Substation autonomous cruise inspection system based on three-dimensional laser point cloud and path planning

Abstract

The application relates to the technical field of intelligent inspection robots and path planning, in particular to a transformer substation autonomous cruise inspection system based on three-dimensional laser point clouds and path planning, which comprises an initialization module, a perception mapping module and a flow field solving module.The initialization module is used for acquiring global topological nodes and establishing an initial virtual gravitational potential field between a starting point and a target point; the perception mapping module is used for collecting real-time point cloud data through a three-dimensional laser radar, calculating point cloud density and time series variance of the point cloud density, and generating a local space impedance tensor field; the flow field solving module is used for introducing the initial virtual gravitational potential field into the local space impedance tensor field, performing local iteration through solving of a Poisson equation, and generating a continuous velocity vector field; and the servo control module is used for extracting a local streamline tangent vector and a local streamline curvature, mapping the local streamline tangent vector and the local streamline curvature into a longitudinal linear velocity and a steering angular velocity, and generating a control instruction based on the longitudinal linear velocity and the steering angular velocity to drive a chassis motor; and the application effectively reduces mechanical wear and energy consumption of the chassis motor.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection robot path planning technology, specifically to an autonomous substation patrol system based on three-dimensional laser point cloud and path planning. Background Technology

[0002] As a key piece of equipment for intelligent operation and maintenance, the smoothness and safety of the autonomous movement of substation inspection robots in scenarios such as power system protection and unattended inspections directly depend on the environmental adaptability of the navigation system. Therefore, building a robust path planning and control architecture is a prerequisite for achieving all-weather inspection.

[0003] Existing autonomous navigation methods have many problems. For example, they often use conventional discrete graph search algorithms for geometric pathfinding. However, due to the highly dynamic and strongly coupled characteristics of high-voltage cable deflection and temporary fences in substations, it is difficult to handle replanning requests for sudden obstacles with limited computing power. In particular, when faced with unstructured dynamic disturbances caused by high-voltage cable deflection, temporary fences, or moving personnel, which lack fixed geometric shapes and have highly random motion states, traditional algorithms are prone to computational deadlock and frequent sudden stops. This results in severe mechanical wear of the mobile platform in the substation and an inability to maintain a continuous motion posture that conforms to dynamics. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of computational deadlock and frequent replanning faced by conventional discrete graph search algorithms in the highly dynamic and strongly coupled environment of substations. This allows the mobile platform to avoid sudden braking and replanning when encountering unexpected obstacles, and instead exhibit a smooth fluid-like circumvention posture that conforms to dynamics. This effectively reduces the mechanical wear and energy consumption of the chassis motor, and gives the system high real-time performance and robustness when navigating complex equipment groups. It can also locate and report hidden dynamic risks in substations in real time.

[0005] To achieve the above objectives, this invention provides an autonomous substation patrol and inspection system based on three-dimensional laser point clouds and path planning. Specifically, the system includes a mobile platform, which is equipped with a chassis motor, and the following components are mounted on the mobile platform: The initialization module is used to read the pre-built global topology map of the substation to obtain global topology nodes and establish an initial virtual gravitational potential energy field between the starting point and the target point; The perception mapping module is used to collect real-time point cloud data through a 3D LiDAR mounted on a mobile platform, and extract point cloud features within a local perception domain based on the real-time point cloud data. The local perception domain is a spherical spatial region with the geometric center of the mobile platform as the origin and the effective detection range of the 3D LiDAR as the radius. The point cloud features include point cloud density and the temporal variance of the point cloud density. A local spatial impedance tensor field is generated based on the point cloud density and the temporal variance of the point cloud density. The flow field solution module is used to introduce the initial virtual gravitational potential energy field into the local spatial impedance tensor field, and generate a continuous velocity vector field by solving the Poisson equation through local iteration. The servo control module is used to extract the local streamline tangent vector and local streamline curvature corresponding to the current physical coordinates of the mobile platform in the continuous velocity vector field, map the local streamline tangent vector to longitudinal linear velocity, map the local streamline curvature to steering angular velocity, and generate control commands based on the longitudinal linear velocity and steering angular velocity to drive the chassis motor.

[0006] As a further aspect of the present invention: the perception mapping module includes: a voxelization unit, used to divide the local perception domain into multiple spatial voxels; the local perception domain is a spherical spatial region with the geometric center of the mobile platform as the origin and the effective detection distance of the three-dimensional lidar as the radius; The feature calculation unit is used to count the number of point clouds in each spatial voxel to obtain the point cloud density, and to extract the point cloud density from multiple consecutive data frames to calculate the temporal variance of the point cloud density.

[0007] As a further aspect of the present invention, the perception mapping module further includes: Tensor mapping unit is used to substitute the point cloud density and temporal variance of the point cloud density corresponding to each spatial voxel into the preset impedance mapping model. The impedance mapping model generates a local spatial impedance tensor field, which is composed of the impedance tensor matrix corresponding to each spatial voxel. The condition judgment unit is used to set the dynamic perturbation variance threshold. When the temporal variance of the point cloud density of a single spatial voxel is greater than or equal to the dynamic perturbation variance threshold, the impedance tensor matrix of the corresponding spatial voxel is multiplied by a preset amplification factor greater than one. When the temporal variance of the point cloud density of a single spatial voxel is less than the dynamic perturbation variance threshold, the impedance tensor matrix of the corresponding spatial voxel is kept as the basic mapping value.

