uses piezoelectric materials as sensing elements, and the vibration sensor converts the vibration of the outer shell into an electrical signal output.

[0019] Optionally, the biomimetic sensor network also includes the biomimetic sensor network construction method and information transmission and collaborative working mechanism:

[0020] Biomimetic sensor network construction method: Construct a self-organizing topology, where each sensor node has the function of receiving, processing, and transmitting information; the nodes are interconnected locally to form a network structure;

[0021] Information transmission and collaborative working mechanism: Sensor nodes exchange information through local communication protocols; when a node detects local information, it transmits the information to its connected neighboring nodes; the receiving node integrates and processes its own local information and the received information, and then transmits it to several nodes.

[0022] Optionally, the computer vision monitoring unit includes a camera and a hierarchical convolutional neural network image processing unit; the feature extraction process of the hierarchical convolutional neural network image processing unit is as follows:

[0023] Images of the charging station and its surrounding environment are input into a low-level convolutional neural network. Through the operation of the low-level convolutional kernels, basic feature maps of edges and textures are obtained. Then, the low-level feature maps are fed into a high-level convolutional neural network. The high-level convolutional kernels process these feature maps, extract abstract features, and through multiple convolutional operations, feature vectors that can represent key information in the image are obtained.

[0024] Optionally, the computer vision monitoring unit further includes a self-attention mechanism, which operates as follows:

[0025] By calculating the attention scores between different locations in the image, we can determine which regions need more attention. For each location in the image, we calculate its attention weight relative to other locations using a similarity function divided by a scaling factor. Then, we normalize the attention weights using a softmax function to obtain the final attention weight matrix.

[0026] Optionally, the fault diagnosis model includes a swarm intelligence algorithm model and a division of labor and cooperation sub-model; the swarm intelligence algorithm model is a particle swarm optimization algorithm.

[0027] The particle swarm optimization-assisted diagnostic methods are as follows:

[0028] Particle encoding and initialization: Each possible failure mode is encoded as a position vector of a particle; each dimension corresponds to a failure mode, with a value of 0 or 1, indicating whether the failure mode has occurred; multiple particles are randomly generated in the solution space as the initial population;

[0029] Fitness function construction: A fitness function is constructed based on the goal of fault diagnosis; the fitness function is evaluated based on the degree of matching between the fault mode and sensor data and image data.

[0030] Optionally, the simulation method of the division of labor and cooperation sub-model is as follows:

[0031] Sub-model division: Based on the fault type and data processing characteristics, the fault diagnosis task is divided into different sub-models. Fault diagnosis based on sensor data is one sub-model, and fault diagnosis based on image data is another sub-model. Each sub-model is responsible for analyzing a specific type of data and making a preliminary judgment on the fault.

[0032] Collaboration and Fusion Mechanism: After the two sub-models obtain preliminary fault diagnosis results, they collaborate and merge through the fusion module. The fusion module uses a weighted fusion method to comprehensively analyze the results of the two sub-models.

[0033] Optionally, the genetically programmed evolutionary maintenance strategy module operates as follows:

[0034] Input data collection and processing: Collect information related to fault diagnosis results, including fault type, charging pile equipment characteristics, and historical maintenance experience; clean, process, and encode the collected data;

[0035] Classification and coding: The above data is classified and coded so that it can be used as input for genetic programming;

[0036] Initial maintenance strategy generation: A set of initial maintenance strategies is randomly generated. The maintenance strategy is represented in the form of gene coding, and each gene represents a specific maintenance procedure or operation step.

[0037] Fitness assessment: A fitness function is established based on actual maintenance feedback; the fitness function is based on a comprehensive evaluation of multiple factors.

[0038] Optionally, the genetic programming evolutionary maintenance strategy module also includes genetic operations and evolution, as follows:

[0039] Selection operation: Calculate the fitness value of each individual based on the fitness function, use roulette wheel selection to calculate the relative fitness of each individual, and then assign the probability of each individual being selected based on the relative fitness value;

[0040] Crossover operation: Two individuals are randomly selected from the population after the selection operation as parents, and crossover operation is performed according to the pre-set crossover probability; using the single-point crossover method, a crossover point is randomly selected, and some genes of the two parent individuals before and after the point are exchanged to generate two new offspring individuals;

[0041] Mutation operation: For each individual in the population after the crossover operation, a mutation operation is performed according to a pre-set mutation probability; the gene position inversion method is used to randomly select a gene position of an individual and invert its value;

[0042] Iterative evolution and optimization: The new generation of population after selection, crossover and mutation operations is taken as the current population, and the fitness evaluation, selection, crossover and mutation operations are repeated; the individual with the largest fitness value in the final population is selected as the optimal solution of the maintenance strategy.

[0043] An artificial intelligence-based remote monitoring and early warning system for charging piles includes:

[0044] Bionic sensor network: It consists of various sensors distributed inside, on top, bottom and side of the charging pile, including temperature sensor, liquid pressure sensor, gas sensor, chemical sensor, pressure sensor and vibration sensor. It is connected to the central processing system through wireless communication module to form a distributed data acquisition network.

[0045] Computer vision monitoring unit: includes a high-definition camera and an image acquisition card. The camera is responsible for capturing images of the charging pile, and the image acquisition card converts analog signals into digital signals and transmits them to the central processing system.

[0046] Data acquisition and transmission module: including sensor driver, data acquisition program and wireless communication program, responsible for controlling the sensor and camera to acquire data, and transmitting the acquired data to the central processing system through the network;

[0047] Central processing system: including storage devices and processors, used to receive, store and process sensor data and image data, and run fault diagnosis models, maintenance strategies and algorithms;

[0048] Early warning unit: Based on the results of the central processing system, when an abnormality or fault is detected, it issues an early warning signal;

[0049] Control unit: In an emergency, automatically controls the charging station to stop working;

[0050] Human-machine interface: used to display fault diagnosis results, alarm information and system operating status; users configure and manage the system through the human-machine interface.

