Intelligent positioning and risk control method of power operation environment based on three-dimensional model

Abstract

The present invention relates to an intelligent positioning and risk control method for an electric power operation environment based on a three-dimensional model, comprising the following steps: S1: constructing a three-dimensional model of an electric power operation environment, and spatially stratifying and annotating the three-dimensional model to form a three-dimensional panoramic operation map with attributes; S2: marking the positions of operators and equipment in real time to generate dynamic positioning in the electric power operation scene; S3: recording and predicting the trajectory of operators, and warning whether they enter a safe area or a dangerous area; and using a camera to collect operator images, identifying the irregular behavior of operators in real time through an AI algorithm, and generating an operator behavior risk index based on the detection results; S4: training a risk prediction model based on sensor data, detecting possible risk events, and obtaining the operation risk index of the equipment; S5: calculating the comprehensive risk index in real time. The present invention effectively improves the efficiency of safety management while ensuring the comprehensive safety of personnel, equipment and environment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent management and control, and in particular to an intelligent positioning and risk management method for an electric power operation environment based on a three-dimensional model. Background Art

[0002] In the power operation environment, the operation scenes are complex and changeable, and the potential safety risks are diverse, covering equipment operation risks such as high-voltage arc discharge, equipment overload operation, abnormal mechanical vibration, and behavioral risks such as operators not operating according to specifications and entering dangerous areas. In addition, power equipment is mostly arranged outdoors, in substations, or along transmission and distribution lines. Most of these environments have large spatial scales, scattered operation areas, and contain potential dangerous areas (such as high-voltage areas, overloaded equipment, and damaged insulation areas), which makes real-time monitoring of equipment and operators particularly difficult. With the parallel operation of multiple devices and the collaborative work of multiple positions during the operation process, the operation site presents complex and changeable characteristics due to the dynamic adjustment of work tasks and environmental changes, which increases the difficulty of safety management. Traditional monitoring and risk control methods (such as manual inspections and manual records) are not only inefficient, but also easily lead to delayed or even missed risk information, making it difficult to provide timely warnings and interventions. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide an intelligent positioning and risk management method for the power operation environment based on a three-dimensional model, which can comprehensively perceive and predict the behavior trajectory of operators, equipment status and environmental risks, thereby achieving more efficient risk management and more accurate safety warnings.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for intelligent positioning and risk control of power operation environment based on a three-dimensional model, comprising the following steps:

[0006] S1: construct a three-dimensional model of the power operation environment, and perform spatial stratification and equipment annotation on the three-dimensional model to form a three-dimensional panoramic operation map with attributes;

[0007] S2: Mark the positions of operators and equipment in real time in a three-dimensional panoramic operation map with attributes, and generate dynamic visual positioning in the power operation scene;

[0008] S3: Record and predict the trajectory of operators, and warn whether they have entered a safe zone or a dangerous zone; use cameras to collect images of operators, use AI algorithms to identify irregular behaviors of operators in real time, and generate a behavioral risk index for operators based on the detection results;

[0009] S4: Train the risk prediction model based on sensor data, detect possible risk events, and obtain the operation risk index of the equipment;

[0010] S5: Dynamically integrate the operator behavior risk and equipment operation risk, calculate the comprehensive risk index in real time, and evaluate the comprehensive risk of the power operation environment in real time.

[0011] Furthermore, S1 is specifically:

[0012] S11: Use LiDAR equipment to capture high-precision point cloud data in the power operation environment. The point cloud data includes power facilities and the operation site environment, and retains the three-dimensional spatial geometric information of the environment;

[0013] S12: preprocessing the original point cloud data;

[0014] S13: Based on the pre-processed point cloud data, a complete three-dimensional model of the power operation environment is generated using point cloud processing software. The oblique photography data is fused with the point cloud through a structured method to complete the visual blind spots with poor modeling accuracy;

[0015] S14: According to the safety regulations of power operations, the working environment is divided into high-voltage area, approach area and low-voltage area, a spatial grid is constructed in the generated three-dimensional model, and the partition boundaries are drawn using an algorithm;

[0016] S15: Finally, the high pressure area, the approach area and the low pressure area are presented in the three-dimensional model with different colors or transparencies to obtain a three-dimensional panoramic operation map with attributes.

