Urban rail transit substation operation and maintenance method and system based on inspection robot

Abstract

The invention relates to the field of substation operation and maintenance, and provides an urban rail transit substation operation and maintenance method and system based on an inspection robot, and the method comprises the steps: obtaining the motion posture data of the inspection robot, and collecting environment point cloud data, visual image data and substation equipment state data in a substation in real time through the inspection robot; by fusing the environment point cloud data, the visual image data and the moving posture data, constructing a substation internal map by adopting a particle filtering algorithm; based on a Dijkstra algorithm and the internal map of the substation, planning an inspection path of the inspection robot; enabling the inspection robot to advance along the inspection path, obtaining the coordinates of the obstacle in real time through the environment point cloud data, and carrying out dynamic obstacle avoidance based on an artificial potential field method; constructing an equipment fault diagnosis model, and identifying the substation equipment state data to obtain a substation equipment anomaly category; and pushing the maintenance notification and the maintenance suggestion to a manager. According to the invention, the operation and maintenance safety and working efficiency of the substation equipment are improved.

Description

Technical Field

[0001] The present invention relates to the field of substation operation and maintenance, and in particular to an urban rail transit substation operation and maintenance method and system based on a patrol robot. Background Art

[0002] Urban rail transit substations are critical infrastructure within the urban rail transit system. They are primarily responsible for receiving and converting high-voltage electricity from the power system, converting it to voltage levels suitable for rail transit train operation and station equipment, and distributing and transmitting this electricity. Substations typically consist of main transformers, high-voltage distribution equipment, DC traction power supply systems, uninterruptible power supply systems, and monitoring systems. They are the core power source that ensures the normal operation of train power supply, station lighting, ventilation, signaling, and other systems. As the energy center of rail transit, the operating status of substations is directly related to the safety, reliability, and stability of the entire rail transit network, and is a critical foundation for the normal operation of urban public transportation.

[0003] Existing substation operations and maintenance utilize a monitoring method based on a fixed sensor network. This involves installing a variety of fixed sensors, such as those for temperature, vibration, and sound, on key substation equipment. This system collects equipment operating parameters in real time through a SCADA system, and uses simple threshold judgments to generate anomaly alarms. However, fixed sensors can only monitor predetermined measurement points and cannot comprehensively survey the substation environment. Furthermore, simple threshold judgments cannot identify complex equipment failure modes and abnormal conditions caused by the coupling of multiple parameters, resulting in a high false alarm rate and limited ability to predict potential failures. When equipment anomalies occur, the system can only generate simple alarms, lacking the ability to diagnose faults and generate maintenance recommendations. Summary of the Invention

[0004] In view of this, the present invention proposes an urban rail transit substation operation and maintenance method and system based on a patrol robot, which solves the problem that the existing technology can only monitor predetermined measurement points, cannot conduct a comprehensive inspection of the substation environment, cannot identify complex equipment failure modes and abnormal conditions of multi-parameter coupling, resulting in a high false alarm rate and limited ability to predict potential faults, and lacks the ability to diagnose faults and generate maintenance suggestions.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides an urban rail transit substation operation and maintenance method based on a patrol robot, comprising the following steps:

[0006] Obtain the motion posture data of the inspection robot, and use the inspection robot to collect environmental point cloud data, visual image data, and substation equipment status data in real time within the substation;

[0007] By fusing environmental point cloud data, visual image data, and motion posture data, a particle filter algorithm is used to construct an internal map of the substation.

[0008] Plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation;

[0009] The inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method;

[0010] Building an equipment fault diagnosis model based on a deep learning network, identifying substation equipment status data based on the equipment fault diagnosis model, and obtaining abnormal categories of substation equipment;

[0011] According to the abnormality category of the substation equipment, the corresponding maintenance notification and maintenance suggestion are pushed to the management personnel.

[0012] On the basis of the above technical solution, preferably, the acquisition of the motion posture data of the inspection robot, and the real-time collection of environmental point cloud data, visual image data and substation equipment status data by the inspection robot in the substation, include:

[0013] The inspection robot is equipped with an inertial measurement unit, a laser radar, a camera, a temperature sensor, and a sound sensor;

[0014] The inertial measurement unit carried by the inspection robot collects the acceleration and angular velocity of the inspection robot during operation to obtain the motion posture data of the inspection robot;

[0015] The inspection robot is equipped with a laser radar and camera to collect environmental point cloud data and visual image data respectively, and the substation equipment operating status data is collected based on temperature sensors and sound sensors.

[0016] On the basis of the above technical solution, preferably, the method of constructing the internal map of the substation by fusing environmental point cloud data, visual image data and motion posture data and using a particle filter algorithm includes:

[0017] The environmental point cloud data is processed by noise filtering and downsampling, the visual image data is subjected to contrast adjustment and edge refinement, the motion posture data is normalized, and the data is fused using a weighted fusion algorithm to obtain fused observation data.