[0051] (III) Beneficial Effects

[0052] This invention provides a remote monitoring and early warning method and system for charging piles based on artificial intelligence, which has the following beneficial effects:

[0053] 1. This invention, by deploying a biomimetic sensor network, enables multi-dimensional real-time monitoring of charging piles. These sensors can collect physicochemical parameters, allowing the system to gain a more comprehensive understanding of the charging pile's status and improve the accuracy of fault detection. The chemical sensors in the biomimetic sensor network can simulate the function of biological olfactory receptor cells, accurately detecting the characteristic gases produced by the thermal decomposition of insulating materials caused by local overheating due to electrical faults. This sensitive detection mechanism allows the system to issue early warnings at the early stages of electrical faults.

[0054] 2. The self-attention mechanism of this invention determines which areas need more attention by calculating the attention scores between various positions in the image. The system can adaptively focus on key areas in the image, thereby improving the ability to capture key information. Based on the visual saliency model, the saliency value of each position in the image is calculated according to the color, brightness and texture features of the image. The system can more accurately identify the status of the indicator lights on the charging pile, thereby improving the accuracy and reliability of the entire monitoring system.

[0055] 3. This invention transforms the fault diagnosis problem into a problem of finding the optimal particle position in the solution space through particle swarm optimization algorithm. By sharing information and cooperating among particles, it quickly converges to the optimal fault diagnosis result, thereby improving the accuracy of fault diagnosis. By using neural networks and convolutional neural networks to process sensor data and image data, it can accurately identify and classify fault features, further improving the accuracy of diagnosis.

[0056] 4. This invention generates optimal maintenance strategies through genetic programming evolution, which can significantly shorten maintenance time and improve maintenance efficiency. The genetic programming evolution method can adapt to different fault types and equipment characteristics, generate targeted maintenance strategies, improve the system's adaptability and flexibility, and by collecting and analyzing historical maintenance data, the maintenance strategy module can optimize based on practical experience, providing data-driven decision support for future maintenance work. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the artificial intelligence-based remote monitoring and early warning method for charging piles according to the present invention.

[0058] Figure 2 This is a schematic diagram of the modules of the artificial intelligence-based remote monitoring and early warning system for charging piles according to the present invention. Detailed Implementation

[0059] 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.

[0060] Example 1

[0061] Please see Figure 1 A remote monitoring and early warning method for charging piles based on artificial intelligence includes the following steps:

[0062] Step 1: Sensor Data Acquisition: Deploy a biomimetic sensor network and computer vision monitoring unit to monitor the charging pile in real time and collect data including physicochemical parameters and image information;

[0063] Step 2: Data Preprocessing: Preprocess the collected sensor data and image data. For sensor data, filtering and denoising operations are performed to improve data quality. For example, median filtering is used to remove noise points from temperature data. For image data, grayscale conversion, cropping, and scaling are performed to facilitate subsequent analysis. For example, color images are converted to grayscale images, the image portion containing the indicator light area is cropped, and the image size is standardized to 64×64 pixels. In the image preprocessing stage, the image needs to be segmented to separate the charging pile area from the background. The edge detection-based image segmentation method uses the Canny operator to detect edges in the image. The calculation formula of the Canny operator involves Gaussian filtering, gradient calculation, and non-maximum suppression steps. First, a Gaussian filter is used to smooth the image to reduce the influence of noise. Then, the gradient magnitude and direction of the image are calculated. Finally, non-maximum suppression is used to preserve important edge information in the image, resulting in the segmented image.

[0064] Step 3: Fault Diagnosis: Input the collected data into the fault diagnosis model, and use swarm intelligence algorithms and collaborative sub-models to analyze the data, determine whether a fault exists and identify the type of fault.

[0065] Step 4: Fault Diagnosis Result Output and Decision-Making: Based on the integrated fault diagnosis results, determine whether the charging pile has a fault and the specific type of fault. If the result indicates that a certain fault exists (such as overheating), issue an alarm in a timely manner and take corresponding measures (such as stopping charging and starting the heat dissipation system). At the same time, record the fault information for subsequent maintenance and repair.

[0066] Step 5: Maintenance Strategy Generation and Optimization: Based on the fault diagnosis results, activate the genetic programming evolution maintenance strategy module to generate an initial maintenance strategy, and then evaluate and optimize it.

[0067] The biomimetic sensor network includes current sensors, voltage sensors, temperature sensors, liquid / gas pressure sensors, gas sensors, chemical sensors, pressure sensors, and vibration sensors installed on the charging pile. The sensors transmit the collected data to the central processing system via a wireless communication network. The current and voltage sensors monitor the output current and voltage data of the charging pile during charging. The temperature sensor collects temperature data every minute. The liquid / gas pressure sensor monitors the internal liquid or gas pressure in real time. The gas sensor detects whether there is a flammable gas leak.

[0068] Chemical Sensor (Simulated Olfactory Receptor Cells): Biological olfactory receptor cells bind to odor molecules through specific receptor proteins, triggering a series of biochemical reactions within the cell and generating electrical signals, thereby enabling the perception of odors. Chemical sensors utilize materials with highly specific recognition capabilities as sensing elements, such as nanomaterials and conductive polymers. When characteristic gases (such as formaldehyde and benzene volatile organic compounds) generated by the thermal decomposition of insulating materials due to localized overheating caused by electrical faults come into contact with the sensing element, these gases react chemically or are physically adsorbed onto specific receptors on the surface of the sensing element, causing changes in the electrical properties (such as resistance and capacitance) of the sensing element. The detection circuit monitors these changes in the electrical properties of the sensing element in real time and converts them into corresponding electrical signals, thereby achieving accurate detection of the characteristic gases and sensitive capture of early signs of electrical faults in charging piles.

