Substation route inspection planning method based on deep learning

Abstract

The invention discloses a substation route inspection planning method based on deep learning. The method comprises the following steps: acquiring substation equipment state data and environment monitoring data; performing multi-dimensional feature analysis on the equipment state data based on a preset depth feature extraction model to generate an equipment health degree evaluation vector; based on the environment monitoring data and the equipment health degree evaluation vector, multiple candidate paths are generated through a reinforcement learning algorithm, an initial inspection path is obtained through screening in the multiple candidate paths, and the initial inspection path is dynamically optimized by fusing geographic information in real time; according to unmanned aerial vehicle positioning data and equipment state feedback, if it is detected that the health degree is lower than a threshold value or the environmental risk exceeds the limit, local path re-planning is triggered, and navigation to high-risk equipment is preferentially carried out. The method has the following advantages and effects that the intelligence, accuracy and safety of routing inspection path planning can be improved, and it is ensured that the unmanned aerial vehicle can efficiently and reliably execute the substation equipment routing inspection task.

Description

Technical Field

[0001] The present invention relates to the inspection route planning of unmanned aerial vehicles (UAVs), and particularly to a method for planning the inspection route of a substation based on deep learning. Background Art

[0002] With the continuous expansion of the scale of the power system, as an important part of the power transmission and distribution network, the stable operation of the equipment in the substation directly affects the safety and reliability of the power system. Therefore, the inspection work of the substation has become an important link in ensuring the safety of the power system. The traditional inspection methods of substations mainly rely on manual inspection and regular detection, but these two methods have problems such as low efficiency, small coverage, and slow response speed, and it is difficult to meet the inspection requirements of modern substations.

[0003] With the development of UAV technology and deep learning technology, the inspection scheme based on UAVs has gradually become a hot topic in the power industry. UAVs have advantages such as high efficiency, flexibility, and automation, and can quickly obtain the operating status of substation equipment, monitor the health status of equipment in real time, and provide more comprehensive and accurate inspection data. However, in practical applications, due to the complexity of the inspection area of the substation and the uncertainty of the environment, traditional route planning methods are difficult to cope with complex working environments, such as problems like the dynamic changes of equipment status and the real-time changes of obstacles.

[0004] The existing UAV-based inspection route planning methods mainly have the following deficiencies: First, they do not fully consider the impact of equipment status and environmental risks on the inspection path, resulting in low safety and efficiency of the inspection path; second, most of the path planning lacks real-time response to changes in equipment health, and cannot adjust the inspection path in time to cope with equipment failures or environmental changes; third, most existing methods ignore the need for multi-objective optimization and cannot take into account multiple factors such as efficiency, safety, and emergency tasks at the same time.

[0005] Therefore, in view of the above problems, the present invention proposes a method for planning the inspection route of a substation based on deep learning, aiming to improve the intelligence, accuracy, and safety of the inspection path planning, and ensure that the UAV can efficiently and reliably perform the inspection task of substation equipment. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for planning the inspection route of a substation based on deep learning to solve the problems raised in the background art.

[0007] The above technical purpose of the present invention is achieved through the following technical solutions: A method for planning the inspection route of a substation based on deep learning includes the following steps: S100. Obtain the substation equipment status data and environmental monitoring data; wherein, the equipment status data includes equipment temperature, vibration frequency, and insulation resistance value, and the environmental monitoring data includes meteorological parameters, light intensity, and obstacle distribution; S200. Based on a preset deep feature extraction model, perform multi-dimensional feature analysis on the equipment status data to generate an equipment health assessment vector, and the generated equipment health assessment vector can reflect different states of the equipment; wherein, the deep feature extraction model includes a dual-modal structure of a convolutional neural network and a long short-term memory network, which is used to extract the spatial features and time series of the equipment status data; S300. Based on the environmental monitoring data and the equipment health assessment vector, generate multiple candidate paths through a reinforcement learning algorithm, screen an initial inspection path from the multiple candidate paths, and dynamically optimize the initial inspection path by fusing geographical information in real time; S400. According to the UAV positioning data and the equipment status feedback, if it is detected that the health degree is lower than the threshold or the environmental risk exceeds the limit, trigger local path replanning and preferentially navigate to high-risk equipment.

[0008] By adopting the above technical solution, integrating a variety of equipment status data and environmental monitoring data provides an efficient and intelligent solution for substation inspection planning; substation inspection tasks usually include complex equipment monitoring and environmental assessment requirements. The present invention can accurately judge the health status of equipment, thereby formulating the optimal inspection path; the deep feature extraction model conducts equipment health assessment, which can extract effective features from various operating states of the equipment to form a comprehensive health assessment vector, which can not only quantify the operating state of the equipment, but also reflect the potential fault risks existing in the equipment, thereby providing important information for the optimization of the inspection path; at the same time, the present invention also enhances the dynamic adaptability of path planning. The UAV can dynamically adjust the inspection path according to real-time environmental data and equipment health assessment results. This mechanism significantly improves the flexibility of the inspection task and the ability to cope with complex environmental changes, thereby realizing efficient and safe UAV inspection in a complex industrial environment such as a substation.