[0017] Further, preprocessing is as follows:

[0018] Align the point clouds obtained from multiple scans. For point cloud groups with large initial position deviations, use the NDT method for initial positioning and the ICP algorithm for precise registration.

[0019] Downsample the point cloud with local density greater than the threshold to reduce the amount of point cloud data, and use the SOR algorithm to remove isolated points and noise points;

[0020] High-voltage cables are identified and extracted from the point cloud, and the cable trajectories are identified by fitting straight lines using RANSAC. Strong constraints are applied and the cable information is stored as an independent data layer for further partitioning.

[0021] Furthermore, the point clouds obtained from multiple scans are aligned. For point cloud groups with large deviations in the starting positions, the NDT method is used for initial positioning, and the ICP algorithm is used for precise registration, as follows:

[0022] The NDT method creates a probability density function by counting the distribution characteristics of the point cloud in the area, and completes the initial matching by aligning the target point cloud and the source point cloud. First, the point cloud is divided into a three-dimensional grid V of a fixed size. Assume that the points in each grid obey a Gaussian distribution, and calculate the grid center point u and the covariance matrix Σ:

[0023] ;

[0024] Among them, the subscript Indicates the grid number; For Grid The points; For Grid The number of midpoints; the superscript T indicates transposition;

[0025] Define the cost function for point q in each grid S ( q,V ):

[0026]

[0027] Through the gradient descent method, the point-to-grid distribution matching deviation between the target point cloud and the source point cloud is minimized and the transformation matrix is ​​optimized;

[0028] The ICP algorithm aligns the closest points of two point clouds through iterative optimization. For each point p in the source point cloud s , find the nearest point p in the target point cloud t :

[0029] ;

[0030] Among them, T is the target point cloud set; Represents the source point p s and the Euclidean distance between the target point p;

[0031] After the matching is completed, the closest point pair set {(p s,i ,p t,i )}, where i=1,2,…,N, N is the number of matching point pairs;

[0032] According to the closest point matching pair, calculate the transformation matrix T k+1 :

[0033] ;

[0034] According to the obtained T k+1 The transformation updates the source point cloud and iterates until convergence.

[0035] Furthermore, the point clouds with local density greater than the threshold are downsampled to reduce the amount of point cloud data, and the SOR algorithm is used to remove isolated points and noise points, as follows:

[0036] Set the grid dimension to l x ,l y ,l z , map the point cloud coordinates to grid coordinates;

[0037] Place the point into the specified voxel grid and record all the points in the voxel;

[0038] For each voxel grid, retain one of the points as a representative to achieve downsampling;

[0039] For the i-th point in the point cloud array p i , calculate the mean Euclidean distance to the k nearest points in its neighborhood:

[0040] ;

[0041] in, d i For point p i The neighborhood mean distance of ; p j For point p i The jth point in the neighborhood of ;

[0042] Calculate the mean distance μ of all points d and standard deviation σ d ;

[0043] Eliminate points whose distance distribution exceeds the set threshold:

[0044] ;

[0045] Among them, λ is the scale parameter.

[0046] Furthermore, S2 is specifically:

[0047] S21: Use the GPS positioning module to collect the real-time location information of the equipment; deploy environmental sensors and equipment status sensors to monitor the working environment in real time;

[0048] S22: Operators are equipped with wearable devices with IMU to ensure high-precision positioning in dynamic scenes. Multi-source fusion positioning method is used to integrate GPS and IMU to achieve high-precision real-time positioning of operators. For indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and 3D models.

[0049] S23: Integrate positioning information based on a multi-source fusion algorithm, map the positioning results to a three-dimensional panoramic operation map with attributes, and perform real-time rendering in the three-dimensional panoramic operation map with attributes, dynamically mark personnel and equipment, and display attributes and status.

[0050] Furthermore, a multi-source fusion positioning method is used to integrate GPS and IMU to perform high-precision real-time positioning of operators. For indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and three-dimensional models, as follows:

[0051] Outdoor scenarios: Use GPS to provide global absolute coordinates, and IMU to perform high-frequency position compensation to handle short-term signal loss issues;

[0052] GPS and IMU data are processed through extended Kalman filter (EKF) to overcome GPS jitter.