[0069] Pressure and vibration sensors (refer to tactile neurons): Tactile neurons sense pressure and vibration mechanical stimuli by sensing mechanosensitive ion channels on their cell membranes, converting mechanical signals into electrical signals that are transmitted along nerve fibers. Mimicking this mechanism, pressure sensors use piezoelectric materials as sensing elements. When the charging pile casing undergoes minor deformation due to internal mechanical loosening or external impact, the piezoelectric material generates an electric charge, causing a change in the resistance of the strain gauge. These changes are detected by a measuring circuit and converted into electrical signals. Vibration sensors utilize the vibration principle of a mass, spring, and damping system. When the casing vibrates, the mass block inside the sensor moves relative to the spring. By measuring the displacement or acceleration of the mass block, it is converted into an electrical signal output, thus sensitively capturing minute deformation and vibration signals.

[0070] Bionic sensor networks also include bionic sensor network construction methods and information transmission and collaborative working mechanisms:

[0071] Biomimetic sensor network construction method: mimicking the connection patterns in biological neural networks, a self-organizing topology is constructed. Each sensor node is similar to a neuron, with the functions of receiving, processing, and sending information. The nodes are interconnected locally to form a network structure. This connection has a certain degree of sparsity and randomness, while following certain rules to ensure efficient information transmission and collaborative work. For example, a scale-free network model can be used to construct the topology of the sensor array.

[0072] Information transmission and collaborative working mechanism: Sensor nodes interact with each other through local communication protocols. When a node detects local information (such as temperature or pressure changes), it transmits this information to its neighboring nodes. The receiving node integrates and processes the information based on its own local information and the received information, and then transmits it to several nodes. This information transmission method is similar to action potential conduction in neural networks. Through the collaborative work of multiple nodes, a comprehensive perception of the overall status of the charging pile is achieved. For example, when a sensor in a certain area detects an abnormal increase in temperature, it transmits this information to the surrounding sensor nodes. These nodes, based on their location and functional characteristics, further determine whether the entire charging pile is overheating or if it is only a local fault, thereby achieving an accurate assessment of the charging pile's status.

[0073] In this embodiment, by deploying a biomimetic sensor network, the charging pile can be monitored in real time from multiple dimensions. These sensors can collect various data including physicochemical parameters (such as temperature, pressure, and gas). This multi-dimensional monitoring method enables the system to have a more comprehensive understanding of the charging pile's status and improves the accuracy of fault detection. The chemical sensors in the biomimetic sensor network can simulate the function of biological olfactory receptor cells and accurately detect the characteristic gases produced by the thermal decomposition of insulating materials caused by local overheating due to electrical faults. This sensitive capture mechanism enables the system to issue early warnings in the early stages of electrical faults, thereby preventing further deterioration of the fault.

[0074] The biomimetic sensor network constructs a self-organizing topology. Each sensor node is similar to a neuron, with the functions of receiving, processing, and sending information. The nodes are interconnected locally to form a network structure, ensuring efficient information transmission and collaborative work. This information transmission method is similar to the action potential conduction in a neural network. Through the collaborative work of multiple nodes, a comprehensive perception of the overall status of the charging pile can be achieved.

[0075] The construction method and information transmission and collaborative working mechanism of the biomimetic sensor network enable the system to maintain high reliability when facing complex and ever-changing environments. Even if some nodes fail or malfunction, other nodes can still continue to work, ensuring the normal operation of the system. Through the real-time monitoring and early fault warning functions of the biomimetic sensor network, the system can promptly detect and handle potential problems, thereby reducing maintenance costs and downtime caused by faults. This is of great significance for improving the service life of charging piles and reducing operating costs.

[0076] Example 2

[0077] This embodiment is a further optimization based on Embodiment 1. Specifically, the computer vision monitoring unit includes a camera and a hierarchical convolutional neural network image processing unit. The camera's shooting angle and frame rate are set, and the camera acquires images of the charging pile's appearance and indicator light status. After acquiring the images, image frames are captured as samples at preset time intervals. The image data reflects whether the charging pile's outer shell is damaged and whether the indicator light's color and status are normal. For example, the high-definition camera captures a set of images of the charging pile at regular intervals (e.g., 5 minutes), including the appearance from different angles and close-ups of the indicator lights. The image data is preprocessed, such as cropping, scaling, and grayscale operations. The image size is unified to a fixed size required for model input, such as 224×224 pixels.

[0078] The feature extraction process of the hierarchical convolutional neural network image processing unit is as follows:

[0079] Images of charging stations and their surrounding environment are input into a low-level convolutional neural network. Through the operations of the low-level convolutional kernels, basic feature maps of edges and textures are obtained. Then, the low-level feature maps are fed into a high-level convolutional neural network. The high-level convolutional kernels process these feature maps and extract abstract features, such as the outline of the charging station and the shape of the indicator lights. Through multi-layer convolution operations, feature vectors that can represent key information in the image are obtained, thereby realizing the effective recognition and analysis of images of charging stations and their surrounding environment.