[0009] A further setting is that in S100, obtaining the substation equipment status data and environmental monitoring data specifically is: Collect the equipment temperature, vibration frequency, and insulation resistance value through a distributed sensor network, and perform normalization processing on the equipment status data; wherein, the distributed sensor network is deployed at key monitoring nodes of substation equipment and includes an infrared sensor, an acceleration sensor, and an insulation monitoring unit; Obtain the meteorological parameters, light intensity, and obstacle distribution through a multi-spectral camera and lidar carried by the UAV.

[0010] By adopting the above technical solution, through the distributed sensor network and the sensor system on the unmanned aerial vehicle, comprehensive monitoring of the equipment and the environment is achieved; the equipment status data is an important basis for judging the health of the equipment, while the environmental monitoring data provides detailed information on the external environment for path planning. The acquisition of multi-source data helps the model better identify and evaluate the health status of the equipment; the deployment of the sensor network enables each key node of the substation to feedback the working status of the equipment in real time, and the distributed monitoring improves the efficiency of the patrol inspection and avoids the blind spots and errors that may be brought by manual patrol inspection; at the same time, the real-time nature of data acquisition ensures a rapid response to emergencies; the sensors such as the multi-spectral camera and lidar carried by the man-machine provide all-round sensing capabilities for environmental monitoring. Especially in the complex environment inside the substation, it can accurately obtain information such as obstacle distribution, light intensity, and meteorological changes, thus providing real-time and accurate environmental data support for path planning.

[0011] A further setting is that the S200 is specifically as follows: S201. Input the equipment temperature, vibration frequency, and insulation resistance value into the parallel channels of the convolutional neural network respectively, and extract local abnormal features through multi-scale convolutional kernels; S202. Input the local abnormal features and the time series of historical equipment status data into the long short-term memory network; S203. Perform tensor splicing on the spatial features and time series features, and map them into an equipment health degree evaluation vector through a fully connected layer.

[0012] By adopting the above technical solution, through the convolutional neural network and the long short-term memory network, the spatial features and time series features in the equipment status data can be fully mined. The convolutional neural network extracts local abnormal features through multi-scale convolutional kernels, which is of great significance for monitoring the abnormal status of the equipment and can effectively identify minor faults or early fault signs of the equipment; the long short-term memory network can deeply analyze the historical status data of the equipment and capture the dynamic change rules in the time series, which is particularly crucial during the operation of the equipment because the health status of the equipment is not only affected by the current environment but also closely related to its historical operation conditions. The long short-term memory network can effectively process data with a long time span and accurately predict the future health status of the equipment; after performing tensor splicing on the spatial features and time features to generate a health degree evaluation vector, it can not only comprehensively evaluate the operation status of the equipment but also reflect the risk degree of the equipment. This comprehensive evaluation method provides a scientific basis for subsequent path planning and can issue warnings in a timely manner when the equipment status is abnormal to ensure the efficient execution of the patrol inspection task.

[0013] A further setting is that the S201 is specifically as follows: Standardize the device temperature data, vibration frequency signals, and insulation resistance values respectively to eliminate the dimensional differences; Input the standardized device temperature data, vibration frequency signals, and insulation resistance values into three independent input channels of a convolutional neural network respectively. Each channel corresponds to a data type to achieve parallel processing of multi-modal data and generate local anomaly feature tensors; Specifically, S202 is: Arrange the local anomaly features in chronological order as a time series and align the timestamps with the historical device status data; Add time encoding parameters to the historical data; Construct a long short-term memory network. Its input layer receives the concatenated local anomaly features and time encoding parameters, and dynamically adjusts the information flow through the gating mechanisms of the input gate, forget gate, and output gate; Specifically, S203 is: Concatenate the spatial features output by the convolutional neural network and the temporal features output by the long short-term memory network to construct a fully connected network, and constrain the output value within the range of 0.0 - 1.0 to generate a device health assessment vector: If the health degree is greater than the set first health degree threshold, it indicates a normal state; If the health degree is between the first health degree threshold and the second health degree threshold, it indicates the existence of potential risks; If the health degree is less than the second health degree threshold, it indicates a high-risk state and triggers an alarm.

[0014] By adopting the above technical solutions, the dimensional differences between different sensor data are eliminated through standardizing the device data, ensuring that the data can be effectively fused in the same model; Standardization is a key step in data preprocessing, which ensures the balance of various types of data when input into the deep learning model and avoids the interference of data with different scales on the model learning process; The parallel processing of multi-modal data is achieved through three independent input channels of the convolutional neural network. Each channel corresponds to a data type. This parallel processing method can efficiently extract the features of device data. Especially when facing complex device states, it can fully explore the potential information in various sensor data, which helps to improve the accuracy of fault prediction; Concatenating the spatial features output by the convolutional neural network and the temporal features output by the long short-term memory network to generate a fully connected network not only improves the accuracy of the assessment but also ensures the stability of the device health assessment; By constraining the output value within the range of 0.0 - 1.0, different states of device health can be clearly defined, providing clear guidance for subsequent inspection decisions.