[0053] Let the state of the system be x t Including the operator's current position, speed and posture;

[0054] The acceleration and angular velocity obtained from the IMU are used to make short-term inertial navigation predictions to obtain the predicted values. :

[0055] ;

[0056] Among them, F is the state transfer matrix; G is the control input matrix; u t The acceleration and angular velocity matrices obtained by IMU;

[0057] Position update formula:

[0058] ;

[0059] Among them, Δt is the time step; a t v is the vector acceleration value measured by IMU; t is the current speed of the operator;

[0060] Speed ​​update formula:

[0061] ;

[0062] Indoor scene:

[0063] IMU outputs acceleration and angular velocity in real time; the camera's continuous frame image stream obtains environmental features through SLAM;

[0064] Through IMU state prediction, the next position and posture are calculated and passed to SLAM as the initial value;

[0065] Use feature point matching and PnP to solve the inter-frame pose changes;

[0066] Combine the IMU and SLAM results, reduce drift through filtering, and output the final position and attitude.

[0067] Furthermore, S3 is specifically:

[0068] Build a trajectory prediction model based on LSTM to predict future trajectory sequences based on the operator's current and historical trajectory sequences;

[0069] According to the spatial stratification of the three-dimensional panoramic operation map, the dangerous area D and the safe area S are constructed, and the high-pressure area is regarded as the dangerous area D, the low-pressure area is regarded as the safe area S, and the approach area is regarded as the transition area D. transition ;

[0070] Use the geometric conflict detection R-Tree to check whether the current trajectory point or the predicted trajectory point enters the danger zone:

[0071] Danger zone D, approach zone D transition , and the safety zone S are stored as a bounding-box, and an R-Tree index tree is constructed;

[0072] Each leaf node stores a representation of the bounding box of a spatial object:

[0073] Bounding Box: [(X min ,X max ), (Y min ,Y max ), (Z min ,Z max )];

[0074] Among them, (X min ,X max ) is the minimum and maximum value on the x-axis; (Y min ,Y max ) are the minimum and maximum values ​​on the y-axis; (Z min ,Z max ) are the minimum and maximum values ​​on the x-axis;

[0075] The current or predicted trajectory point (x, y, z) performs a quick search on the R-Tree index tree; determines whether the point is within the stored region boundary:

[0076] if (X min ≤x≤X max )∧(Y min ≤y≤Y max )∧(Z min ≤z≤Z max );

[0077] Use the target detection model to classify and regionally match operators and equipment, and identify irregular behaviors of operators;

[0078] Based on posture estimation key point extraction and angle calculation, analyze whether the operator has any dangerous actions.

[0079] Furthermore, S4 is specifically as follows: collecting equipment temperature, current, voltage, and operating data through sensors and preliminarily processing them through edge computing modules; denoising, feature extraction, and time series construction of sensor data; using trained machine learning to perform real-time reasoning and risk prediction to obtain prediction results, and obtaining risk levels based on the prediction results.

[0080] Furthermore, the operator behavior risk and equipment operation risk are dynamically integrated to calculate the comprehensive risk index in real time and evaluate the comprehensive risk of the power operation environment in real time, as follows:

[0081] Operator Behavior Risk Index (BRI):

[0082] ;

[0083] in, For the The weight of each behavioral risk factor; For the The real-time detection results of behavioral risk events; t is the current moment; is the number of irregular behavior types;

[0084] Equipment operation risk index ERI (t):

[0085] ;

[0086] in, For the The weight of each equipment operation risk factor; For the The operation risk value of each equipment parameter; M is the number of operation risk types;

[0087] The comprehensive risk index CRI (t) is calculated by dynamic weighting:

[0088] ；

[0089] in, α and β is the weight coefficient.