[0080] Low-level convolutional kernels (simulating the receptive fields of simple cells): Low-level convolutional kernels are small in size and are responsible for extracting local features in the image, such as edges and basic texture elements; for example, 3×3 or 5×5 convolutional kernels are used, and their parameters are initialized with reference to some pre-trained models or randomly initialized and then optimized through training; these convolutional kernels slide on the image and calculate the weighted sum of pixels in the local region, just like the receptive fields of simple cells in the retina perform preliminary processing of visual information;

[0081] High-level convolutional kernels (simulating the comprehensive processing capabilities of complex cells): High-level convolutional kernels are built on the foundation of low-level feature maps and are larger in size. They are used to further extract more abstract shape and object contour features. They integrate information from different local features by performing convolution operations on multiple low-level feature maps. For example, a 5×5 or larger convolutional kernel can be used, and its parameters are also determined through training. The mathematical formula for high-level convolution operations is similar to that of low-level operations, except that the input data is a collection of multiple low-level feature maps.

[0082] The computer vision monitoring unit also includes a self-attention mechanism, which operates as follows:

[0083] By calculating attention scores between different locations in the image, we can determine which regions require more attention; for each location i in the image, we calculate its attention weight e with respect to other locations j. ij The calculation formula uses a similarity function (such as dot product or cosine similarity) divided by a scaling factor (such as d). k The square root of d k (These are the feature dimensions of positions i and j); then the attention weights are normalized using the softmax function to obtain the final attention weight matrix A; in application, the attention weight matrix A is multiplied by the original image feature matrix X to obtain the feature matrix Y after focusing on the key region, i.e., Y = AX; in this way, the system adaptively focuses on the key regions in the image, such as the area near the charging port or the indicator light display area.

[0084] The method based on the visual saliency model: The visual saliency model calculates the saliency value of each position in the image based on the color, brightness and texture features of the image; the higher the saliency value, the easier it is to attract the attention of the system; for example, the Itti model is used to calculate the saliency value; after obtaining the saliency value matrix S, the saliency value is also processed by the softmax function to obtain the attention weight matrix, thereby realizing the focus on key areas.

[0085] In this embodiment, by setting up a camera and a hierarchical convolutional neural network (CNN) image processing unit, the system can perform multi-dimensional real-time image acquisition and processing of the appearance and indicator status of the charging pile. This multi-dimensional monitoring method enables the system to have a more comprehensive understanding of the charging pile's status and improve the accuracy of fault detection. The hierarchical convolutional neural network extracts basic edge and texture features and more abstract shape and object contour features from the image through the operation of low-level and high-level convolutional kernels. After these features are processed through multiple convolution operations, they form feature vectors that can represent key information in the image, thereby achieving effective recognition and analysis of images of the charging pile and its surrounding environment.

[0086] The self-attention mechanism calculates attention scores between different locations in an image to determine which areas require more attention. The system can adaptively focus on key areas in the image, such as the area near the charging port and the indicator light display area, thereby improving the ability to capture key information. Based on the visual saliency model, the system calculates the saliency value of each location in the image according to the color, brightness, and texture features. The system can more accurately identify the status of indicator lights on the charging pile (e.g., red indicates a fault, green indicates normal), whether there is damage to the equipment appearance (e.g., cracked or deformed casing), and whether there are foreign objects blocking the charging port. Through the application of the self-attention mechanism and the visual saliency model, the system can more accurately identify and analyze the status of the charging pile, improving the accuracy and reliability of the entire monitoring system.

[0087] The camera captures image frames as samples at pre-set time intervals. Combined with a hierarchical convolutional neural network and a self-attention mechanism, it enables real-time monitoring and early warning of charging piles, allowing for timely detection and handling of potential problems.

[0088] Example 3

[0089] This embodiment is a further optimization based on Embodiments 1 and 2. Specifically, the fault diagnosis model includes a swarm intelligence algorithm model and a division of labor and cooperation sub-model. The swarm intelligence algorithm model is a particle swarm optimization algorithm, which transforms the fault diagnosis problem into a problem of finding the optimal particle position in the solution space. Each particle represents a possible fault mode. Through information sharing and cooperation among particles, the optimal fault diagnosis result is quickly converged.

[0090] The following are methods for assisting in diagnostics using particle swarm optimization (PSO):

[0091] Particle encoding and initialization: Each possible fault mode is encoded as a particle position vector; for example, for a system with 3 possible faults (overheating, overvoltage, undervoltage), a 3D vector is used to represent the particle position, each dimension corresponds to a fault mode, and the value is 0 or 1, indicating whether the fault mode has occurred; multiple particles are randomly generated in the solution space as the initial population;

[0092] Fitness function construction: A fitness function is constructed based on the goal of fault diagnosis. The fitness function is evaluated based on the degree of matching between the fault mode and sensor data and image data. For example, the fitness function is designed as the reciprocal of the error function between the fault mode and the actual observed data. If the error between the predicted sensor data and the actual collected data is small under a certain fault mode, and the image features are consistent with the typical image features under that fault mode, then the fitness value is high.

[0093] Division of labor and cooperation sub-model: Simulates the division of labor and cooperation mode in biological populations, assigns different fault diagnosis tasks to different sub-models or algorithm modules, and the modules work together to complete complex fault diagnosis tasks.

[0094] The simulation method for the division of labor and cooperation sub-model is as follows:

[0095] Sub-model division: Based on the fault type and data processing characteristics, the fault diagnosis task is divided into different sub-models. Fault diagnosis based on sensor data is one sub-model, and fault diagnosis based on image data is another sub-model. Each sub-model is responsible for analyzing a specific type of data and making a preliminary judgment on the fault.