[0015] A further setting is that specifically, S300 is: S301. Define the 3D perception space of the substation area, including equipment coordinates, obstacle distribution, and environmental risks; divide the substation area into grid cells, and each grid cell is marked as one of the following states: Passable area, indicating that there are no obstacles inside; Low-risk obstacle, indicating a stationary obstacle; High-risk obstacle, indicating a dynamic obstacle or a high-risk environmental area; S302. Design a multi-objective reward function; S303. Optimize the initial path through a path smoothing algorithm to ensure that the UAV flight trajectory is continuous and conforms to kinematic constraints.

[0016] By adopting the above technical solutions, by defining the 3D perception space of the substation area and dividing the grid cells, a comprehensive perception of the substation environment can be achieved. This area division can clearly mark the passable area, low-risk obstacle area, and high-risk obstacle area, which helps the UAV avoid obstacles during the inspection process and ensure flight safety; at the same time, the management after zoning can improve the flexibility of path planning, enabling the UAV to more efficiently plan the inspection path; the design of the multi-objective reward function takes into account various factors such as efficiency, safety, and emergency tasks, making the path planning process not only focus on the shortest distance of the path, but also reasonably balance the task priority and flight safety. This multi-faceted objective optimization ensures that the UAV can efficiently and safely complete the inspection task; by optimizing the initial path through the path smoothing algorithm, the discontinuity in the path can be eliminated, avoiding sharp turns and vibrations of the UAV during flight, and improving the flight smoothness and task execution accuracy.

[0017] A further setting is that the multi-objective reward function in S302 includes an efficiency reward function, a safety reward function, and an emergency task reward function; For the efficiency reward function, the value R of the efficiency reward function is calculated based on the path time consumption and a preset threshold time ; For the safety reward function, for each successful bypass of a high-risk obstacle, the value R of the safety reward function is calculated based on the actual obstacle avoidance distance and the safety distance safe ; For the emergency task reward function, for each inspection of a device in a high-risk state with a health degree less than the second health degree threshold, the value R of the emergency task reward function is calculated based on the health degree of the device in the high-risk state emergency ; Weights are assigned to the efficiency reward function, the safety reward function, and the emergency task reward function respectively, and the value R of the multi-objective reward function is obtained through calculation total .

[0018] By adopting the above technical solutions, the multi-objective reward function can optimize multiple aspects in path planning, including the efficiency, safety of the path, and the response priority of emergency tasks; through this multi-objective optimization, the UAV can not only quickly complete regular inspection tasks, but also prioritize the inspection of high-risk equipment when the equipment health is poor, thereby improving the emergency response ability of the inspection; the safety reward function can effectively guide the UAV to avoid high-risk obstacles and ensure the safety during flight; while the efficiency reward function calculates through the path time consumption and preset thresholds to ensure the optimization of the path, enabling the inspection task to be completed in the shortest time; the emergency task reward function ensures timely response in case of emergencies by prioritizing the handling of high-risk equipment. By assigning weights to each reward function, the priorities of different objectives can be flexibly adjusted according to the actual situation, enabling the UAV to make the most appropriate choice in different task scenarios.

[0019] A further setting is that the S303 is specifically as follows: S3031. Generate multiple candidate paths through the Monte Carlo tree search algorithm, and screen the paths that meet the following conditions: The path length is less than 1.2 times the theoretical shortest path; Cover more than 90% of high-risk equipment; S3032. Calculate the value R of the multi-objective reward function for each screened candidate path total , and select the candidate path with the highest score as the initial inspection path; S3033. Perform cubic Bezier curve interpolation on the waypoint sequence of the initial inspection path to generate a continuous trajectory, and constrain the path curvature to prevent the UAV from making sharp turns; S3034. Dynamically adjust the flight speed of the UAV according to the value R of the path multi-objective reward function total : If the value R of the multi-objective reward function total is greater than the set first function threshold, increase the flight speed of the UAV to the first speed; If the value R of the multi-objective reward function total is between the first function threshold and the second function threshold, the UAV maintains the default flight speed; If the value R of the multi-objective reward function total is less than the second function threshold, then reduce the flight speed of the UAV to the second speed; S3035. If there is equipment in the substation with a health degree less than the third health degree threshold, add a hovering waypoint to the path and trigger an audible and visual alarm.