[0018] Construct the original map, use the particle filter algorithm to predict the motion state of each particle, project the fused observation data into the original map coordinate system, and update the occupancy probability of each small unit in the original map using the occupancy grid map method to obtain the internal map of the substation;

[0019] The calculation formula of the occupancy probability is:

[0020]

[0021] Among them, p(m j |z 1:k ) is k moment m j The occupancy probability of the unit, N1 is the total number of particles in the particle filter, and each particle represents an estimation of the state of the inspection robot. is the weight of the i-th particle at time k, is the number m in the local map corresponding to the i-th particle j The occupancy probability of a cell.

[0022] On the basis of the above technical solution, preferably, the planning of the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation includes:

[0023] Discretizing the internal map of the substation into a weighted graph model, wherein each graph node corresponds to a grid in the map, and assigning a weight to each edge of the weighted graph model;

[0024] The Dijkstra algorithm is used on the weighted graph model to search for the shortest path and plan the inspection path of the inspection robot from the starting point to the target point.

[0025] On the basis of the above technical solution, preferably, the inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method, including:

[0026] Using the environmental point cloud data collected by the inspection robot in real time, the three-dimensional coordinate information of obstacles can be obtained in real time through data preprocessing and target detection methods;

[0027] Based on the artificial potential field method, combined with the three-dimensional coordinate information of the obstacle, a synthetic potential field for driving obstacle avoidance is calculated and generated. The motion trajectory of the inspection robot is adjusted according to the synthetic potential field to perform dynamic obstacle avoidance.

[0028] The calculation formula of the synthetic potential field is:

[0029]

[0030] d n =||P robot -P obs,n ||;

[0031] Among them, F total is the synthetic potential field, α2 is the attractive potential field coefficient, P target 、P robot 、P obs,nare the position vectors of the inspection robot, the target point, and the nth obstacle, N2 is the number of obstacles detected, and δ n is the repulsive potential field coefficient of the nth obstacle, d n is the Euclidean distance between the inspection robot and the nth obstacle, d th is the preset safety distance threshold, r and s are the first positive exponent and the second positive exponent respectively, and ||·|| is the Euclidean norm.

[0032] On the basis of the above technical solution, preferably, the device fault diagnosis model is constructed based on the deep learning network, and the substation device status data is identified based on the device fault diagnosis model to obtain the abnormality category of the substation equipment, including:

[0033] Obtain historical substation equipment status data, and perform normalization and standardization preprocessing on the historical substation equipment status data based on the deep learning network;

[0034] Extracting nonlinear features from historical substation equipment status data through a deep learning network, wherein the deep learning network includes an input layer, a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a flattening layer, a first fully connected layer, and an output layer;

[0035] The first convolutional layer includes 64 3×3 convolution kernels with a stride of 1 and a ReLU activation function. The second convolutional layer includes 128 3×3 convolution kernels with a stride of 1 and a ReLU activation function, and batch normalization is used. The first and second maximum pooling layers both use a 2×2 pooling window. The first fully connected layer includes 256 neurons and uses a ReLU activation function to output the abnormality discrimination score of each category of the device abnormality classification. The number of neurons in the output layer is the same as the total number of categories of the preset device abnormality classification. The softmax activation function is used to convert the abnormality discrimination score of each category of the device abnormality classification into the corresponding abnormality category probability distribution and output it.

[0036] The back-propagation algorithm is used to optimize the model parameters of the deep learning network, and the cross-entropy loss function is used as the objective function of model training to obtain the equipment fault diagnosis model;

[0037] Using the equipment fault diagnosis model to identify and feature-match substation equipment status data, identifying and classifying abnormal categories of substation equipment, the substation equipment status data includes multiple groups of equipment status data, each group of equipment status data corresponds to one substation equipment;

[0038] The substation equipment abnormality category includes an equipment abnormality classification category corresponding to each substation equipment.

[0039] On the basis of the above technical solution, preferably, the method of pushing corresponding maintenance notifications and maintenance suggestions to management personnel according to the abnormality type of the substation equipment includes:

[0040] According to the abnormality categories of the substation equipment and the preset fault knowledge base, maintenance notifications and maintenance suggestions for various abnormal conditions are generated, and the maintenance notifications and maintenance suggestions are pushed to management personnel through multiple channels.