[0090] The present invention has the following beneficial effects:

[0091] 1. The present invention constructs the power operation environment into a three-dimensional model, and combines real-time positioning, equipment sensor data and AI analysis to comprehensively perceive and predict the behavior trajectory of operators, equipment status and environmental risks, thereby achieving more efficient risk management and more accurate safety warnings;

[0092] 2. The present invention can efficiently optimize point cloud data through uniform thinning (Voxel downsampling) and SOR algorithm, improve data calculation efficiency and structure retention accuracy, and realize fast and accurate point cloud registration through the combination of NDT+ICP, effectively optimizing the three-dimensional modeling and data quality of the power operation environment; further constructing a high-precision, visual, dynamic and real-time updated three-dimensional panoramic operation map with attributes, laying a technical foundation for risk management and safety monitoring of the power operation environment;

[0093] 3. The present invention utilizes multi-source fusion technology and combines it with a three-dimensional environmental model to provide operators with high-precision, real-time positioning services in dynamic scenes. It can not only achieve precise outdoor positioning, but also achieve stable and reliable continuous positioning in complex occlusion or indoor environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0095] The following is combined with Figure 1 The present invention is further described in detail with specific embodiments:

[0096] A method for intelligent positioning and risk control of power operation environment based on a three-dimensional model, comprising the following steps:

[0097] S1: Build a three-dimensional model of the power operation environment, and perform spatial stratification (high-voltage area, low-voltage area, proximity area) and equipment annotation on the three-dimensional model to form a three-dimensional panoramic operation map with attributes;

[0098] S2: Mark the positions of operators and equipment in real time in a three-dimensional panoramic operation map with attributes, and generate dynamic visual positioning in the power operation scene;

[0099] S3: Record and predict the trajectory of operators, and warn whether they enter the safe zone or the dangerous zone; use cameras to collect images of operators, use AI algorithms to identify irregular behaviors of operators in real time (such as not wearing safety tools, entering dangerous areas, etc.), and generate behavioral risk indexes for operators based on the detection results;

[0100] S4: Train the risk prediction model based on sensor data, detect possible risk events, and obtain the operation risk index of the equipment;

[0101] S5: Dynamically integrate the operator behavior risk and equipment operation risk, calculate the comprehensive risk index in real time, and evaluate the comprehensive risk of the power operation environment in real time.

[0102] In this embodiment, S1 is specifically:

[0103] S11: Use LiDAR equipment to capture high-precision point cloud data in the power operation environment. The point cloud data includes power facilities (poles, towers, transformers, etc.) and the work site environment (terrain, buildings, obstacles), and retains the three-dimensional spatial geometric information of the environment;

[0104] S12: preprocessing the original point cloud data;

[0105] S13: Based on the pre-processed point cloud data, point cloud processing software (such as "BentleyContextCapture" or "Pix4D") is used to generate a complete 3D model of the power operation environment. The oblique photography data is fused with the point cloud through the structured method (SfM) to complete the visual blind spots with poor modeling accuracy.

[0106] S14: According to the safety regulations of power operations, the working environment can be divided into high-voltage area, proximity area and low-voltage area. High-voltage area (dangerous area): such as the high-voltage area near the wire, within the standard safety distance of the power grid. Low-voltage area (operating area): such as box-type transformers, switch cabinets and other areas. Proximity area (protection area): an area close to the equipment but still within the scope of the operating safety regulations; construct a spatial grid in the generated three-dimensional model, and use an algorithm to draw the partition boundaries;

[0107] S15: Finally, the high pressure area, the approach area and the low pressure area are presented in the three-dimensional model with different colors or transparencies to obtain a three-dimensional panoramic operation map with attributes.

[0108] In this embodiment, the preprocessing is as follows:

[0109] Align the point clouds obtained from multiple scans. For point cloud groups with large initial position deviations, use the NDT method for initial positioning and the ICP algorithm for precise registration.

[0110] Downsample (uniformly thin) point clouds with local density greater than the threshold to reduce the amount of point cloud data, and use the SOR algorithm to remove isolated points (such as small vegetation or floating debris) and noise points;

[0111] Identify and extract high-voltage cables in the point cloud, identify cable trajectories by fitting straight lines using RANSAC, and apply strong constraints, such as cables always extending from the tops of adjacent towers, to improve extraction efficiency. Store cable information as an independent data layer for further partitioning.