[0020] By adopting the above technical solutions, multiple candidate paths are generated through the Monte Carlo tree search algorithm, and the optimal path is selected through the multi-objective reward function. This can not only avoid the uncertainty of path selection, but also perform dynamic optimization based on various factors such as equipment health status and environmental risks, ensuring the comprehensiveness and effectiveness of the inspection path; Bezier curve interpolation can generate a smooth path, avoiding sharp turn problems caused by excessive path curvature, which is crucial for the flight stability of the UAV and the accuracy of task execution. Especially in complex environments, it can ensure that the UAV successfully completes the task; the strategy of dynamically adjusting the flight speed enables the UAV to automatically adjust the flight speed according to the value of the multi-objective reward function, thereby ensuring the execution efficiency and safety of different inspection tasks. By flexibly adjusting the flight speed, the UAV can maintain the optimal flight state under different environmental and task requirements.

[0021] A further setting is that the S400 is specifically as follows: S401. If there is a device in the substation with a decreased health level and less than the second health threshold, increase the weight of its emergency task reward function in the multi-objective reward function; if the health level of this device further decreases and is less than the third health threshold, force the UAV to prioritize the inspection and update the path score; S402. If a sudden increase in wind speed is detected, immediately reduce the weight of the safety reward function and recalculate the path.

[0022] By adopting the above technical solutions, dynamically adjusting the weight of the emergency task reward function can ensure that when the health level of the equipment decreases, the UAV prioritizes the inspection of these high-risk devices; through this mechanism, it can effectively respond to sudden equipment failures or potential risks, preventing the impact of faulty equipment from spreading to the entire system; by increasing the weight of the emergency task, the UAV can timely adjust the inspection path, so that when the health level of the equipment further decreases, it can quickly adjust the priority and optimize the path to ensure that critical equipment is inspected in a timely manner.

[0023] A further setting is that after the S400, the following steps are also included: S500. Collect the path tracking error, equipment misdetection rate, and obstacle avoidance failure events in the actual inspection and generate a report.

[0024] By adopting the above technical solutions, the generated report provides detailed inspection records for the inspection personnel, helps to discover potential problems and provides improvement measures. This function can not only improve the inspection efficiency, but also ensure that the health status of the equipment is comprehensively evaluated, thereby improving the management level and safety of substation equipment.

[0025] In summary, the present invention has the following beneficial effects: The innovations of the present invention in equipment health assessment, path planning optimization, and real-time emergency response contribute to improving the efficiency and safety of inspection tasks, reducing the cost of manual intervention, and providing strong technical support for the efficient operation and maintenance of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flowchart of an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The present invention will be further described in detail below with reference to the accompanying drawings.

[0028] As shown in the Figure 1 accompanying drawings; This embodiment discloses a substation route inspection planning method based on deep learning, including the following steps: S100. Obtain substation equipment status data and environmental monitoring data; wherein, the equipment status data includes equipment temperature, vibration frequency, and insulation resistance value, and the environmental monitoring data includes meteorological parameters, light intensity, and obstacle distribution; S200. Perform multi-dimensional feature analysis on the equipment status data based on a preset deep feature extraction model to generate an equipment health assessment vector, and the generated equipment health assessment vector can reflect different states of the equipment; wherein, the deep feature extraction model includes a dual-modal structure of a convolutional neural network and a long short-term memory network for extracting spatial features and time series of the equipment status data; S300. Based on the environmental monitoring data and the equipment health assessment vector, generate multiple candidate paths through a reinforcement learning algorithm, screen an initial inspection path from the multiple candidate paths, and dynamically optimize the initial inspection path by real-time fusing geographic information; S400. According to the UAV positioning data and the equipment status feedback, if it is detected that the health degree is lower than the threshold or the environmental risk exceeds the limit, trigger local path replanning and preferentially navigate to high-risk equipment.

[0029] Among them, in S100, obtaining the substation equipment status data and environmental monitoring data specifically includes: Collect the equipment temperature, vibration frequency, and insulation resistance value through a distributed sensor network, and perform normalization processing on the equipment status data; wherein, the distributed sensor network is deployed at key monitoring nodes of substation equipment and includes an infrared sensor, an acceleration sensor, and an insulation monitoring unit; Obtain meteorological parameters, light intensity, and obstacle distribution through a multi-spectral camera and a lidar carried by a UAV, and perform spatial grid segmentation on the environmental monitoring data; wherein, the granularity of the spatial grid segmentation is adaptively adjusted according to the substation area and the obstacle density.

[0030] Among them, S200 specifically is: S201: Input the device temperature, vibration frequency, and insulation resistance value into the parallel channels of the convolutional neural network respectively, and extract local anomaly features through multi-scale convolutional kernels; S202: Input the local anomaly features and the time series of historical device state data into the long short-term memory network; S203: Perform tensor splicing on the spatial features and time series features, and map them into a device health assessment vector through a fully connected layer; among them, the dimension of the device health assessment vector corresponds one-to-one with the number of substation device types.