[0041] In a second aspect, the present invention further provides an urban rail transit substation operation and maintenance system based on a patrol robot, the system comprising:

[0042] The data acquisition module is used to obtain the motion posture data of the inspection robot, and collect environmental point cloud data, visual image data and substation equipment status data in real time through the inspection robot in the substation;

[0043] A map construction module is used to construct a substation internal map by fusing environmental point cloud data, visual image data, and motion posture data using a particle filter algorithm;

[0044] The path planning module is used to plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation;

[0045] A dynamic obstacle avoidance module is used for the inspection robot to move along the inspection path, obtain obstacle coordinates in real time through environmental point cloud data, and perform dynamic obstacle avoidance based on the artificial potential field method;

[0046] A fault diagnosis module is used to build an equipment fault diagnosis model based on a deep learning network, identify substation equipment status data based on the equipment fault diagnosis model, and obtain the abnormality category of the substation equipment;

[0047] The maintenance push module is used to push corresponding maintenance notifications and maintenance suggestions to management personnel based on the abnormality category of the substation equipment.

[0048] In a third aspect, the present invention further provides an electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus;

[0049] Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a method for operating and maintaining an urban rail transit substation based on a patrol robot.

[0050] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions enable a computer to implement steps of an urban rail transit substation operation and maintenance method based on a patrol robot.

[0051] The urban rail transit substation operation and maintenance method and system based on the inspection robot of the present invention have the following beneficial effects compared with the prior art:

[0052] (1) Through multimodal sensor data fusion, a map is constructed based on the particle filter algorithm, the inspection path is planned using the Dijkstra algorithm, dynamic obstacle avoidance is achieved based on the artificial potential field method, and an equipment fault diagnosis model is constructed using a deep learning network. The process from data collection, environmental perception to anomaly identification is completed, and autonomous inspection, fault diagnosis and maintenance suggestion generation are achieved in the entire substation, thereby improving the safety and work efficiency of substation equipment operation and maintenance;

[0053] (2) By filtering out noise and downsampling environmental point cloud data, adjusting contrast and refining edges of visual image data, normalizing motion posture data, and fusing data using a weighted fusion algorithm, a particle filter algorithm and an occupancy grid map method are combined to construct an internal map of the substation, thereby improving map accuracy and dynamic update capabilities, and providing patrol robots with accurate and highly real-time environmental perception and positioning capabilities.

[0054] (3) By utilizing the environmental point cloud data collected by the inspection robot in real time, combined with data preprocessing and target detection methods, the three-dimensional coordinate information of obstacles is obtained in real time, and a synthetic potential field for driving obstacle avoidance is generated based on the artificial potential field method. The robot's motion trajectory is dynamically adjusted, which improves the flexibility and safety of the inspection robot in the substation. While ensuring the smooth completion of the inspection task, the inspection efficiency and environmental adaptability of the inspection robot are improved, and the dynamic obstacle avoidance function of the inspection robot in complex environments is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 This is a flow chart of an urban rail transit substation operation and maintenance method based on a patrol robot according to the present invention;

[0057] Figure 2This is a structural diagram of an urban rail transit substation operation and maintenance system based on a patrol robot according to the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] See also Figure 1 The present invention provides an urban rail transit substation operation and maintenance method based on a patrol robot, comprising the following steps:

[0060] Obtain the motion posture data of the inspection robot, and use the inspection robot to collect environmental point cloud data, visual image data, and substation equipment status data in real time within the substation;

[0061] By fusing environmental point cloud data, visual image data, and motion posture data, a particle filter algorithm is used to construct an internal map of the substation.

[0062] Plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation;

[0063] The inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method;

[0064] Building an equipment fault diagnosis model based on a deep learning network, identifying substation equipment status data based on the equipment fault diagnosis model, and obtaining abnormal categories of substation equipment;

[0065] According to the abnormality category of the substation equipment, the corresponding maintenance notification and maintenance suggestion are pushed to the management personnel.

[0066] Specifically, this embodiment fuses multimodal sensor data, builds a map based on the particle filter algorithm, plans inspection routes using the Dijkstra algorithm, implements dynamic obstacle avoidance based on the artificial potential field method, and uses a deep learning network to build an equipment fault diagnosis model. It completes the process from data collection, environmental perception to anomaly identification, and realizes full-range autonomous inspection, fault diagnosis, and maintenance suggestion generation of the substation, thereby improving the safety and work efficiency of substation equipment operation and maintenance.

[0067] The acquisition of the motion posture data of the inspection robot, and the real-time collection of environmental point cloud data, visual image data, and substation equipment status data by the inspection robot in the substation, include:

[0068] The inspection robot is equipped with an inertial measurement unit, a laser radar, a camera, a temperature sensor, and a sound sensor;

[0069] The inertial measurement unit carried by the inspection robot collects the acceleration and angular velocity of the inspection robot during operation to obtain the motion posture data of the inspection robot.