[0008] As a further aspect of the present invention: the flow field solution module includes: The equation building unit is used to construct the local Poisson equation with the local spatial impedance tensor field as the boundary condition and the initial virtual gravitational potential energy field as the driving source. The iterative solution unit is used to numerically iterate the local Poisson equation to obtain the local potential function distribution, and obtain the velocity vector of each coordinate point in space by calculating the negative gradient of the local potential function. The velocity vectors of all coordinate points are combined to generate a continuous velocity vector field.

[0009] As a further aspect of the present invention: the servo control module includes: The state extraction unit is used to substitute the current physical coordinates of the mobile platform into the continuous velocity vector field, extract the velocity vector of the corresponding coordinate point as the local streamline tangent vector, and calculate the local streamline curvature. The instruction decoding unit is used to convert the local streamline tangent vector into the longitudinal linear velocity control signal of the chassis motor and the local streamline curvature into the steering angular velocity control signal of the chassis motor through a preset kinematic model.

[0010] As a further aspect of the present invention, the servo control module further includes: The closed-loop actuator is used to input the longitudinal linear velocity control signal and the steering angular velocity control signal into the proportional-integral-derivative controller, and output the motor drive current to the chassis motor through the proportional-integral-derivative controller.

[0011] As a further aspect of the present invention, the system also includes: The anomaly recording module is used to extract the Frobenius norm of the impedance tensor matrix in the local spatial impedance tensor field, and to extract the coordinates of points where the Frobenius norm is greater than a preset impedance threshold. The heatmap generation module records the coordinates of points where the Frobenius norm of the impedance tensor matrix is ​​greater than a preset impedance threshold and aggregates them to generate a heatmap of dynamic interference in the microenvironment. The heatmap of dynamic interference in the microenvironment is then reported as an early warning parameter. For coordinates where the Frobenius norm of the impedance tensor matrix is ​​less than or equal to the preset impedance threshold, they are determined to be safe areas and are not recorded.

[0012] As a further aspect of the present invention: the mobile platform is an inspection robot configured in a substation.

[0013] The beneficial effects of this invention are: 1) This invention generates a continuous velocity vector field by constructing a local spatial impedance tensor field and solving the Poisson equation through local iteration, transforming discrete geometric pathfinding into continuous physical field control. This effectively solves the problem that conventional algorithms are prone to computational deadlock in the high dynamic environment of substations, and enables the mobile platform to avoid sudden braking and replanning when facing unstructured dynamic interference such as wind deflection of high-voltage cables, thus achieving smooth and compliant obstacle avoidance in accordance with dynamics. 2) This invention utilizes the local streamline tangent vector and local streamline curvature to directly map longitudinal linear velocity and steering angular velocity to generate chassis motor control commands, establishing a closed-loop mapping from sensing the flow field to motor execution; this mechanism avoids the frequent abrupt stops and turns of traditional path planning, ensuring that the mobile platform exhibits a continuous and smooth motion posture in the substation, significantly reducing the mechanical wear and energy consumption of the chassis motor, and improving the stability of inspection operations and equipment lifespan; 3) This invention introduces point cloud density temporal variance calculation, quantifying environmental dynamic features into a preset amplification factor of impedance tensor; the system can keenly capture and actively avoid implicit dynamic risks such as high-voltage cable wind deflection and static equipment boundaries. When the temporal variance is greater than or equal to the dynamic disturbance variance threshold, the impedance tensor matrix of the corresponding spatial voxel is multiplied by a preset amplification factor greater than one, forming a high impedance repulsion zone in the virtual flow field, thereby ensuring the inherent safety of the mobile platform in the complex coupled substation environment and solving the problem of dynamic interference being difficult to quantify and avoid. 4) This invention adopts a strategy of introducing the initial virtual gravitational potential energy field into the local spatial impedance tensor field for coupling solution, unifying macroscopic topological guidance and microscopic real-time obstacle avoidance within the framework of the local Poisson equation; this design not only ensures the global convergence of the robot toward the target, but also solves the local minimum trap by utilizing the fluid bypass characteristics, ensuring efficiency and robustness in shuttling through static equipment groups. 5) This invention has an abnormal micro-environment perception function. By recording the coordinates of points where the Frobenius norm of the impedance tensor matrix is ​​greater than a preset impedance threshold and aggregating them to generate a micro-environment dynamic interference heat map, the system can automatically identify and report the micro-environment dynamic interference heat map as an early warning parameter while performing patrol tasks. This can provide early warning of potential risk areas such as wind-induced dynamic disturbances of high-voltage cables in substations, thereby improving the overall efficiency of unattended substation patrols. Attached Figure Description