[0031] Among them, S201 specifically is: Perform standardization processing on the device temperature data, vibration frequency signal, and insulation resistance value respectively to eliminate the dimension difference; specifically, perform standardization processing on the device temperature data and convert it into a grayscale thermal map of 128×128 pixels, where the grayscale value of each pixel corresponds to the level of the device surface temperature; perform short-time Fourier transform on the vibration frequency signal according to a 0.1-second time window to generate a time-frequency diagram to characterize the change of vibration intensity at different frequencies over time; generate a two-dimensional time series matrix for the insulation resistance value through a sliding window, with the window length configured as 10 seconds and the step size configured as 5 seconds; Input the device temperature data, vibration frequency signal, and insulation resistance value after standardization processing into three independent input channels of the convolutional neural network respectively, with each channel corresponding to one data type, to achieve parallel processing of multi-modal data and generate a local anomaly feature tensor; specifically, input the grayscale thermal map, time-frequency diagram, and time series matrix into three independent channels of the convolutional neural network respectively, and each channel uses 3×3, 5×5, and 7×7 multi-scale convolutional kernels for feature extraction, where: The 3×3 convolutional kernel is used to detect local minute anomalies on the device surface, including regions of sudden temperature rise or instantaneous spikes in the vibration spectrum; The 5×5 convolutional kernel is used to identify associated anomalies between device components, including temperature gradient anomalies or spatial diffusion patterns of vibration fluctuations; The 7×7 convolutional kernel is used to analyze the overall trend, including the uniformity of the temperature distribution across the entire substation or the periodic pattern of vibration; Finally, fuse the multi-scale convolution results through residual connection, and weight the important features using the channel attention mechanism to generate a local anomaly feature tensor.

[0032] S202 specifically is: Arrange the local anomaly features in chronological order as a time series and align the timestamps with the historical device state data; the historical device state data includes records of the device temperature, vibration frequency, and insulation resistance value in the past 24 hours; Add time encoding parameters to historical data, including seasonal factors, day-night markers, and load cycle identifiers; Construct a long short-term memory network, whose input layer receives the concatenated local anomaly features and time encoding parameters, and dynamically adjusts the information flow through the gating mechanisms of the input gate, forget gate, and output gate; The input gate assigns weights according to the importance of the current features. If it is detected that the insulation resistance value drops by more than 5% within the same load cycle, the weight of the features at this time step is increased; The forget gate decides whether to retain or discard information based on the relevance of historical data. If a certain temperature fluctuation is determined to be occasional noise, its long-term impact is reduced; Output the hidden state to characterize the degradation trend of the device, and predict the remaining life based on the Weibull distribution model. The calculation formula is: ; where η is the characteristic life parameter; β is the shape parameter, which is dynamically adjusted by the hidden variable output by the long short-term memory network; t is the time, and R(t) represents the reliability of the device after time t, that is, the probability that the device has not failed from the initial state to time t.

[0033] Specifically, S203 is as follows: Compress the spatial features output by the convolutional neural network into a 512-dimensional vector through global max pooling, compress the time features output by the long short-term memory network into a 256-dimensional vector, perform tensor concatenation on the spatial features output by the convolutional neural network and the time features output by the long short-term memory network to generate a 768-dimensional mixed feature vector, and reduce the dimension to 256 through principal component analysis to eliminate redundancy; construct a fully connected network, including three hidden layers, with the activation function using Leaky ReLU, the dimension of the output layer strictly matching the number of substation equipment types, and constrain the output value within the range of 0.0 - 1.0 through the Sigmoid function to generate a device health assessment vector: If the health degree is greater than the set first health degree threshold, it indicates a normal state; If the health degree is between the first health degree threshold and the second health degree threshold, it indicates the existence of potential risks; If the health degree is less than the second health degree threshold, it indicates a high-risk state and triggers an alarm.

[0034] More specifically, in this embodiment, the first health degree threshold is set to 0.8, and the second health degree threshold is set to 0.5; If the health degree > 0.8, it indicates a normal state and is marked green at the same time; If 0.5 ≤ health degree ≤ 0.8, it indicates potential risks and is marked yellow at the same time; If the health degree < 0.5, it indicates a high-risk state and triggers an alarm, and is marked red at the same time.

[0035] Set dynamic thresholds according to device types, with the temperature threshold for outdoor devices being 5% higher than that for indoor devices and the vibration tolerance for frequently used devices being reduced by 10%.

[0036] Among them, S300 specifically is as follows: S301. Define the three-dimensional perception space of the substation area, including device coordinates, obstacle distribution, and environmental risks; divide the substation area into grid cells of 0.5 m × 0.5 m, and each grid cell is marked as one of the following states: Passable area, indicating that there are no obstacles inside; Low-risk obstacle, indicating a static obstacle; High-risk obstacle, indicating a dynamic obstacle or a high-risk environmental area; among them, dynamic obstacles such as birds, and high-risk environmental areas such as wind speed greater than 10 m / s; S302. Design a multi-objective reward function; S303. Optimize the initial path through a path smoothing algorithm to ensure that the UAV flight trajectory is continuous and conforms to kinematic constraints.