[0070] In a specific embodiment, the inertial measurement unit includes an accelerometer and a gyroscope, and the collected acceleration and angular velocity data are fused through a Kalman filter algorithm to calculate the real-time three-axis motion posture of the inspection robot.

[0071] The inspection robot is equipped with a laser radar and camera to collect environmental point cloud data and visual image data respectively, and the substation equipment operating status data is collected based on temperature sensors and sound sensors.

[0072] In a specific embodiment, the point cloud data collected by the lidar is filtered through a preset noise reduction algorithm to remove external interference noise, and the accuracy of the point cloud data is improved through a spatial registration algorithm; the visual image data is feature extracted and segmented through an edge computing device, and combined with the point cloud data to achieve higher-precision data collection results.

[0073] Specifically, this embodiment improves the accuracy and reliability of the inspection robot's motion data by equipping the inspection robot with an inertial measurement unit, a lidar, a camera, a temperature sensor, and a sound sensor, and using a Kalman filter algorithm to fuse the data of the accelerometer and the gyroscope to obtain real-time three-axis motion posture.

[0074] By collecting point cloud data through lidar and processing it using noise reduction algorithms and spatial registration algorithms, the accuracy of point cloud data has been greatly improved. At the same time, by collecting visual image data through cameras and using edge computing devices for feature extraction and segmentation, combined with point cloud data, higher-precision environmental data perception capabilities have been achieved.

[0075] The method of constructing a substation internal map by fusing environmental point cloud data, visual image data, and motion posture data and using a particle filter algorithm includes:

[0076] The environmental point cloud data is processed by noise filtering and downsampling, the visual image data is subjected to contrast adjustment and edge refinement, the motion posture data is normalized, and the data is fused using a weighted fusion algorithm to obtain fused observation data.

[0077] Construct an original map, use a particle filter algorithm to predict the motion state of each particle, project the fused observation data into the original map coordinate system, and update the occupancy probability of each small cell in the original map using the occupancy grid map method. Based on the motion state prediction of the particle at each moment and the fused observation data, the occupancy probability of each small cell is gradually updated to obtain the internal map of the substation.

[0078] The calculation formula of the occupancy probability is:

[0079]

[0080] Among them, p(m j |z 1:k ) is k moment m j The occupancy probability of the unit, N1 is the total number of particles in the particle filter, and each particle represents an estimation of the state of the inspection robot. is the weight of the i-th particle at time k, is the number m in the local map corresponding to the i-th particle j The occupancy probability of a cell.

[0081] Specifically, this embodiment performs noise filtering and downsampling on environmental point cloud data, performs contrast adjustment and edge refinement on visual image data, normalizes motion posture data, and adopts a weighted fusion algorithm for data fusion. It combines the particle filter algorithm with the occupancy grid map method to construct an internal map of the substation, thereby improving the map accuracy and dynamic update capability, and providing the inspection robot with accurate and highly real-time environmental perception and positioning capabilities.

[0082] The inspection path of the inspection robot is planned based on the Dijkstra algorithm and the internal map of the substation, including:

[0083] Discretize the substation internal map into a weighted graph model, where each graph node corresponds to a grid in the map, and assign a weight describing the distance and environmental risk (e.g., obstacle occupancy probability) to each edge of the weighted graph model;

[0084] The Dijkstra algorithm is used on the weighted graph model to search for the shortest path and plan the inspection path of the inspection robot from the starting point to the target point.

[0085] In a specific embodiment, the calculation formula of the inspection path is:

[0086] d(m)=min{d(m),d(l)+f(l,m)};

[0087]

[0088] Among them, d(l) and d(m) are the shortest path costs of the inspection robot from the starting point to the target point l and the target point m, respectively. f(l,m) is the path cost of the inspection robot from the target point l to the target point m. d(l,m) is the Euclidean distance between the target point l and the target point m. p(m lm ) is the average occupancy probability of the area between target point l and target point m, p max is the maximum threshold of the occupancy probability, α1, β1, and γ1 are the weights of the distance cost, obstacle avoidance cost, and environmental uncertainty cost, respectively, and Δp(l,m) is the standard deviation of the occupancy probability in the boundary area from target point l to target point m.

[0089] Specifically, this embodiment discretizes the internal map of the substation into a weighted graph model that takes distance and environmental risks into consideration, and uses the Dijkstra algorithm combined with a multi-factor comprehensive evaluation path calculation formula to search for the shortest path, thereby achieving intelligent planning of the inspection path.

[0090] This embodiment not only takes into account the physical distance of the inspection path, but also comprehensively evaluates various environmental factors such as obstacle occupancy probability and regional uncertainty. By dynamically adjusting weight parameters, it generates a safe and efficient inspection path, thereby improving the path planning efficiency and safety of the inspection robot in the complex substation environment, enabling the inspection operation to avoid high-risk areas while ensuring the inspection coverage and task completion efficiency.