[0014] Figure 1 A schematic diagram of a substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning, provided in an embodiment of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, the substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning includes a mobile platform, which is equipped with a chassis motor, and an initialization module configured on the mobile platform, which is used to read the global topology map pre-built by the substation to obtain global topology nodes and establish an initial virtual gravitational potential energy field between the starting point and the target point. The perception mapping module is used to collect real-time point cloud data through a 3D LiDAR mounted on a mobile platform, and extract point cloud features within a local perception domain based on the real-time point cloud data. The local perception domain is a spherical spatial region with the geometric center of the mobile platform as the origin and the effective detection range of the 3D LiDAR as the radius. The point cloud features include point cloud density and the temporal variance of the point cloud density. A local spatial impedance tensor field is generated based on the point cloud density and the temporal variance of the point cloud density. The flow field solution module is used to introduce the initial virtual gravitational potential energy field into the local spatial impedance tensor field, and generate a continuous velocity vector field by solving the Poisson equation through local iteration. The servo control module is used to extract the local streamline tangent vector and local streamline curvature corresponding to the current physical coordinates of the mobile platform in the continuous velocity vector field, map the local streamline tangent vector to longitudinal linear velocity, map the local streamline curvature to steering angular velocity, and generate control commands based on the longitudinal linear velocity and steering angular velocity to drive the chassis motor.

[0017] This embodiment provides an autonomous patrol and inspection system for substations based on 3D laser point cloud and path planning. It aims to solve the problems of computational deadlock and frequent replanning faced by conventional discrete graph search algorithms in the highly dynamic and strongly coupled environment of substations. The mobile platform in this system can be a patrol robot entity configured in a confined space such as a substation. The initialization module acquires global topology nodes, which are used to form a global guidance path at the macro level; and establishes an initial virtual gravitational potential energy field between the starting point and the target point. This potential energy field constructs a spatial potential energy gradient based on the macro topology without considering microscopic obstacles, so as to provide the mobile platform with a macroscopic driving force to move towards the target. Specifically, the initialization module extracts the coordinates of the starting point and the target point, and constructs a virtual potential energy distribution pointing from each point in space to the target point. In practical applications, due to the vast area of ​​the substation, directly establishing a single potential energy field between the starting point and the final target point is prone to getting trapped in local minima. Therefore, the initialization module obtains topological nodes by reading the pre-constructed global topology map of the substation. Using a graph search algorithm, specifically the A* algorithm, with the Euclidean distance between nodes as the heuristic function and the actual physical path length between nodes as the cost function, a global guidance path composed of a series of topological nodes is planned at the macro level. The next topological node on this global guidance path that is closest to the current position of the mobile platform is taken as the local target point, and a virtual potential energy distribution pointing from each point in space to this local target point is constructed. The specific calculation formula for this initial virtual gravitational potential energy field is as follows: in, For the initial virtual gravitational potential energy field, Let these be the physical coordinates of any grid point in space. The physical coordinates of the local target point. The preset potential energy conversion coefficient is used to form a funnel-shaped gravitational potential energy field in the local space with the local target point as the unique minimum point, thereby achieving effective decoupling between global macroscopic pathfinding and local microscopic potential field. During the movement of the mobile platform, the initialization module calculates the Euclidean distance between the current physical coordinates of the mobile platform and the current local target point in real time. When the Euclidean distance is less than the preset arrival judgment threshold, such as 0.5m, the local target point is dynamically updated. That is, the next adjacent topological node is extracted from the global guidance path as the new local target point, and the initial virtual gravitational potential energy field is recalculated based on the physical coordinates of the new local target point. This process is repeated until the mobile platform reaches the final target point of the global guidance path. The perception mapping module acquires real-time point cloud data through 3D LiDAR, extracts point cloud features within the local perception domain, and calculates the point cloud density and its temporal variance. Based on the point cloud density and temporal variance, it generates a local spatial impedance tensor field. This tensor field mathematically transforms discrete physical obstacles and their dynamic evolution characteristics into anisotropic viscosity coefficient distributions in a fluid medium to characterize the degree of obstruction to the passage of the mobile platform in different areas of space. The perception mapping module transforms the complex physical environment of high-voltage cable wind deflection and temporary insulating fences in the substation into a continuous tensor distribution. The point cloud density is used to extract and characterize the static equipment boundary features in the substation, and the temporal variance of the point cloud density is used to extract and characterize the dynamic disturbance features of high-voltage cable wind deflection. High-density or high-variance regions are represented by high-impedance tensors. The flow field solving module introduces the initial virtual gravitational potential energy field into the local spatial impedance tensor field. By solving the Poisson equation and performing local iteration, a continuous velocity vector field is generated. After solving the differential equation in the local space, a smooth velocity distribution is obtained to provide collision-free continuous motion guidance for the mobile platform. During this process, the flow field solving module couples the macroscopic driving force with the microscopic impedance environment, so that the generated virtual fluid can naturally bypass the high impedance region. The servo control module extracts the local streamline tangent vector and local streamline curvature corresponding to the current physical coordinates of the mobile platform in the continuous velocity vector field. It maps the local streamline tangent vector to longitudinal linear velocity and the local streamline curvature to steering angular velocity. Based on the longitudinal linear velocity and steering angular velocity, it generates control commands to drive the chassis motor. The servo control module directly extracts differential geometric features from the continuous velocity vector field and converts them into low-level motor control signals that conform to the kinematic constraints of the mobile platform, thus realizing a closed loop from perception to control. This system, through the collaborative efforts of the initialization module, perception mapping module, flow field solution module, and servo control module, corrects the simplified reduced-order modeling method that treats robot navigation as discrete geometric pathfinding, transforming it into a continuous control problem in a physical field. In the substation inspection scenario, this architecture eliminates the serial computational delay of traditional map building, path search, and trajectory tracking, reducing the computational complexity of dynamic obstacle avoidance from the exponential level of graph search to the linear level of solving local partial differential equations. This allows the mobile platform to exhibit a smooth fluid-like circumvention posture that conforms to dynamics when facing sudden obstacles without the need for abrupt braking and replanning, effectively reducing the mechanical wear and energy consumption of the chassis motor.