[0037] Among them, the multi-objective reward function in S302 includes an efficiency reward function, a safety reward function, and an emergency task reward function; For the efficiency reward function, based on the path time consumption and a preset threshold T max to calculate the value R of the efficiency reward function time ; if the path time consumption T is less than the preset threshold T max , the value R of the efficiency reward function time = 1 - T / T max ; For the safety reward function, for each successful bypass of a high-risk obstacle, calculate the value R of the safety reward function based on the actual obstacle avoidance distance and the safety distance safe ; for each successful bypass of a high-risk obstacle, the value R of the safety reward function safe = (D safe - d) / D safe , where d is the actual obstacle avoidance distance and D safe is the safety distance, set to 2 m; For the emergency task reward function, for each inspection of a device in a high-risk state with a health degree less than the second health degree threshold, calculate the value R of the emergency task reward function based on the health degree of the device in the high-risk state emergency ; for each inspection of a device in a high-risk state with a health degree less than the second health degree threshold, the value R of the emergency task reward function emergency = 2×(1 - health degree); Assign weights to the efficiency reward function, the safety reward function, and the emergency task reward function respectively, and calculate the value R of the multi-objective reward functiontotal 。

[0038] The value R of the multi-objective reward function total = a 1 R time + a 2 R safe + a 3 R emergency , where a 1 , a 2 and a 3 respectively represent the weights of the efficiency reward function, the safety reward function, and the emergency task reward function in the multi-objective reward function; the initial value R of the multi-objective reward function total = 0.5R time + 0.3R safe + 0.2R emergency 。

[0039] Among them, S303 is specifically as follows: S3031. Generate multiple candidate paths through the Monte Carlo tree search algorithm, and filter the paths that meet the following conditions: The path length is less than 1.2 times the theoretical shortest path; Cover more than 90% of high-risk devices; S3032. Calculate the value R of the multi-objective reward function for each filtered candidate path total , and select the candidate path with the highest score as the initial inspection path; S3033. Perform cubic Bezier curve interpolation on the waypoint sequence of the initial inspection path to generate a continuous trajectory, and constrain the maximum path curvature to 0.1 per meter to prevent the UAV from making sharp turns; S3034. Dynamically adjust the UAV flight speed according to the value R of the path multi-objective reward function total : If the value R of the multi-objective reward function total is greater than the set first function threshold, increase the UAV flight speed to the first speed; If the value R of the multi-objective reward function total is between the first function threshold and the second function threshold, the UAV maintains the default flight speed; If the value R of the multi-objective reward function total is less than the second function threshold, reduce the UAV flight speed to the second speed; More specifically, in this embodiment, the first function threshold is set to 0.8, the second function threshold is set to 0.5, the first speed is set to 3 m / s, the default flight speed is set to 2 m / s, and the second speed is set to 1 m / s; If the value R of the multi-objective reward function totalIf it is greater than 0.8, increase the flight speed of the drone to 3 m / s; If 0.5 ≤ the value R of the multi-objective reward function total ≤ 0.8, the drone maintains the default flight speed of 2 m / s; If the value R of the multi-objective reward function total < 0.5, reduce the flight speed of the drone to 1 m / s.

[0040] S3035: If there is a device in the substation with a health level less than the third health level threshold, add a hovering waypoint to the path, trigger an audible and visual alarm, and at the same time, the drone will stay for 10 seconds to take pictures for evidence; the third health level threshold is set to 0.3.

[0041] Among them, S400 is specifically as follows: S401: If there is a device in the substation with a decreasing health level and less than the second health level threshold, increase the weight of its emergency task reward function to 0.3 in the multi-objective reward function; if the health level of the device further decreases and is less than the third health level threshold, force the drone to prioritize inspection and update the path score; S402: If a sudden increase in wind speed is detected, immediately reduce the weight of the safety reward function to 0.2 and recalculate the path; at the same time, if the detected light intensity is lower than the set light intensity threshold, automatically add a searchlight command to the drone and increase the flight height to 3 meters.

[0042] Among them, after S400, the following steps are also included: S500: Collect the path tracking error, equipment misdetection rate, and obstacle avoidance failure events in the actual inspection and generate a report.

[0043] Embodiment 1 The purpose is to verify the efficient inspection under normal conditions; conduct daily inspections on a certain 500 kV substation. The equipment health levels are generally good, the environmental conditions are stable, the wind speed is 3 m / s, and the light intensity is 800 lux; the goal is to complete the inspection of 120 pieces of equipment within 30 minutes; In S100, the equipment status data is as follows: Equipment temperature: The highest temperature on the surface of transformer A is 52 °C, lower than the threshold of 55 °C, and the temperature of insulator B is 48 °C; Vibration frequency: The vibration amplitude of circuit breaker C is 0.2 mm, within the normal range of 0.1 - 0.3 mm; Insulation resistance value: The resistance value of lightning arrester D is 500 MΩ, within the normal range; The environmental monitoring data is as follows: The electric poles and fences are marked as low-risk obstacles; Meteorological parameters: Wind speed 3 m / s, no precipitation.