[0091] The inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method, including:

[0092] Using the environmental point cloud data collected by the inspection robot in real time, the three-dimensional coordinate information of obstacles can be obtained in real time through data preprocessing and target detection methods;

[0093] Based on the artificial potential field method, combined with the three-dimensional coordinate information of the obstacle, a synthetic potential field for driving obstacle avoidance is calculated and generated. The motion trajectory of the inspection robot is adjusted according to the synthetic potential field to perform dynamic obstacle avoidance.

[0094] The calculation formula of the synthetic potential field is:

[0095]

[0096] d n =||P robot -P obs,n ||;

[0097] Among them, F total is the synthetic potential field, α2 is the attractive potential field coefficient, P target 、P robot 、P obs,nare the position vectors of the inspection robot, the target point, and the nth obstacle, N2 is the number of obstacles detected, and δ n is the repulsive potential field coefficient of the nth obstacle, d n is the Euclidean distance between the inspection robot and the nth obstacle, d th is the preset safety distance threshold, r and s are the first positive exponent and the second positive exponent respectively, and ||·|| is the Euclidean norm.

[0098] Specifically, this embodiment utilizes the environmental point cloud data collected by the inspection robot in real time, combines data preprocessing and target detection methods, obtains the three-dimensional coordinate information of obstacles in real time, and generates a synthetic potential field to drive obstacle avoidance based on the artificial potential field method, dynamically adjusts the robot's motion trajectory, and improves the flexibility and safety of the inspection robot in the substation. While ensuring the smooth completion of the inspection task, it improves the inspection efficiency and environmental adaptability of the inspection robot, and realizes the dynamic obstacle avoidance function of the inspection robot in complex environments.

[0099] The device fault diagnosis model is constructed based on the deep learning network, and the substation device status data is identified based on the device fault diagnosis model to obtain the abnormality category of the substation equipment, including:

[0100] Obtain historical substation equipment status data, and perform normalization and standardization preprocessing on the historical substation equipment status data based on the deep learning network;

[0101] Extracting nonlinear features from historical substation equipment status data through a deep learning network, wherein the deep learning network includes an input layer, a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a flattening layer, a first fully connected layer, and an output layer;

[0102] The first convolutional layer includes 64 3×3 convolution kernels with a stride of 1 and a ReLU activation function. The second convolutional layer includes 128 3×3 convolution kernels with a stride of 1 and a ReLU activation function, and batch normalization is used. The first and second maximum pooling layers both use a 2×2 pooling window. The first fully connected layer includes 256 neurons and uses a ReLU activation function to output the abnormality discrimination score of each category of the device abnormality classification. The number of neurons in the output layer is the same as the total number of categories of the preset device abnormality classification. The softmax activation function is used to convert the abnormality discrimination score of each category of the device abnormality classification into the corresponding abnormality category probability distribution and output it.

[0103] The back-propagation algorithm is used to optimize the model parameters of the deep learning network, and the cross-entropy loss function is used as the objective function of model training to obtain the equipment fault diagnosis model;

[0104] Using the equipment fault diagnosis model to identify and feature-match substation equipment status data, identifying and classifying abnormal categories of substation equipment, the substation equipment status data includes multiple groups of equipment status data, each group of equipment status data corresponds to one substation equipment;

[0105] The substation equipment abnormality category includes an equipment abnormality classification category corresponding to each substation equipment.

[0106] In a specific embodiment, the calculation formula of the abnormality discrimination score and the abnormality category probability distribution is:

[0107]

[0108]

[0109] in, is the abnormality discrimination score of the Kth device abnormality classification category of the hth group of device status data, is the probability that the h-th group of device status data is classified into the K-th device abnormality classification category, and are the linear discriminant weight and nonlinear discriminant weight of the K-th device abnormality classification category, θ K and b K are the weight vector and bias term of the K-th device anomaly classification category, ||·|| is the Euclidean norm, Φ h (x) is the input vector of the hth group of device status data, ε is the anti-zero constant, μ K is the representative center of the Kth device anomaly classification category, N3 is the total number of device anomaly classification categories, is the abnormality discrimination score of the g-th device abnormality classification category of the h-th group of device status data, and exp(·) is the exponential function.

[0110] Specifically, this embodiment adopts a deep learning architecture with two convolutional layers, two pooling layers and a fully connected layer. The first convolutional layer uses 64 3×3 convolution kernels to extract basic features, and the second convolutional layer uses 128 3×3 convolution kernels to capture more advanced features. This progressive feature extraction capability improves the ability to characterize complex device status signals.