[0018] In this embodiment, the perception mapping module includes: a voxelization unit, used to divide the local perception domain into multiple spatial voxels; the local perception domain is a spherical spatial region with the geometric center of the mobile platform as the origin and the effective detection distance of the three-dimensional lidar as the radius; and a feature calculation unit, used to count the number of point clouds in each spatial voxel to obtain the point cloud density, and extract the point cloud density of multiple consecutive data frames to calculate the temporal variance of the point cloud density.

[0019] This embodiment has refined the design of the specific structure and operation mechanism of the perception mapping module; the voxelization unit divides the local perception domain into multiple spatial voxels, which are the smallest cubic units after discretizing the continuous three-dimensional physical space, so as to provide a structured statistical carrier for massive unordered point cloud data. The voxelization unit establishes a local coordinate system to mesh the three-dimensional space in front of the mobile platform. The feature calculation unit counts the number of points within each spatial voxel to obtain the point cloud density. Simultaneously, it extracts and records the 3D physical coordinates of each point cloud within each spatial voxel for subsequent covariance matrix construction. It also extracts the point cloud density from multiple consecutive data frames to calculate the temporal variance of the point cloud density. To quantify the static occupancy and dynamic perturbation of spatial voxels, the feature calculation unit calculates the temporal variance of the point cloud density using the following formula: in, Let be the temporal variance of the point cloud density; The number of consecutive data frames, derived from the system sampling rate preset; For the first Spatial voxel point cloud density of a frame; For the current voxel continuum Mean point cloud density of the frame; Building upon this foundation, the feature calculation unit introduces temporal variance, enabling the system to identify dynamic environmental characteristics. For instance, wind-blown drooping cables can cause significant fluctuations in the number of points within the same voxel, leading to a substantial increase in the variance of that voxel. Through the collaboration between the voxelization unit and the feature calculation unit, this solution not only efficiently compresses the amount of unstructured point cloud data but, more importantly, in the complex environment of substations, by extracting the temporal variance of point cloud density, the perception mapping module can keenly capture implicit dynamic constraints such as wind deflection of high-voltage cables, providing a solid data foundation for the subsequent construction of a tensor field reflecting the environmental hazard gradient.

[0020] In this embodiment, the perception mapping module further includes: a tensor mapping unit, used to substitute the point cloud density and temporal variance of the point cloud density corresponding to each spatial voxel into a preset impedance mapping model, the impedance mapping model generating a local spatial impedance tensor field, the local spatial impedance tensor field being composed of the impedance tensor matrix corresponding to each spatial voxel; and a condition judgment unit, used to set a dynamic perturbation variance threshold, when the temporal variance of the point cloud density of a single spatial voxel is greater than or equal to the dynamic perturbation variance threshold, multiplying the impedance tensor matrix of the corresponding spatial voxel by a preset amplification factor greater than one; when the temporal variance of the point cloud density of a single spatial voxel is less than the dynamic perturbation variance threshold, maintaining the impedance tensor matrix of the corresponding spatial voxel as the basic mapping value.