[0044] In S200, the convolutional neural network detects local temperature hotspots in transformer A, but the overall distribution is uniform; the long short-term memory network analyzes historical equipment status data and predicts that its remaining life is more than 5 years; The equipment health assessment vector is as follows: The health of transformer A is 0.85, the health of insulator B is 0.78, the health of circuit breaker C is 0.82, and the health of the lightning arrester is 0.93, all indicating normal status and marked in green.

[0045] In S300, the multi-objective reward function R is selected by calculation total The candidate path with the highest value of the multi-objective reward function R is used as the initial inspection path; the calculation process of the value of the multi-objective reward function R for this initial inspection path is as follows: total The calculation process of its value is as follows: The theoretical shortest path L min = 1.2 km, and the path length L = 1.3 km; The value of the efficiency reward function R time = 1 - 28 / 30 ≈ 0.07; There is no obstacle avoidance event, and the value of the safety reward function R safe = 0; There is no high-risk equipment, and the value of the emergency task reward function R emergency = 0; The value of the multi-objective reward function R total = 0.5×0.07 + 0.3×0 + 0.2×0 = 0.035; The unmanned flight is set at a speed of 3 m / s.

[0046] In S400, the unmanned aerial vehicle flies according to the planned path, completes the inspection in 29 minutes, and the path deviation is less than 0.2 meters.

[0047] The result analysis is as follows: The actual time consumption is 51.7% shorter than the traditional average 60-minute manual inspection; The consistency rate of the equipment health assessment and manual review reaches 98%.

[0048] Example 2 To verify the dynamic response in a complex environment, during a certain inspection, a strong wind of 15 m / s and a flock of birds interference are suddenly encountered, and the path needs to be adjusted urgently and high-risk equipment needs to be inspected preferentially; In S100, the equipment status data is as follows: The insulation resistance of the mutual inductor E drops to 150 MΩ; The environmental monitoring data is as follows: Obstacle update: 3 more birds are added, and at the same time the wind speed rises to 15 m / s; Light intensity: Decreased to 200 lux due to cloud cover.

[0049] In S200, The convolutional neural network detected abnormal partial insulation resistance of the mutual inductor E, and the long short-term memory network predicted that its remaining life was less than 6 months; The health degree of the mutual inductor E was 0.25 and was marked in red; In S400, the weight of its emergency task reward function was increased to 0.3, and the weight of the safety reward function was decreased to 0.2; The path was replanned; the Monte Carlo tree search generated a new path around the bird, covering the mutual inductor E, and the path length increased to 1.5 km; at the same time, the flight speed was decreased to 1 m / s, and a hover instruction was added.

[0050] The result analysis is as follows: All birds were successfully avoided, and the minimum obstacle avoidance distance was 1.8 meters; The fault of the mutual inductor E was confirmed in time and was later analyzed as a damaged internal insulation layer, avoiding potential power outages.

[0051] As can be seen from the above Embodiment 1 and Embodiment 2, the present invention shows significant advantages in both conventional and complex scenarios. The method provides reliable technical support for intelligent substation inspection and has broad application prospects.

[0052] This specific embodiment is only an explanation of the present invention and is not a limitation of the present invention. Those skilled in the art can make modifications to this embodiment without creative contributions according to needs after reading this specification, but as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A substation route inspection planning method based on deep learning, characterized in that: The following steps are included: S100, obtaining substation equipment status data and environmental monitoring data; wherein the equipment status data includes equipment temperature, vibration frequency and insulation resistance value, and the environmental monitoring data includes meteorological parameters, light intensity and obstacle distribution; S200, performing multi-dimensional feature analysis on the device status data based on a preset deep feature extraction model to generate a device health assessment vector, where the generated device health assessment vector can reflect different states of the device; wherein the deep feature extraction model includes a bimodal structure of a convolutional neural network and a long short-term memory network, which is used to extract spatial features and time series features of the device status data; S300, based on environmental monitoring data and equipment health assessment vectors, generates multiple candidate paths through reinforcement learning algorithms, selects the initial inspection path from the multiple candidate paths, and dynamically optimizes the initial inspection path by integrating geographic information in real time; S400: Based on the drone positioning data and device status feedback, if it is detected that the health level is lower than the threshold or the environmental risk exceeds the limit, local path replanning is triggered, and navigation to high-risk devices is prioritized.

2. According to a deep learning-based substation route inspection planning method according to claim 1, it is characterized by: In S100, the substation equipment status data and environmental monitoring data are obtained specifically as follows: The distributed sensor network collects the equipment temperature, vibration frequency and insulation resistance value, and normalizes the equipment status data; the distributed sensor network is deployed at the key monitoring nodes of the substation equipment, including infrared sensors, acceleration sensors and insulation monitoring units; Meteorological parameters, light intensity and obstacle distribution are obtained through the multispectral camera and lidar carried by the drone.