[0111] Through normalization and standardization preprocessing, the differences between data of different dimensions are eliminated, and the network convergence process is accelerated; the batch normalization technology introduced in the second convolutional layer effectively suppresses the gradient vanishing problem and improves the model training stability and generalization ability.

[0112] The dual discrimination strategy combining linear discrimination weights and nonlinear discrimination weights is introduced, which not only takes into account the linear separability of the feature space, but also takes into account the nonlinear characteristics of the equipment status data, greatly improving the model's ability to recognize boundary conditions and complex fault modes.

[0113] The softmax activation function is used to convert the anomaly discrimination score into a probability distribution, which not only provides the judgment result of the anomaly category, but also gives the confidence level of the judgment, providing a more comprehensive basis for operation and maintenance decisions.

[0114] By continuously optimizing model parameters through the back-propagation algorithm and cross-entropy loss function, the model is equipped with the ability to continuously learn and adapt. As data accumulates, the diagnostic accuracy can be continuously improved, and it can adapt to new failure modes caused by equipment aging and environmental changes.

[0115] According to the abnormality type of the substation equipment, the corresponding maintenance notification and maintenance suggestion are pushed to the management personnel, including:

[0116] According to the abnormality categories of the substation equipment and the preset fault knowledge base, maintenance notifications and maintenance suggestions for various abnormal conditions are generated.

[0117] In a specific embodiment, the specific process of generating the maintenance notification and maintenance suggestion includes:

[0118] Match the abnormal classification results of substation equipment with the preset fault knowledge base, which contains the typical characteristics, possible causes, treatment methods and historical case library of various abnormal conditions;

[0119] Based on the matching results, the corresponding maintenance suggestion template is automatically extracted for the specific abnormality type, and the maintenance suggestion is personalized and supplemented based on the real-time collected equipment status data;

[0120] Based on the importance of equipment, degree of abnormality and potential risks, the fuzzy comprehensive evaluation method is used to assess the risk level of abnormal status, and the risk level is divided into three levels: emergency, important and routine, which serve as the basis for the push strategy.

[0121] Maintenance notifications and maintenance recommendations are pushed to managers through multiple channels.

[0122] In a specific embodiment, the specific process of the multi-channel targeted push includes:

[0123] For emergency-level abnormal conditions, relevant management personnel will be notified immediately through SMS, phone calls, App push, etc., and the sound and light warning system will be triggered;

[0124] For critical abnormal conditions, relevant managers will be notified via App push and work group notifications, requiring them to respond within the specified time.

[0125] For abnormal conditions of regular levels, they will be pushed to relevant managers in the form of daily work reports for regular processing.

[0126] Specifically, this embodiment matches the equipment anomaly classification results with a preset fault knowledge base, and utilizes the typical characteristics of abnormal conditions, possible causes, processing methods, and historical case libraries contained in the knowledge base to quickly generate highly targeted maintenance notifications and suggestions.

[0127] By automatically extracting maintenance suggestion templates and combining them with real-time equipment status data for personalized supplementation, the targetedness and operability of maintenance suggestions are enhanced, the time cost of manual analysis and judgment is reduced, and maintenance efficiency is improved.

[0128] The risk level of abnormal conditions is assessed through a comprehensive evaluation method, and divided into three levels: emergency, important, and routine according to the importance of the equipment, the degree of abnormality, and the potential risks. This effectively distinguishes the processing priorities of different abnormal conditions, avoids the situation where urgent problems are ignored or delayed, and ensures that high-risk problems can be paid attention to and resolved by managers in a timely manner.

[0129] Emergency-level abnormal conditions are notified in real time via SMS, phone calls, app push notifications, and audio and visual warning systems, ensuring synchronization and timeliness of notifications. Important-level conditions are notified via app notifications and work group push notifications to reduce information omissions and clearly stipulate response times. Routine-level conditions are communicated via daily work reports to facilitate regular processing and data archiving. This multi-level push mechanism meets the notification needs of abnormal conditions at different levels and avoids confusion and omissions in information processing.

[0130] By matching equipment anomaly classification with the knowledge base, we can automatically generate operation and maintenance recommendations, significantly reducing reliance on expert experience and improving the intelligent operation and maintenance of substations. Furthermore, the integration of risk assessment and push strategies further enhances the proactiveness and reliability of operation and maintenance responses, providing managers with efficient and intuitive intelligent decision-making support.

[0131] Real-time, multi-channel notifications for emergency-level abnormal conditions enable rapid response to sudden equipment failures, preventing serious accidents. Categorized management of critical and routine conditions optimizes daily O&M resource allocation and reduces the likelihood of low-risk issues triggering cascading failures. Overall, this shortens maintenance response time and significantly reduces downtime and economic losses caused by equipment anomalies.