[0021] This embodiment further defines the specific logic of tensor generation in the perception mapping module; the tensor mapping unit substitutes the point cloud density and temporal variance of the point cloud density corresponding to each spatial voxel into a preset impedance mapping model to generate a local spatial impedance tensor field. This tensor mapping unit uses the impedance mapping matrix to transform scalar features into tensor features, making the spatial impedance anisotropic; specifically, the tensor mapping unit calculates the basic mapping value according to the following formula. : in, It is the basic impedance coefficient, an empirical value pre-calibrated based on the dimensions of the mobile platform and chassis dynamics; It is a natural constant; The density-sensitive factor is a constant obtained by fitting the point cloud distribution of typical obstacles through actual measurements; the impedance mapping matrix is ​​constructed using tensor mapping units. Its structural form is as follows: in, It is the identity matrix. This is the covariance matrix of the point cloud distribution within the current voxel. This covariance matrix is ​​obtained by calculating the deviation covariance of the three-dimensional physical coordinates of each point within the current voxel relative to the geometric center of the point cloud within the voxel, based on the three-dimensional physical coordinates of each point cloud transmitted by the feature computing unit. Specifically, suppose the current voxel contains A point cloud, For positive integers, point cloud Coordinates are Its geometric center coordinates are Then the covariance matrix The calculation formula is: Among them, superscript This is the matrix transpose symbol; when the number of point clouds within a voxel is lower than the preset effective calculation threshold, directly set... The zero matrix is ​​used to map the scalar density into a second-order tensor that reflects the dominant extension trend of the obstacle in different spatial directions; Specifically, the physical significance of introducing the covariance matrix lies in giving the impedance field anisotropic characteristics: when the point cloud presents a linear distribution, the covariance matrix makes the impedance increment along the extension direction of the object smaller, while the impedance increment perpendicular to the surface of the object larger, thereby guiding the virtual fluid to flow along the surface of the obstacle rather than passing through it perpendicularly, forming a bypass trajectory that conforms to fluid dynamics. The condition judgment unit sets a dynamic disturbance variance threshold. When the time order variance is greater than or equal to the dynamic disturbance variance threshold, the condition judgment unit multiplies the impedance tensor matrix of the corresponding space voxel by a preset amplification factor greater than one. When the time order variance is less than the dynamic disturbance variance threshold, the condition judgment unit keeps the impedance tensor matrix of the corresponding space voxel as the basic mapping value. To achieve proactive avoidance of highly dynamic hazardous areas, the condition judgment unit introduces a nonlinear impedance amplification mechanism, which maps the impedance matrix... The mathematical expression of this mechanism, which uses the underlying tensor for overall numerical scaling, is as follows: in, This is the impedance tensor matrix of the corresponding spatial voxel in the final output, i.e., the tensor matrix. The preset amplification factor is set by the substation safety distance specification; The fundamental impedance mapping matrix; This is the dynamic disturbance variance threshold, set for statistical analysis of historical normal environmental fluctuations; In a typical application scenario of this embodiment, for a 500kV substation environment, when the sampling rate of the three-dimensional lidar is 10Hz, the dynamic disturbance variance threshold is... The value range is usually set to to This is to effectively distinguish between the inherent noise of the sensor and actual dynamic disturbances; The condition judgment unit ensures, through the above logic, that low-variance regions such as static equipment cabinets only exhibit conventional rigid boundaries, while high-variance regions such as violently swinging cables or moving personnel will trigger a sharp amplification of the impedance tensor, which will manifest as regions with increased viscosity and strong repulsive force in the virtual flow field. The combination of tensor mapping unit and condition judgment unit cleverly transforms the special dynamic physical coupling environment of substation into the underlying parameters of navigation control. When encountering dynamic obstacles, the nonlinear rapid expansion of the local spatial impedance tensor field forces the virtual fluid to divert in advance, enabling the mobile platform to smoothly bypass the dynamic hazard source at a safe distance. This effectively solves the problem of lag and deadlock caused by frequent replanning in traditional algorithms.

[0022] In this embodiment, the flow field solution module includes: an equation construction unit, used to construct a local Poisson equation with the local spatial impedance tensor field as the boundary condition and the initial virtual gravitational potential energy field as the driving source; and an iterative solution unit, used to perform numerical iterative solution of the local Poisson equation to obtain the local potential function distribution, and to obtain the velocity vector of each coordinate point in space by calculating the negative gradient of the local potential function, and to combine the velocity vectors of all coordinate points to generate a continuous velocity vector field.