[0112] In this embodiment, the point clouds obtained by multiple scans are aligned. For the point cloud group with large initial position deviation, the NDT method is used for initial positioning, and the ICP algorithm is used for precise registration, as follows:

[0113] The NDT method creates a probability density function (PDF) by counting the distribution characteristics of the point cloud in the area, and completes the initial matching by aligning the target point cloud and the source point cloud. First, the point cloud is divided into a three-dimensional grid V of a fixed size. Assume that the points in each grid obey a Gaussian distribution, and calculate the grid center point u and the covariance matrix Σ:

[0114] ;

[0115] Among them, the subscript Indicates the grid number; For Grid The points; For Grid The number of midpoints; the superscript T indicates transposition;

[0116] Define the cost function for point q in each grid S ( q,V ):

[0117]

[0118] Through the gradient descent method, the point-to-grid distribution matching deviation between the target point cloud and the source point cloud is minimized and the transformation matrix is ​​optimized;

[0119] The ICP algorithm aligns the closest points of two point clouds through iterative optimization. For each point p in the source point cloud s , find the nearest point p in the target point cloud t :

[0120] ;

[0121] Where T is the target point cloud; Represents the source point p s and the Euclidean distance between the target point p;

[0122] After the matching is completed, the closest point pair set {(p s,i ,p t,i )}, where i=1,2,…,N, N is the number of matching point pairs;

[0123] According to the closest point matching pair, calculate the transformation matrix T k+1 :

[0124] ;

[0125] According to the obtained T k+1 The transformation updates the source point cloud and iterates until convergence.

[0126] In this embodiment, the point cloud with a local density greater than a threshold is downsampled (uniformly thinned) to reduce the amount of point cloud data, and the SOR algorithm is used to remove isolated points (such as small vegetation or floating debris) and noise points, as follows:

[0127] Set the grid dimension to l x ,l y ,l z , mapping point cloud coordinates to grid coordinates;

[0128] Place the point into the specified voxel grid and record all the points in the voxel;

[0129] For each voxel grid, retain one of the points (such as the point at the center of the voxel, or a randomly selected point) as a representative to achieve downsampling;

[0130] For the i-th point in the point cloud array p i , calculate the mean Euclidean distance to the k nearest points in its neighborhood:

[0131] ;

[0132] in, d i For point p i The neighborhood mean distance of ; p j For point p i The jth point in the neighborhood of ;

[0133] Calculate the mean distance μ of all points d and standard deviation σ d ;

[0134] Eliminate points whose distance distribution exceeds the set threshold:

[0135] ;

[0136] Among them, λ is the scale parameter.

[0137] In this embodiment, step S2 is specifically as follows:

[0138] S21: Use the GPS positioning module to collect the real-time location information of the equipment; deploy environmental sensors (temperature and humidity, wind speed, air pressure, etc.) and equipment status sensors (current, voltage, vibration, temperature, etc.) to monitor the working environment in real time;

[0139] S22: Operators are equipped with wearable devices with IMU to ensure high-precision positioning in dynamic scenes. Multi-source fusion positioning method is used to integrate GPS and IMU to achieve high-precision real-time positioning of operators. For indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and 3D models.

[0140] S23: Integrate positioning information based on a multi-source fusion algorithm, map the positioning results to a three-dimensional panoramic operation map with attributes, and perform real-time rendering in the three-dimensional panoramic operation map with attributes, dynamically mark personnel and equipment, and display attributes and status.

[0141] In this embodiment, a multi-source fusion positioning method is used to perform high-precision real-time positioning of operators through the integration of GPS and IMU; for indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and three-dimensional models, as follows:

[0142] Outdoor scenarios: Use GPS to provide global absolute coordinates, and IMU to perform high-frequency position compensation to handle short-term signal loss issues;

[0143] GPS and IMU data are processed through extended Kalman filter (EKF) to overcome GPS jitter.