[0096] The sensor data-based sub-model utilizes a neural network algorithm to process sensor data. Taking a feedforward neural network as an example, the number of nodes in the input layer equals the number of sensor data types (e.g., voltage, current, temperature, pressure, gas concentration). One or more hidden layers are set as needed, and the number of nodes in the output layer equals the number of possible fault modes (e.g., overheating, overvoltage, undervoltage). The input-output relationship of the neural network is y = f(wx + b), where y is the output vector (fault mode probability distribution), f is the activation function (e.g., the sigmoid function), w is the weight matrix, x is the input vector (sensor data, including temperature, pressure, gas concentration, etc.), and b is the bias vector. By training the neural network, it can identify and classify fault modes based on sensor data.

[0097] Image-based sub-model: Convolutional Neural Networks (CNNs) are used to analyze image data. A CNN consists of convolutional layers, pooling layers, and fully connected layers. For fault diagnosis of charging pile images, convolutional layers are first used to extract image features (such as edges and textures), then pooling layers are used to reduce the feature dimensionality, and finally fully connected layers are used to map the features to the fault mode space. For example, a simple CNN architecture includes two convolutional layers (each with a kernel size of 3×3, a stride of 1, and 16 and 32 channels respectively), one pooling layer (with a pooling window size of 2×2 and a stride of 2), and one fully connected layer (with the number of neurons equal to the number of fault modes). The training process of a CNN is to adjust the network parameters through the backpropagation algorithm so that the network can accurately classify fault features in the image.

[0098] Collaboration and Fusion Mechanism: After the two sub-models obtain preliminary fault diagnosis results, they collaborate and fuse through the fusion module. The fusion module uses a weighted fusion method to comprehensively analyze the results of the two sub-models. For example, different weights are assigned to the results of each sub-model based on the accuracy and reliability of the sub-model. Assuming that the result of the sub-model based on sensor data is R1 and the result of the sub-model based on image data is R2, the final result after fusion is R = w1R1 + w2R2, where w1 and w2 are the weight coefficients of the two sub-models, and w1 + w2 = 1.

[0099] In this embodiment, the Particle Swarm Optimization (PSO) algorithm transforms the fault diagnosis problem into finding the optimal particle position in the solution space. Each particle represents a possible fault mode. Through information sharing and cooperation among particles, the algorithm quickly converges to the optimal fault diagnosis result, thereby improving the accuracy of fault diagnosis. By using neural networks and convolutional neural networks (CNNs) to process sensor data and image data, fault features can be accurately identified and classified, further improving the accuracy of diagnosis. The PSO algorithm continuously updates the velocity and position of particles, causing them to gradually converge to the optimal position, i.e., the most likely combination of fault modes, thereby accelerating the speed of fault diagnosis.

[0100] The collaborative sub-model assigns different fault diagnosis tasks to different sub-models or algorithm modules. The modules work together to complete complex fault diagnosis tasks, improving the system's response speed. By constructing a fitness function, the matching degree between fault modes and sensor data and image data is considered, ensuring the reliability of the diagnosis results. The fusion module uses a weighted fusion method to comprehensively analyze the results of the two sub-models. Different weights are assigned to the results of each sub-model according to their accuracy and reliability, further enhancing the system's stability.

[0101] The entire fault diagnosis process is completed automatically by artificial intelligence algorithms, reducing human intervention and realizing intelligent and automated fault diagnosis. The system can automatically adjust the diagnosis strategy based on real-time data, improving the system's adaptability. Accurate fault diagnosis and fast response speed can reduce equipment downtime and lower maintenance costs.

[0102] Example 4

[0103] This embodiment is a further optimization based on embodiment 3. Specifically, the working method of the genetic programming evolution maintenance strategy module is as follows:

[0104] Input data collection and processing: The data acquisition module interfaces with the charging pile's monitoring system and maintenance record database to obtain real-time and historical data. It collects information related to fault diagnosis results, including fault type (e.g., overheating, overvoltage, undervoltage, component damage), charging pile equipment characteristics (model, specifications, service life, parameters of each component), and historical maintenance experience (repair methods, success rates, repair time, and costs for similar faults in the past). The collected data is then cleaned, processed, and encoded. For example, noise and errors are removed, and different data types are converted into formats suitable for genetic programming input. This is achieved using data cleaning algorithms, encoding rule tools, and techniques.

[0105] Classification and coding: The above data is classified and coded to serve as input for genetic programming; numerical and binary coding methods are used, for example, different numerical values ​​are used to represent different fault types, and equipment characteristic parameters are represented by binary strings;

[0106] Initial maintenance strategy generation: A set of initial maintenance strategies is randomly generated. The maintenance strategy is represented in the form of gene coding, with each gene representing a specific maintenance procedure or operation step; for example, a gene sequence can be used to represent a maintenance process such as "check the temperature sensor - replace the temperature sensor - recalibrate the temperature system".

[0107] Set the initial population size, such as N individuals (maintenance strategy), and determine the size of N based on the complexity of the problem and computer resource factors;

[0108] Fitness assessment: A fitness function is established based on actual repair performance feedback; the fitness function is based on a comprehensive evaluation of multiple factors, such as repair time (T), repair success rate (S), and cost (C);

[0109] The fitness function formula is: Among them, fitness is the fitness value, and w1, w2, and w3 are weighting coefficients that reflect the importance of different factors in the evaluation.

[0110] For example, if a repair strategy can complete the repair within 1 hour (T=1), achieve a 90% success rate (S=0.9), and has a cost of 100 yuan (C=100), and the weighting coefficients are w1=0.4, w2=0.3, and w3=0.3 respectively, then its fitness is: ×0.9-0.3×100.