[0132] Maintenance recommendations are dynamically supplemented based on real-time collected equipment status data, and the fault knowledge base is continuously improved in combination with the historical case library, so that maintenance recommendations have the ability to be personalized and sustainably optimized, and gradually adapt to the abnormal modes and processing requirements of different equipment.

[0133] See also Figure 2 The present invention also provides an urban rail transit substation operation and maintenance system based on a patrol robot, the system comprising:

[0134] The data acquisition module is used to obtain the motion posture data of the inspection robot, and collect environmental point cloud data, visual image data and substation equipment status data in real time through the inspection robot in the substation;

[0135] A map construction module is used to construct a substation internal map by fusing environmental point cloud data, visual image data, and motion posture data using a particle filter algorithm;

[0136] The path planning module is used to plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation;

[0137] A dynamic obstacle avoidance module is used for the inspection robot to move along the inspection path, obtain obstacle coordinates in real time through environmental point cloud data, and perform dynamic obstacle avoidance based on the artificial potential field method;

[0138] A fault diagnosis module is used to build an equipment fault diagnosis model based on a deep learning network, identify substation equipment status data based on the equipment fault diagnosis model, and obtain the abnormality category of the substation equipment;

[0139] The maintenance push module is used to push corresponding maintenance notifications and maintenance suggestions to management personnel based on the abnormality category of the substation equipment.

[0140] Specifically, an urban rail transit substation operation and maintenance system based on a patrol robot in this embodiment realizes full process coverage of equipment status data collection, exception handling and efficient maintenance through data acquisition, map construction, path planning, dynamic obstacle avoidance, fault diagnosis and maintenance push modules; it adopts methods such as deep learning, Dijkstra algorithm and artificial potential field method, precise path planning, dynamic obstacle avoidance and fault diagnosis capabilities, as well as multi-channel push strategies to improve operation and maintenance efficiency, safety and reliability, and provide an efficient overall solution for substation operation and maintenance.

[0141] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement an urban rail transit substation operation and maintenance method based on a patrol robot.

[0142] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the patrol robot-based urban rail transit substation operation and maintenance method described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An urban rail transit substation operation and maintenance method based on a patrol robot, characterized in that: The following steps are involved: Obtain the motion posture data of the inspection robot, and use the inspection robot to collect environmental point cloud data, visual image data, and substation equipment status data in real time within the substation; By fusing environmental point cloud data, visual image data, and motion posture data, a particle filter algorithm is used to construct an internal map of the substation. Plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation; The inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method; Building an equipment fault diagnosis model based on a deep learning network, identifying substation equipment status data based on the equipment fault diagnosis model, and obtaining abnormal categories of substation equipment; According to the abnormality category of the substation equipment, the corresponding maintenance notification and maintenance suggestion are pushed to the management personnel.

2. The urban rail transit substation operation and maintenance method based on the inspection robot according to claim 1, characterized in that: The acquisition of the motion posture data of the inspection robot, and the real-time collection of environmental point cloud data, visual image data, and substation equipment status data by the inspection robot in the substation, include: The inspection robot is equipped with an inertial measurement unit, a laser radar, a camera, a temperature sensor, and a sound sensor; The inertial measurement unit carried by the inspection robot collects the acceleration and angular velocity of the inspection robot during operation to obtain the motion posture data of the inspection robot; The inspection robot is equipped with a laser radar and camera to collect environmental point cloud data and visual image data respectively, and the substation equipment operating status data is collected based on temperature sensors and sound sensors.

3. The urban rail transit substation operation and maintenance method based on the inspection robot according to claim 1, characterized in that: The method of constructing a substation internal map by fusing environmental point cloud data, visual image data, and motion posture data and using a particle filter algorithm includes: The environmental point cloud data is processed by noise filtering and downsampling, the visual image data is subjected to contrast adjustment and edge refinement, the motion posture data is normalized, and the data is fused using a weighted fusion algorithm to obtain fused observation data. Construct the original map, use the particle filter algorithm to predict the motion state of each particle, project the fused observation data into the original map coordinate system, and update the occupancy probability of each small unit in the original map using the occupancy grid map method to obtain the internal map of the substation; The calculation formula of the occupancy probability is: Among them, p(m j |z 1:k ) is k moment m j The occupancy probability of the unit, N1 is the total number of particles in the particle filter, and each particle represents an estimation of the state of the inspection robot. is the weight of the i-th particle at time k, is the number m in the local map corresponding to the i-th particle j The occupancy probability of a cell.

4. The urban rail transit substation operation and maintenance method based on the inspection robot according to claim 3 is characterized in that: The inspection path of the inspection robot is planned based on the Dijkstra algorithm and the internal map of the substation, including: Discretizing the internal map of the substation into a weighted graph model, wherein each graph node corresponds to a grid in the map, and assigning a weight to each edge of the weighted graph model; The Dijkstra algorithm is used on the weighted graph model to search for the shortest path and plan the inspection path of the inspection robot from the starting point to the target point.