[0144] Let the state of the system be x t Including the operator's current position, speed and posture;

[0145] The acceleration and angular velocity obtained from the IMU are used to make short-term inertial navigation predictions to obtain the predicted values. :

[0146] ;

[0147] Among them, F is the state transfer matrix; G is the control input matrix; u t The acceleration and angular velocity matrices obtained by IMU;

[0148] Position update formula:

[0149] ;

[0150] Among them, Δt is the time step; a t v is the vector acceleration value measured by IMU; t is the current speed of the operator;

[0151] Speed ​​update formula:

[0152] ;

[0153] Indoor scene:

[0154] IMU outputs acceleration and angular velocity in real time; the camera's continuous frame image stream obtains environmental features through SLAM;

[0155] Through IMU state prediction, the next position and posture are calculated and passed to SLAM as the initial value;

[0156] Use feature point matching and PnP to solve the inter-frame pose changes;

[0157] Combine the IMU and SLAM results, reduce drift through filtering, and output the final position and attitude.

[0158] In this embodiment, S3 is specifically:

[0159] Build a trajectory prediction model based on LSTM to predict future trajectory sequences based on the operator's current and historical trajectory sequences;

[0160] According to the spatial stratification of the three-dimensional panoramic operation map, the dangerous area D and the safe area S are constructed, and the high-pressure area is regarded as the dangerous area D, the low-pressure area is regarded as the safe area S, and the approach area is regarded as the transition area D. transition ;

[0161] Use geometric conflict detection (R-Tree) to check whether the current trajectory point or the predicted trajectory point enters the danger zone:

[0162] Danger zone D, approach zone D transition , and the safety zone S are stored as a bounding box (Bounding-Box), and an R-Tree index tree is constructed;

[0163] Each leaf node stores a representation of the bounding box of a spatial object:

[0164] Bounding Box: [(X min ,X max ), (Y min ,Y max ), (Z min ,Z max )];

[0165] Among them, (X min ,X max ) is the minimum and maximum value on the x-axis; (Y min ,Y max ) are the minimum and maximum values ​​on the y-axis; (Z min ,Z max ) are the minimum and maximum values ​​on the x-axis;

[0166] The current or predicted trajectory point (x, y, z) performs a quick search on the R-Tree index tree; determines whether the point is within the stored region boundary:

[0167] if (X min ≤x≤X max )∧(Y min ≤y≤Y max )∧(Z min ≤z≤Z max );

[0168] Use the target detection model to classify and regionally match operators and equipment, and identify irregular behaviors of operators;

[0169] Based on posture estimation key point extraction and angle calculation, analyze whether the operator has any dangerous actions.

[0170] In this embodiment, S4 specifically includes: collecting device temperature, current, voltage, and operating data through sensors and preliminarily processing them through the edge computing module; denoising, feature extraction, and time series construction of sensor data; using trained machine learning to perform real-time reasoning and risk prediction to obtain prediction results, and obtaining risk levels based on the prediction results.

[0171] In this embodiment, the operator behavior risk and equipment operation risk are dynamically integrated, the comprehensive risk index is calculated in real time, and the comprehensive risk of the power operation environment is evaluated in real time, as follows:

[0172] Operator Behavior Risk Index (BRI):

[0173] ;

[0174] in, For the The weight of each behavioral risk factor; For the The real-time detection results of behavioral risk events; t is the current moment; is the number of irregular behavior types;

[0175] Equipment operation risk index ERI (t):

[0176] ;

[0177] in, For the The weight of each equipment operation risk factor; For the The operation risk value of each equipment parameter; M is the number of operation risk types;

[0178] The comprehensive risk index CRI (t) is calculated by dynamic weighting:

[0179] ；

[0180] in, α and β is the weight coefficient.