[0023] This embodiment elaborates on the internal operation mechanism of the flow field solution module; the equation construction unit uses the local spatial impedance tensor field as the anisotropic guiding medium coefficient and the initial virtual gravitational potential energy field as the driving source to construct the local Poisson equation. The local Poisson equation is a second-order partial differential equation describing the relationship between the potential energy distribution in space and the impedance source, so as to constrain the flow pattern of the virtual fluid in the complex impedance environment. In this process, the specific mathematical implementation of introduction and coupling is manifested as follows: the initial virtual gravitational potential energy field is substituted into the source term on the right side of the Poisson equation as the driving source, and the local spatial impedance tensor field is substituted into the divergence term on the left side of the equation as the anisotropic diffusion coefficient matrix. Thus, the equation building unit unifies the macroscopic navigation intention and the microscopic obstacle avoidance requirement under the same differential equation framework. In the physical analogy corresponding to this mathematical form, the left side of the equation describes the steady-state seepage and diffusion process of a virtual fluid in an anisotropic impedance medium, while the source term on the right side of the equation is analogous to a distributed source term with elastic restoring force. The physical principle behind using the difference in potential functions as a source term in the Poisson equation lies in forcing the local potential function of the microscopic solution to be solved. In the unobstructed region with low impedance, the difference term infinitely approaches the macroscopically set initial virtual gravitational potential energy field. This is equivalent to potential energy deviation, which, when multiplied by a relaxation coefficient, forms a driving force that pulls the microscopic flow field back to the macroscopic guiding path. This ensures that the moving platform can quickly resume its macroscopic motion toward the target after bypassing high-impedance obstacles, clearly realizing the logical transformation from macroscopic driving to solving the microscopic potential field. Specifically, the local Poisson equation is constructed in an anisotropic diffusion form: in, Let be the local potential function to be solved. The local spatial impedance tensor field serves as the fluid permeation resistance. To guide the tensor, and to prevent local spatial impedance tensor fields in unobstructed or coplanar point cloud regions. In the case of singular matrices that are not invertible, before performing the inversion operation, [the following steps are taken]. Introducing a very small positive real regularization constant That is, the derived tensor actually substituted into the equation is To ensure the absolute stability of numerical calculations, For vector differential operators, The initial virtual gravitational potential energy field, which manifests as a scalar potential value at a specific spatial grid point, is calculated using the following formula: in, This represents the Euclidean distance from the spatial grid point to the target point. The relaxation coefficient is used to control the smoothness of the potential field distribution during the iterative process; The iterative solution unit performs numerical iterative solutions to the local Poisson equation, obtains the velocity vector of each coordinate point in space, and combines the velocity vectors of all coordinate points to generate a continuous velocity vector field. In order to achieve ultra-fast solutions at the microsecond level on the edge computing nodes, the iterative solution unit uses the finite difference method to discretize and iterate the Poisson equation. In this process, for the left side of the equation containing variable coefficients divergence term The iterative solution element employs a conserved difference scheme on an interlaced mesh to handle interfaces with drastically changing impedances in space, specifically targeting... The second-order difference term in the axial direction is discretized using the following formula: Among them, symbols This represents a numerical discretization approximation. For grid step size, For grid nodes and The equivalent guiding coefficient of the interface between them. For grid nodes and The equivalent guiding coefficient of the interface between them. These are the corresponding grid nodes. , , The local potential function value at point A; its harmonic mean is calculated using the following formula: in, Take the principal component in the corresponding direction. Grid nodes and The principal component of the guiding tensor at the point is used to ensure the flux continuity of the virtual fluid when crossing the high impedance boundary; the Gauss-Seidel iterative method is used to update the potential function distribution point by point until the convergence condition is met. The convergence condition is specifically set as follows: the maximum absolute value of the difference between the potential functions of all spatial grid points in the current iteration and the previous iteration is less than the preset convergence tolerance threshold, or the number of iterations reaches the preset maximum number of iterations. The velocity vector is extracted as follows: in, The velocity vector of a point in space; Based on this, the iterative solution unit calculates the negative gradient of the local potential function to obtain the optimal flow direction and velocity at each discrete coordinate point in the current local sensing domain, thereby forming a complete continuous velocity vector field. The equation building unit and iterative solution unit cleverly transform the global graph search problem into a local matrix differential equation solving problem. This design limits the computational complexity to a constant level within the local grid. In the strong electromagnetic interference environment of substations, the control delay of dynamic obstacle avoidance can be strictly compressed within the radar scanning cycle, ensuring the efficient and stable operation of edge computing nodes. This gives the system high real-time performance and robustness when navigating complex equipment clusters.

[0024] In this embodiment, the servo control module includes: a state extraction unit, used to substitute the current physical coordinates of the mobile platform into a continuous velocity vector field, extract the velocity vector of the corresponding coordinate point as a local streamline tangent vector, and calculate the local streamline curvature; and an instruction decoding unit, used to convert the local streamline tangent vector into a longitudinal linear velocity control signal of the chassis motor and convert the local streamline curvature into a steering angular velocity control signal of the chassis motor through a preset kinematic model.

[0025] This embodiment further standardizes the instruction decoding process of the servo control module; the state extraction unit substitutes the current physical coordinates of the mobile platform into the continuous velocity vector field, and calculates the local streamline tangent vector and local streamline curvature at the current physical coordinate position; Specifically, the state extraction unit extracts the velocity vector of the grid point corresponding to the current physical coordinates and directly uses it as the local streamline tangent vector. That is, its amplitude is preserved to characterize the actual acceleration and deceleration dynamics of the flow field on the moving platform, and based on this local streamline tangent vector and the spatial Jacobian matrix of the continuous velocity vector field at that physical coordinate. Calculate the local streamline curvature; Specifically, the state extraction unit extracts the velocity vectors of adjacent grid points in the current physical coordinates and calculates the velocity vectors in each point using the three-dimensional central difference method. , , Partial derivatives in three spatial dimensions, thus constructing Spatial Jacobian Matrix Accurately calculate the local streamline curvature The calculation formula is: in, For the smallest positive real number that is pre-defined to prevent division by zero anomalies, The cross product symbol is used for vectors. The vector 2 norm symbol is used to calculate the magnitude of the vector; it is used to avoid computational crashes when the mobile platform is in the region of minimum velocity, thereby eliminating the numerical singularity of the calculation of the normal component of the directional derivative; in this way, the state extraction unit constructs a state mapping mechanism between virtual flow field features and physical entity kinematic constraints, accurately capturing the transient hydrodynamic constraints experienced by the mobile platform in the flow field; The instruction decoding unit uses a preset kinematic model to convert local streamline tangent vectors into longitudinal linear velocity control signals for the chassis motor, and local streamline curvature into steering angular velocity control signals for the chassis motor. To map abstract geometric features into specific kinematic commands, the instruction decoding unit executes the following conversion logic: in, The longitudinal linear velocity, For speed mapping coefficients, This refers to the steering angular velocity; The instruction decoding unit ensures that the linear velocity of the mobile platform is proportional to the flow velocity of the virtual fluid, and at the same time, it directly calculates the angular velocity required for turning by using the product of curvature and linear velocity, thus ensuring the smoothness of the motion trajectory. By combining the state extraction unit and the instruction decoding unit, this scheme establishes a lossless conversion channel from continuous differential geometric features to discrete physical motor control. In the narrow inspection channel of the substation, this direct mapping based on streamline curvature avoids the complex trajectory sampling and cost function evaluation in the traditional local path planning algorithm, and directly outputs deterministic smooth instructions that conform to the control boundary of non-electrical variables, which greatly improves the continuity and response speed of chassis movement.