3. According to a deep learning-based substation route inspection planning method according to claim 2, it is characterized by: The S200 is specifically: S201, inputting the device temperature, vibration frequency and insulation resistance value into the parallel channels of the convolutional neural network respectively, and extracting local abnormal features through a multi-scale convolution kernel; S202, inputting the time series of local abnormal features and historical device status data into a long short-term memory network; S203: concatenate the spatial features and the time series features into tensors, and map them into a device health assessment vector through a fully connected layer.

4. A substation route inspection planning method based on deep learning according to claim 3, characterized in that: The S201 is specifically: Standardize the equipment temperature data, vibration frequency signal and insulation resistance value to eliminate dimensional differences; The standardized device temperature data, vibration frequency signal and insulation resistance value are input into three independent input channels of the convolutional neural network respectively. Each channel corresponds to a data type to achieve multi-modal data parallel processing to generate local abnormal feature tensors. The S202 is specifically: arranging the local abnormal features into a time series in chronological order, and aligning the timestamps with the historical device status data; Add time coding parameters to historical data; Construct a long short-term memory network, whose input layer receives the concatenated local abnormal features and time encoding parameters, and dynamically adjusts the information flow through the gating mechanism of input gate, forget gate and output gate; The S203 is specifically as follows: tensor splicing the spatial features output by the convolutional neural network and the temporal features output by the long short-term memory network, constructing a fully connected network, and constraining the output value within the range of 0.0-1.0 to generate a device health assessment vector: If the health level is greater than the set first health level threshold, it indicates a normal state; If the health level is between the first health level threshold and the second health level threshold, it indicates that there is a potential risk; If the health level is lower than the second health level threshold, it indicates a high-risk state and triggers an alarm.

5. A substation route inspection planning method based on deep learning according to claim 4, characterized in that: The S300 is specifically: S301. Define the three-dimensional perception space of the substation area, including equipment coordinates, obstacle distribution and environmental risks; divide the substation area into grid units, and mark each grid unit as one of the following states: A traversable area means there are no obstacles within it; Low-risk obstacles, which means stationary obstacles; High-risk obstacles, indicating dynamic obstacles or high-risk areas in the environment; S302, designing a multi-objective reward function; S303, optimizing the initial path through a path smoothing algorithm to ensure that the UAV flight trajectory is continuous and complies with kinematic constraints.

6. A substation route inspection planning method based on deep learning according to claim 5, characterized in that: The multi-objective reward function in S302 includes an efficiency reward function, a safety reward function and an emergency task reward function; For the efficiency reward function, the value R of the efficiency reward function is calculated based on the path time and the preset threshold time ; For the safety reward function, each time a high-risk obstacle is successfully bypassed, the value of the safety reward function R is calculated based on the actual obstacle avoidance distance and the safety distance. safe ; For the emergency task reward function, each time a device in a high-risk state whose health is less than the second health threshold is inspected, the value R of the emergency task reward function is calculated based on the health of the device in the high-risk state. emergency ; Assign weights to the efficiency reward function, safety reward function, and emergency task reward function respectively, and calculate the value R of the multi-objective reward function total .

7. A substation route inspection planning method based on deep learning according to claim 6, characterized in that: The S303 is specifically: S3031. Generate multiple candidate paths through the Monte Carlo tree search algorithm, and select paths that meet the following conditions: The path length is less than 1.2 times the theoretical shortest path; Covering more than 90% of high-risk equipment; S3032. Calculate the value R of the multi-objective reward function of each selected candidate path total , select the candidate path with the highest score as the initial inspection path; S3033, performing cubic Bezier curve interpolation on the waypoint sequence of the initial inspection path to generate a continuous trajectory, and constraining the path curvature to prevent the UAV from making sharp turns; S3034, according to the value R of the path multi-objective reward function total Dynamically adjust the flight speed of the drone: If the value of the multi-objective reward function R total If it is greater than the set first function threshold, the flight speed of the drone is increased to the first speed; If the value of the multi-objective reward function R total When the drone is between the first function threshold and the second function threshold, the drone maintains the default flight speed; If the value of the multi-objective reward function R total When it is less than the second function threshold, the UAV flight speed is reduced to the second speed; S3035. If there is a device in the substation whose health is less than the third health threshold, a hovering waypoint is added to the path and an audible and visual alarm is triggered.

8. A substation route inspection planning method based on deep learning according to claim 7, characterized in that: The S400 is specifically: S401. If there is a device in the substation whose health has decreased and is less than the second health threshold, increase the weight of its emergency task reward function in the multi-objective reward function; if the health of the device further decreases and is less than the third health threshold, force the drone to give priority to inspection and update the path score; S402: If a sudden increase in wind speed is detected, the weight of the safety reward function is reduced in real time, and the path is recalculated.

9. A substation route inspection planning method based on deep learning according to claim 8, characterized in that: After the step S400, the following steps are also included: S500: Collect path tracking errors, equipment false detection rates, and obstacle avoidance failure events during actual inspections, and generate a report.

Images