[0181] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0182] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0183] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0184] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0185] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for intelligent positioning and risk control of power operation environment based on a three-dimensional model, characterized in that: The following steps are involved: S1: construct a three-dimensional model of the power operation environment, and perform spatial stratification and equipment annotation on the three-dimensional model to form a three-dimensional panoramic operation map with attributes; S2: Mark the positions of operators and equipment in real time in a three-dimensional panoramic operation map with attributes, and generate dynamic visual positioning in the power operation scene; S3: Record and predict the trajectory of operators, and warn whether they enter the safe zone or the dangerous zone; use cameras to collect images of operators, use AI algorithms to identify irregular behaviors of operators in real time, and generate behavioral risk indexes for operators based on the detection results; S4: Train the risk prediction model based on sensor data, detect possible risk events, and obtain the operation risk index of the equipment; S5: Dynamically integrate the operator behavior risk and equipment operation risk, calculate the comprehensive risk index in real time, and evaluate the comprehensive risk of the power operation environment in real time; S1 is specifically: S11: Use LiDAR equipment to capture high-precision point cloud data in the power operation environment. The point cloud data includes power facilities and the operation site environment, and retains the three-dimensional spatial geometric information of the environment; S12: preprocessing the original point cloud data; S13: Based on the pre-processed point cloud data, a complete three-dimensional model of the power operation environment is generated using point cloud processing software. The oblique photography data is fused with the point cloud through a structured method to complete the visual blind spots with poor modeling accuracy; S14: According to the safety regulations of power operations, the working environment is divided into high-voltage area, approach area and low-voltage area, a spatial grid is constructed in the generated three-dimensional model, and the partition boundaries are drawn using an algorithm; S15: finally presenting the high pressure area, the approach area and the low pressure area in different colors or transparencies in the three-dimensional model to obtain a three-dimensional panoramic operation map with attributes; Preprocessing, as follows: Align the point clouds obtained from multiple scans. For point cloud groups whose starting position deviation is greater than the preset value, use the NDT method for initial positioning and the ICP algorithm for precise registration. Downsample the point cloud with local density greater than the threshold to reduce the amount of point cloud data, and use the SOR algorithm to remove isolated points and noise points; Identify and extract high-voltage cables in the point cloud, identify cable trajectories by fitting straight lines using RANSAC, and apply strong constraints to store cable information as an independent data layer for further partitioning; The point clouds obtained by multiple scans are aligned. For the point cloud group whose starting position deviation is greater than the preset value, the NDT method is used for initial positioning, and the ICP algorithm is used for precise registration, as follows: The NDT method creates a probability density function by statistically analyzing the distribution characteristics of the point cloud in the area, and completes the initial matching by aligning the target point cloud and the source point cloud. First, the point cloud is divided into a three-dimensional grid V of a fixed size, and the points in each grid are assumed to obey the Gaussian distribution. The center point of the grid is calculated. and the covariance matrix : ； Among them, the subscript Indicates the grid number; For Grid The points; For Grid The number of midpoints; the superscript T indicates transposition; Define the cost function for point q in each grid S ( q,V ): Through the gradient descent method, the point-to-grid distribution matching deviation between the target point cloud and the source point cloud is minimized and the transformation matrix is ​​optimized; The ICP algorithm aligns the closest points of two point clouds through iterative optimization. For each point p in the source point cloud s , find the nearest point p in the target point cloud t : ； Among them, T is the target point cloud set; Represents the source point p s and the Euclidean distance between the target point p; After the matching is completed, the closest point pair set {(p s,i ,p t,i )}, where i=1,2,…,N, N is the number of matching point pairs; According to the closest point matching pair, calculate the transformation matrix T k+1 : ； According to the obtained T k+1 The transformation updates the source point cloud and iterates until convergence.

2. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 1 is characterized in that: The point cloud with a local density greater than a threshold is downsampled to reduce the amount of point cloud data, and the SOR algorithm is used to remove isolated points and noise points, as follows: Set the grid dimension to l x ,l y ,l z , map the point cloud coordinates to grid coordinates; Place the point into the specified voxel grid and record all the points in the voxel; For each voxel grid, retain one of the points as a representative to achieve downsampling; For the i-th point in the point cloud array p i , calculate the mean Euclidean distance to the k nearest points in its neighborhood: ； in, d i For point p i The neighborhood mean distance of ; p j For point p i The jth point in the neighborhood of ; Calculate the mean distance μ of all points d and standard deviation σ d ; Eliminate points whose distance distribution exceeds the set threshold: ； Among them, λ is the scale parameter.

3. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 1 is characterized in that: The S2 is specifically: S21: Use the GPS positioning module to collect the real-time location information of the equipment; deploy environmental sensors and equipment status sensors to monitor the working environment in real time; S22: Operators are equipped with wearable devices with IMU to ensure high-precision positioning in dynamic scenes. Multi-source fusion positioning method is used to integrate GPS and IMU to achieve high-precision real-time positioning of operators. For indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and 3D models. S23: Integrate positioning information based on a multi-source fusion algorithm, map the positioning results to a three-dimensional panoramic operation map with attributes, and perform real-time rendering in the three-dimensional panoramic operation map with attributes, dynamically mark personnel and equipment, and display attributes and status.

4. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 3 is characterized in that: The multi-source fusion positioning method is used to integrate GPS and IMU to perform high-precision real-time positioning of operators; for indoor or blocked scenes, visual SLAM is used to assist positioning through cameras and three-dimensional models, as follows: Outdoor scenarios: Use GPS to provide global absolute coordinates, and IMU to perform high-frequency position compensation to handle short-term signal loss issues; GPS and IMU data are processed through extended Kalman filter (EKF) to overcome GPS jitter. Let the state of the system be x t Including the operator's current position, speed and posture: The acceleration and angular velocity obtained from the IMU are used to make short-term inertial navigation predictions to obtain the predicted values. : ； Among them, F is the state transfer matrix; G is the control input matrix; u t The acceleration and angular velocity matrices obtained by IMU; Position update formula: ； Among them, Δt is the time step; a t v is the vector acceleration value measured by IMU; t is the current speed of the operator; Speed ​​update formula: ； Indoor scene: IMU outputs acceleration and angular velocity in real time; the camera's continuous frame image stream obtains environmental features through SLAM; Through IMU state prediction, the next position and posture are calculated and passed to SLAM as the initial value; Use feature point matching and PnP to solve the inter-frame pose changes; Combine the IMU and SLAM results, reduce drift through filtering, and output the final position and attitude.

5. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 1 is characterized in that: The S3 is specifically: Build a trajectory prediction model based on LSTM to predict future trajectory sequences based on the operator's current and historical trajectory sequences; According to the spatial stratification of the three-dimensional panoramic operation map, the dangerous area D and the safe area S are constructed, and the high-pressure area is regarded as the dangerous area D, the low-pressure area is regarded as the safe area S, and the approach area is regarded as the transition area D. transition ; Use the geometric conflict detection R-Tree to check whether the current trajectory point or the predicted trajectory point enters the danger zone: Danger zone D, approach zone D transition , and the safety zone S are stored as a bounding-box, and an R-Tree index tree is constructed; Each leaf node stores a representation of the bounding box of a spatial object: Bounding Box: [(X min ,X max ),(Y min ,Y max ),(Z min ,Z max )]; Among them, (X min ,X max ) is the minimum and maximum value on the x-axis; (Y min ,Y max ) are the minimum and maximum values ​​on the y-axis; (Z min ,Z max ) are the minimum and maximum values ​​on the x-axis; The current or predicted trajectory point (x, y, z) performs a quick search on the R-Tree index tree; determines whether the point is within the stored region boundary: if (X min ≤x≤X max )∧(Y min ≤y≤Y max )∧(Z min ≤z≤Z max ); Use the target detection model to classify and regionally match operators and equipment, and identify irregular behaviors of operators; Based on posture estimation key point extraction and angle calculation, analyze whether the operator has any dangerous actions.

6. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 1 is characterized in that: Specifically, S4 includes: collecting device temperature, current, voltage, and operating data through sensors and preliminarily processing them through edge computing modules; denoising, feature extraction, and time series construction of sensor data; using trained machine learning to perform real-time reasoning and risk prediction to obtain prediction results, and obtaining risk levels based on the prediction results.

7. The method for intelligent positioning and risk control of electric power operation environment based on three-dimensional model according to claim 1 is characterized in that: The above method dynamically integrates the operator behavior risk and equipment operation risk, calculates the comprehensive risk index in real time, and evaluates the comprehensive risk of the power operation environment in real time, as follows: Operator Behavior Risk Index (BRI): ； in, For the The weight of each behavioral risk factor; For the The real-time detection results of behavioral risk events; t is the current moment; is the number of irregular behavior types; Equipment operation risk index ERI (t): ； in, For the The weight of each equipment operation risk factor; For the The operation risk value of each equipment parameter; M is the number of operation risk types; The comprehensive risk index CRI (t) is calculated by dynamic weighting: ； in, α and β is the weight coefficient.

Images