[0111] The genetic programming evolutionary maintenance strategy module also includes genetic operations and evolution, as follows:

[0112] Selection operation: Calculate the fitness value of each individual (maintenance strategy) based on the fitness function. Use roulette wheel selection to calculate the relative fitness of each individual (i.e., the ratio of individual fitness to the total fitness of the population). Then, assign the probability of each individual being selected based on the relative fitness. For example, the relative fitness of individual i is... By generating random numbers between [0,1], individuals are selected to enter the next generation according to probability; data input layer p i It is the relative fitness of individual i; It is the overall fitness of the population. i It is the fitness value of individual i;

[0113] Crossover operation: Randomly select two individuals from the population after the selection operation as parents, and crossover according to a pre-set crossover probability (e.g., P). c =0.8) to perform crossover operation; using the single-point crossover method, a crossover point is randomly selected, and some genes of the two parent individuals before and after that point are exchanged to generate two new offspring individuals;

[0114] Mutation operation: For each individual in the population after the crossover operation, a mutation probability (e.g., P) is generated according to a pre-set mutation probability. m =0.05) to perform mutation operation; use the gene position inversion method to randomly select a gene position of an individual and invert its value (i.e. 0 becomes 1, 1 becomes 0);

[0115] Iterative Evolution and Optimization: The new generation of population after selection, crossover, and mutation operations is used as the current population, and the fitness evaluation, selection, crossover, and mutation operations are repeated; an iteration termination condition is set, such as reaching the maximum number of iterations (e.g., 1000 times) or the fitness value no longer showing a significant improvement (the fitness value change is less than a certain threshold for multiple consecutive generations); when the termination condition is met, the evolution process stops; the individual with the largest fitness value is selected from the final population as the optimal solution for the maintenance strategy.

[0116] In this embodiment, the optimal maintenance strategy is generated through genetic programming evolution, which can significantly shorten maintenance time and improve maintenance efficiency. For example, after multiple generations of evolution, individuals in the initial population can find a better maintenance process, thereby reducing unnecessary steps and time in actual operation. The fitness function considers multiple factors, such as maintenance time, repair success rate, and cost, so that the final selected maintenance strategy performs better in all aspects, thereby improving maintenance quality. The genetic programming evolution method can adapt to different fault types and equipment characteristics, generating targeted maintenance strategies, improving the system's adaptability and flexibility. By collecting and analyzing historical maintenance data, the maintenance strategy module can optimize based on practical experience, providing data-driven decision support for future maintenance work. Through rapid iteration and optimization, the maintenance strategy module can quickly generate effective maintenance solutions after a fault occurs, improving the system's rapid response capability.

[0117] By optimizing maintenance strategies, the materials and human resources required for maintenance can be effectively reduced, thereby lowering the overall maintenance cost. For example, some inefficient or high-cost maintenance operations can be gradually eliminated through an evolutionary process. By continuously iterating and optimizing maintenance strategies, the stability and reliability of charging piles after repair can be ensured, reducing the possibility of recurrence of failures.

[0118] The entire maintenance strategy generation and optimization process is completed automatically by artificial intelligence algorithms, reducing human intervention and realizing the intelligence and automation of maintenance strategies. The automated maintenance strategy generation and optimization process can reduce human error and improve the accuracy and consistency of maintenance.

[0119] Please see Figure 2 The present invention also discloses an artificial intelligence-based remote monitoring and early warning system for charging piles, comprising:

[0120] Bionic sensor network: It consists of various sensors distributed inside, on top, bottom and side of the charging pile, including temperature sensor, liquid pressure sensor, gas sensor, chemical sensor, pressure sensor and vibration sensor. It is connected to the central processing system through wireless communication module to form a distributed data acquisition network.

[0121] Computer vision monitoring unit: includes a high-definition camera and an image acquisition card. The camera is responsible for capturing images of the charging pile, and the image acquisition card converts analog signals into digital signals and transmits them to the central processing system.

[0122] Data acquisition and transmission module: including sensor driver, data acquisition program and wireless communication program, responsible for controlling the sensor and camera to acquire data, and transmitting the acquired data to the central processing system through the network;

[0123] Central processing system: including storage devices and processors, used to receive, store and process sensor data and image data (performing preprocessing operations such as cleaning, filtering and normalization of the acquired data to improve data quality and provide accurate data input for subsequent analysis and diagnosis), and run fault diagnosis models, maintenance strategies and algorithms (including swarm intelligence algorithms, division of labor and cooperation sub-models, neural network sub-models, CNN sub-models and fusion decision algorithms. This module is the core part of fault diagnosis and is responsible for fault mode recognition and diagnosis based on the input data).

[0124] Early warning unit: Based on the results of the central processing system, when an abnormality or fault is detected, an early warning signal is issued. The early warning signal is transmitted to relevant personnel through audible and visual alarms, SMS push, and APP notification.

[0125] Control unit: In emergency situations, such as detecting a serious electrical fault or safety hazard, it automatically controls the charging station to stop working to ensure safety;

[0126] Human-machine interface: Provides users with an operating interface to display fault diagnosis results, alarm information and system operating status; users can configure and manage the system through the human-machine interface.

[0127] In summary, the bionic sensor network and computer vision monitoring unit of this invention can collect multi-dimensional data such as temperature, liquid and gas pressure, gas, chemical, pressure, and vibration of the charging pile in real time, as well as high-definition images, providing rich data support for fault diagnosis. The redundant design of the bionic sensor network and computer vision monitoring unit improves the reliability of the system. Even if one sensor fails, other sensors can continue to work, ensuring the continuity of data acquisition.