[0026] In this embodiment, the servo control module further includes a closed-loop execution unit, which is used to input the longitudinal linear velocity control signal and the steering angular velocity control signal into the proportional-integral-derivative controller, and output the motor drive current to the chassis motor through the proportional-integral-derivative controller.

[0027] This embodiment improves the underlying hardware driver closed loop of the servo control module; the closed-loop execution unit, based on the specific configuration of the mobile platform chassis, allocates the longitudinal linear velocity control signal and the steering angular velocity control signal to the target linear velocity of each independent drive wheel through a preset chassis kinematics inverse matrix. For example, for a two-wheel differential chassis, the target speed of the left wheel is... The calculation formula is: Right wheel target speed for: in, The left and right wheel distances are pre-calibrated through actual measurements. The closed-loop execution unit converts the target linear velocity of each independent drive wheel into the target motor speed and inputs the target motor speed into a proportional-integral-derivative (PID) controller. This PID controller employs an industry-standard closed-loop feedback control algorithm module to eliminate the steady-state error between the target motor speed command and the actual motor speed. Specifically, the PID controller uses a discretized position-type PID control law, and its output at the 1st... Target current command for each control cycle The calculation formula is: in, For the first Speed ​​error per control cycle For the first The speed error per control cycle, which is the difference between the target motor speed and the actual speed fed back by the motor encoder; , , These are the pre-calibrated proportional gain, integral gain, and derivative gain, respectively. The closed-loop execution unit outputs a continuous motor drive current to the chassis motor through a proportional-integral-derivative controller. During this period, the closed-loop execution unit receives the ideal speed signal parsed by the instruction decoding unit and, in combination with the actual speed fed back by the motor encoder, adjusts the output current in real time to resist external physical disturbances. The introduction of the closed-loop execution unit provides a solid physical execution guarantee for the entire autonomous cruise inspection system. In the outdoor inspection scenario of substations, in the face of physical disturbances caused by uneven ground or strong electromagnetic interference, the system can accurately resist external deviations through high-frequency closed-loop current regulation, ensuring that the mobile platform strictly follows the smooth streamline of the continuous velocity vector field planning, verifying the high robustness of this technical solution under complex road conditions.

[0028] In this embodiment, the system further includes: an anomaly recording module, used to extract the Frobenius norm of the impedance tensor matrix in the local spatial impedance tensor field, and to extract coordinate points where the Frobenius norm is greater than a preset impedance threshold; and a heatmap generation module, used to record coordinate points where the Frobenius norm of the impedance tensor matrix is ​​greater than the preset impedance threshold and aggregate them to generate a microenvironment dynamic interference heatmap, and to report the microenvironment dynamic interference heatmap as an early warning parameter; for coordinate points where the Frobenius norm of the impedance tensor matrix is ​​less than or equal to the preset impedance threshold, they are determined to be safe areas and the recording is abandoned.