5. The urban rail transit substation operation and maintenance method based on the inspection robot according to claim 1, characterized in that: The inspection robot moves along the inspection path, obtains obstacle coordinates in real time through environmental point cloud data, and performs dynamic obstacle avoidance based on an artificial potential field method, including: Using the environmental point cloud data collected by the inspection robot in real time, the three-dimensional coordinate information of obstacles can be obtained in real time through data preprocessing and target detection methods; Based on the artificial potential field method, combined with the three-dimensional coordinate information of the obstacle, a synthetic potential field for driving obstacle avoidance is calculated and generated. The motion trajectory of the inspection robot is adjusted according to the synthetic potential field to perform dynamic obstacle avoidance. The calculation formula of the synthetic potential field is: d n =||P robot -P obs,n ||; Among them, F total is the synthetic potential field, α2 is the attractive potential field coefficient, P target 、P robot 、P obs,n are the position vectors of the inspection robot, the target point, and the nth obstacle, N2 is the number of obstacles detected, and δ n is the repulsive potential field coefficient of the nth obstacle, d n is the Euclidean distance between the inspection robot and the nth obstacle, d th is the preset safety distance threshold, r and s are the first positive exponent and the second positive exponent respectively, and ||·|| is the Euclidean norm.

6. The urban rail transit substation operation and maintenance method based on the inspection robot according to claim 1, characterized in that: The device fault diagnosis model is constructed based on the deep learning network, and the substation device status data is identified based on the device fault diagnosis model to obtain the abnormality category of the substation equipment, including: Obtain historical substation equipment status data, and perform normalization and standardization preprocessing on the historical substation equipment status data based on the deep learning network; Extracting nonlinear features from historical substation equipment status data through a deep learning network, wherein the deep learning network includes an input layer, a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a flattening layer, a first fully connected layer, and an output layer; The first convolutional layer includes 64 3×3 convolution kernels with a stride of 1 and a ReLU activation function. The second convolutional layer includes 128 3×3 convolution kernels with a stride of 1 and a ReLU activation function, and batch normalization is used. The first and second maximum pooling layers both use a 2×2 pooling window. The first fully connected layer includes 256 neurons and uses a ReLU activation function to output the abnormality discrimination score of each category of the device abnormality classification. The number of neurons in the output layer is the same as the total number of categories of the preset device abnormality classification. The softmax activation function is used to convert the abnormality discrimination score of each category of the device abnormality classification into the corresponding abnormality category probability distribution and output it. The back-propagation algorithm is used to optimize the model parameters of the deep learning network, and the cross-entropy loss function is used as the objective function of model training to obtain the equipment fault diagnosis model; Using the equipment fault diagnosis model to identify and feature-match substation equipment status data, identifying and classifying abnormal categories of substation equipment, the substation equipment status data includes multiple groups of equipment status data, each group of equipment status data corresponds to one substation equipment; The substation equipment abnormality category includes an equipment abnormality classification category corresponding to each substation equipment.

7. The urban rail transit substation operation and maintenance method based on a patrol robot according to claim 1, characterized in that: According to the abnormality type of the substation equipment, the corresponding maintenance notification and maintenance suggestion are pushed to the management personnel, including: According to the abnormality categories of the substation equipment and the preset fault knowledge base, maintenance notifications and maintenance suggestions for various abnormal conditions are generated, and the maintenance notifications and maintenance suggestions are pushed to management personnel through multiple channels.

8. An urban rail transit substation operation and maintenance system based on a patrol robot, used to execute an urban rail transit substation operation and maintenance method based on a patrol robot according to any one of claims 1 to 7, characterized in that: The system comprises: The data acquisition module is used to obtain the motion posture data of the inspection robot, and collect environmental point cloud data, visual image data and substation equipment status data in real time through the inspection robot in the substation; A map construction module is used to construct a substation internal map by fusing environmental point cloud data, visual image data, and motion posture data using a particle filter algorithm; The path planning module is used to plan the inspection path of the inspection robot based on the Dijkstra algorithm and the internal map of the substation; A dynamic obstacle avoidance module is used for the inspection robot to move along the inspection path, obtain obstacle coordinates in real time through environmental point cloud data, and perform dynamic obstacle avoidance based on the artificial potential field method; A fault diagnosis module is used to build an equipment fault diagnosis model based on a deep learning network, identify substation equipment status data based on the equipment fault diagnosis model, and obtain the abnormality category of the substation equipment; The maintenance push module is used to push corresponding maintenance notifications and maintenance suggestions to management personnel based on the abnormality category of the substation equipment.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images