[0128] The fault diagnosis model running in the central processing system can perform in-depth analysis and processing of the collected data, accurately identify and classify fault modes, and improve the accuracy of fault diagnosis. The central processing system processes and analyzes the data in real time. When an abnormality or fault is detected, the early warning unit can quickly issue an early warning signal, and the control unit can also automatically control the charging pile to stop working in an emergency, thus improving the system's response speed.

[0129] The entire system is driven by artificial intelligence algorithms, automating everything from data acquisition, transmission, and processing to fault diagnosis and maintenance strategy generation. This reduces human intervention and improves the system's intelligence level. Accurate fault diagnosis and rapid response speed can reduce equipment downtime and lower maintenance costs. In emergency situations, the control unit can automatically stop the charging pile from working, effectively preventing safety accidents.

[0130] Users can configure and manage the system through the human-computer interaction interface, making the system more flexible and convenient. Through the human-computer interaction interface, users can understand the operating status and fault information of the charging pile in real time. Users can check the operating status and fault information of the charging pile anytime and anywhere, which improves user satisfaction.

[0131] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are 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. An artificial intelligence-based charging pile remote monitoring and early warning method, characterized in that, The method comprises the following steps: Step 1: sensor data collection: deploy a bionic sensor network and a computer vision monitoring unit to monitor the charging pile in real time, and collect physical and chemical parameters and image information; Step 2: data preprocessing: preprocessing the collected sensor data and image data; for sensor data, filtering and denoising operations are performed; for image data, grayscale, cropping, and scaling preprocessing operations are performed; Step 3: fault diagnosis: input the collected data into the fault diagnosis model, and use the swarm intelligence algorithm and the division and cooperation sub-model for analysis to determine whether there is a fault and the fault type; the fault diagnosis model comprises a swarm intelligence algorithm model and a division and cooperation sub-model; the swarm intelligence algorithm model is a particle swarm optimization algorithm; the particle swarm optimization auxiliary diagnosis method is as follows: particle coding and initialization: encode each possible fault mode as a particle position vector; each dimension corresponds to a fault mode, and the value is 0 or 1, indicating whether the fault mode occurs; a plurality of particles are randomly generated in the solution space as the initial population; fitness function construction: construct the fitness function according to the fault diagnosis target; the fitness function is evaluated based on the matching degree of fault modes and sensor data and image data; particle updating and iteration: update the velocity and position of the particle according to the velocity and position updating formula of the particle swarm optimization algorithm; the division and cooperation sub-model simulation method is as follows: sub-model division: according to the fault type and data processing characteristics, the fault diagnosis task is divided into different sub-models, the fault diagnosis based on sensor data is taken as a sub-model, and the fault diagnosis based on image data is taken as another sub-model, each sub-model is responsible for analyzing a specific type of data and preliminarily judging the fault; sensor data-based sub-model: this sub-model uses a neural network algorithm to process sensor data, where the number of input layer nodes is equal to the number of sensor data types, the hidden layer is set according to actual needs, and the number of output layer nodes is consistent with the number of possible fault modes; through training the neural network, the network can identify and classify fault modes according to sensor data; image data-based sub-model: a convolutional neural network is used to analyze image data; CNN is composed of convolutional layers, pooling layers, and fully connected layers; for fault diagnosis of charging pile images, the convolutional layer is used to extract image features, then the pooling layer is used to reduce feature dimension, and finally the fully connected layer is used to map features to fault mode space; collaboration and fusion mechanism: after the two sub-models obtain preliminary fault diagnosis results, the fusion module is used for collaboration and fusion; the fusion module uses a weighted fusion method to comprehensively analyze the results of the two sub-models; Step 4: fault diagnosis result output and decision: according to the fused fault diagnosis result, it is judged whether the charging pile has a fault and the specific fault type; if the result indicates that there is a fault, an alarm is sent in time and corresponding measures are taken, and fault information is recorded. Step five: maintenance strategy generation and optimization: according to the fault diagnosis result, start the genetic programming evolution maintenance strategy module, generate the initial maintenance strategy, and optimize it.

2. The method of claim 1, wherein the method is based on artificial intelligence. The bionic sensor network includes temperature sensors, liquid and gas pressure sensors, gas sensors, chemical sensors, pressure sensors and vibration sensors installed on the charging pile. The sensors transmit the collected data to the central processing system through a wireless communication network. The chemical sensor uses nanomaterials and conductive polymers as sensing elements. When the characteristic gas generated by the thermal decomposition of insulating materials due to local overheating caused by electrical faults contacts the sensing element, the electrical properties of the sensing element change. The change in the electrical properties of the sensing element is monitored in real time by the detection circuit and converted into a corresponding electrical signal output. The pressure sensor uses piezoelectric materials as sensing elements. When the charging pile shell deforms slightly due to internal mechanical loosening or external impact, the resistance value of the strain gauge changes. These changes are detected by the measurement circuit and converted into an electrical signal. When the shell vibrates, the mass inside the vibration sensor moves relative to the spring. The displacement or acceleration of the mass is converted into an electrical signal output.

3. The method of claim 2, wherein the method further comprises: The bionic sensor network also includes a bionic sensor network construction method and an information transmission and cooperative working mechanism: The bionic sensor network construction method is to build a self-organizing topology, and each sensor node has the functions of receiving, processing and transmitting information. The nodes are locally connected to form a network structure. The information transmission and cooperative working mechanism is that the sensor nodes interact with each other through local communication protocols. When a node detects local information, it transmits the information to the adjacent nodes connected to it. The receiving nodes integrate and process the local information and the received information, and then transmit them to several nodes.

4. The method of claim 1, wherein the method further comprises: The computer vision monitoring unit includes a camera and a layered convolutional neural network image processing unit. The camera collects images of the appearance of the charging pile and the status of the indicator lights. The image data is preprocessed to unify the image size to the fixed size required by the model input. The feature extraction process of the layered convolutional neural network image processing unit is as follows: The image of the charging pile and its surrounding environment is input into the low-level convolutional neural network. Through the operation of the low-level convolution kernel, the edge and texture basic feature maps are obtained. Then, the low-level feature maps are input into the high-level convolutional neural network as input. The high-level convolution kernel processes these feature maps to extract abstract features. Through multiple layers of convolution operations, a feature vector is obtained that can represent the key information in the image.

5. The method of claim 4, wherein the method further comprises: The computer vision monitoring unit also includes a self-attention mechanism, which operates as follows: By calculating the attention scores between each position in the image, it determines which areas need more attention. For each position in the image, calculate the attention weight with other positions. The calculation formula uses a similarity function divided by a scaling factor. Then, the attention weights are normalized by the softmax function to obtain the final attention weight matrix.

6. The method of claim 1, wherein the method further comprises: The genetic programming evolution maintenance strategy module works as follows: Input data collection and arrangement: Collect information related to fault diagnosis results, including fault types, charging pile equipment characteristics, and historical maintenance experience; Classification and coding: Classify and code the above data as input for genetic programming; Initial maintenance strategy generation: Randomly generate a set of initial maintenance strategies, with maintenance strategies represented in the form of gene encoding, each gene representing a specific maintenance procedure or operation step; Fitness evaluation: Establish a fitness function based on actual maintenance effect feedback; the fitness function is based on a comprehensive evaluation of multiple factors.

7. The method of claim 6, wherein the method further comprises: The genetic programming evolution maintenance strategy module also includes genetic operations and evolution, as follows: Selection operation: Calculate the fitness value of each individual according to the fitness function, use roulette selection, calculate the relative fitness of each individual, and then distribute the probability of each individual being selected according to the relative fitness size; Cross operation: Randomly select two individuals as parents from the population after selection, and perform cross operation according to the pre-set cross probability; use single-point cross method, randomly select a cross point, and exchange the genes before and after the point of the two parent individuals to generate two new offspring individuals; Mutation operation: For each individual in the population after cross operation, perform mutation operation according to the pre-set mutation probability; use the gene bit inversion method, randomly select a gene bit of the individual, and invert its value; Iterative evolution and optimization: Take the new population after selection, cross and mutation operations as the current population, and repeat the above fitness evaluation, selection, cross and mutation operations; select the individual with the maximum fitness value from the final population as the optimal solution of the maintenance strategy.

8. An artificial intelligence-based charging pile remote monitoring and early warning system, characterized in that, It includes: Bionic sensor network: composed of various sensors distributed in the internal, top, bottom and side parts of the charging pile, including temperature sensor, liquid and gas pressure sensor, gas sensor, chemical sensor, pressure sensor and vibration sensor, connected to the central processing system through the wireless communication module, forming a distributed data acquisition network; Computer vision monitoring unit: including high-definition camera and image acquisition card equipment, the camera is responsible for shooting the image of the charging pile, and the image acquisition card converts the analog signal into digital signal and transmits it to the central processing system; Data acquisition and transmission module: including sensor driver program, data acquisition program and wireless communication program, responsible for controlling sensors and cameras to collect data, and transmitting the collected data to the central processing system through the network; Central processing system: including storage device and processor, used for receiving, storing and processing sensor data and image data, running fault diagnosis model, maintenance strategy and algorithm; the fault diagnosis model includes swarm intelligence algorithm model and division and cooperation sub-model; the swarm intelligence algorithm model is particle swarm optimization algorithm; the particle swarm optimization auxiliary diagnosis method is as follows: Particle encoding and initialization: encode each possible fault mode as a particle's position vector; each dimension corresponds to a fault mode, taking values of 0 or 1, indicating whether the fault mode occurs or not; randomly generate multiple particles in the solution space as the initial population; Fitness function construction: construct the fitness function according to the goal of fault diagnosis; the fitness function is evaluated based on the matching degree of fault modes and sensor data, image data; Particle updating and iteration: update the speed and position of particles iteratively according to the speed and position update formula of particle swarm optimization algorithm; The division and cooperation sub-model simulation mode is as follows: Sub-model division: according to the fault type and data processing characteristics, the fault diagnosis task is divided into different sub-models, the fault diagnosis based on sensor data is taken as a sub-model, and the fault diagnosis based on image data is taken as another sub-model, each sub-model is responsible for analyzing specific type of data and preliminarily judging fault; Sensor data based sub-model: this sub-model uses neural network algorithm to process sensor data, the number of input layer nodes is equal to the number of sensor data types, the hidden layer is set according to actual needs, the number of output layer nodes is consistent with the number of possible fault modes, and the neural network is trained to enable the network to identify and classify fault modes according to sensor data; Image data based sub-model: convolutional neural network is used to analyze image data, CNN is composed of convolutional layer, pooling layer and fully connected layer, for fault diagnosis of charging pile image, convolutional layer is used to extract image features, then pooling layer is used to reduce feature dimension, finally fully connected layer is used to map features to fault mode space; Collaboration and fusion mechanism: after two sub-models get preliminary fault diagnosis results, they are collaborated and fused through fusion module, fusion module uses weighted fusion method to comprehensively analyze the results of two sub-models; Early warning unit: according to the result of central processing system, when abnormal situation or fault is detected, early warning signal is sent out; Control unit: in emergency, automatically control the charging pile to stop working; Human-computer interaction interface: used to display fault diagnosis results, alarm information and system running status; users configure and manage the system through human-computer interaction interface.

Images