[0029] This embodiment expands the system's functionality at the security patrol level; the anomaly recording module extracts the coordinates of the impedance tensor matrix in the local spatial impedance tensor field where the Frobenius norm is greater than a preset impedance threshold. The preset impedance threshold here is derived from the statistical extreme value setting of the impedance tensor characteristics under the condition of no environmental anomalies in the historical security inspection records. In the specific comparison, the anomaly recording module calculates the Frobenius norm of the impedance tensor matrix corresponding to each coordinate point, which is the square root of the sum of the squares of the absolute values ​​of each element in the impedance tensor matrix. The norm is used as a scalar value representing the comprehensive impedance strength of the local space and compared with a preset impedance threshold. When the norm is greater than the preset impedance threshold, the coordinate point is extracted. This module reuses the impedance data of the navigation layer, without the need for additional image recognition algorithms, and directly detects environmental anomalies from the distortion of the physical field. The heatmap generation module records the coordinates of points where the Frobenius norm of the impedance tensor matrix is ​​greater than a preset impedance threshold, and calculates the response value by applying a three-dimensional Gaussian kernel function to each recorded coordinate point. Specifically, for any grid point in space... In its first Recorded coordinate points The three-dimensional Gaussian kernel function response value at [location] The calculation formula is: in, The preset spatial smoothing radius; and All are physical space coordinates; The weighting coefficient is calculated using the following formula: in, This is the early warning gain coefficient, for example, a value of 1.0. For the first The Frobenius norm of the impedance tensor matrix at each recorded coordinate point To preset the impedance threshold, this formula ensures the weighting coefficients. The intensity of dynamic disturbance is quantified by a linear positive correlation with the severity of impedance anomalies. The heatmap generation module linearly superimposes the kernel function response values ​​on the spatial grid to achieve spatial smoothing and weight accumulation. The calculation formula is as follows: in, Spatial grid points The response values ​​of the micro-environment dynamic interference heatmap are aggregated to generate a micro-environment dynamic interference heatmap. This heatmap reflects the multi-dimensional visual distribution of spatial physical structure anomalies or high-frequency dynamic disturbances, providing maintenance personnel with intuitive early warnings of equipment loosening, foreign object intrusion, or cable wind deflection. The heatmap generation module reports the micro-environment dynamic interference heatmap as an early warning parameter. When the Frobenius norm of the impedance tensor matrix is ​​less than or equal to the preset impedance threshold, the heatmap generation module determines the corresponding coordinate point as a safe area and discards the record. This solution achieves underlying data homogeneity between navigation control and security patrol through the anomaly recording module and the heat map generation module. In the scenario of all-weather inspection of substations, this unexpected synergy makes the system not only an efficient mobile platform chassis controller, but also an all-weather environmental anomaly detector. It can locate and report hidden dynamic risks in the substation in real time, which greatly improves the overall application value and computing resource utilization of the automatic inspection system.

[0030] In this embodiment, the mobile platform is an inspection robot configured in the substation; the system extracts the point cloud density and the temporal variance of the point cloud density from the real-time point cloud data, which are used to capture the static equipment boundary features and the dynamic disturbance features of wind deflection of high-voltage cables in the substation, respectively.

[0031] This embodiment clarifies the specific application scenario and hardware requirements of the system; the system limits the mobile platform to an inspection robot configured in a substation; as a typical confined space with strong electromagnetic fields and high risks, the substation places extremely stringent requirements on the navigation safety and real-time performance of the inspection robot. By applying this system to substation inspection robots, this solution specifically overcomes the unique technical difficulties of high-density static equipment and localized high-dynamic disturbances in substation scenarios. The system utilizes the tensor mapping and streamline fitting mechanism unique to rheofield navigators, which not only ensures the safe movement of robots among complex equipment cabinets, but also effectively mitigates dynamic threats such as wind deflection of high-voltage cables. It achieves deterministic and highly reliable autonomous navigation in high-risk environments with zero fault tolerance, demonstrating strong scene adaptability.

[0032] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A substation autonomous patrol and inspection system based on 3D laser point cloud and path planning, characterized in that, Includes a mobile platform, the mobile platform being equipped with a chassis motor, and the following configured on the mobile platform: The initialization module is used to read the pre-built global topology map of the substation to obtain global topology nodes and establish an initial virtual gravitational potential energy field between the starting point and the target point; The perception mapping module is used to collect real-time point cloud data through a three-dimensional lidar mounted on the mobile platform, and extract point cloud features within a local perception domain based on the real-time point cloud data. The local perception domain is a spherical spatial region with the geometric center of the mobile platform as the origin and the effective detection distance of the three-dimensional lidar as the radius. The point cloud features include point cloud density and the temporal variance of the point cloud density. A local spatial impedance tensor field is generated based on the point cloud density and the temporal variance of the point cloud density. The flow field solution module is used to introduce the initial virtual gravitational potential energy field into the local spatial impedance tensor field, and generate a continuous velocity vector field by solving the Poisson equation through local iteration. The servo control module is used to extract the local streamline tangent vector and local streamline curvature corresponding to the current physical coordinates of the mobile platform in the continuous velocity vector field, map the local streamline tangent vector to longitudinal linear velocity, map the local streamline curvature to steering angular velocity, and generate control commands based on the longitudinal linear velocity and the steering angular velocity to drive the chassis motor.

2. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 1, characterized in that, The perception mapping module includes: a voxelization unit, used to divide the local perception domain into multiple spatial voxels; The feature calculation unit is used to count the number of point clouds in each spatial voxel to obtain the point cloud density, and to extract the point cloud density from multiple consecutive data frames to calculate the temporal variance of the point cloud density.

3. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 2, characterized in that, The perception mapping module also includes: Tensor mapping unit is used to substitute the point cloud density and the temporal variance of the point cloud density corresponding to each spatial voxel into a preset impedance mapping model. The impedance mapping model generates the local spatial impedance tensor field, which is composed of the impedance tensor matrix corresponding to each spatial voxel. The condition judgment unit is used to set a dynamic perturbation variance threshold. When the temporal variance of the point cloud density of a single spatial voxel is greater than or equal to the dynamic perturbation variance threshold, the impedance tensor matrix of the corresponding spatial voxel is multiplied by a preset amplification factor greater than one. When the temporal variance of the point cloud density of a single spatial voxel is less than the dynamic perturbation variance threshold, the impedance tensor matrix of the corresponding spatial voxel is kept as the basic mapping value.

4. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 1, characterized in that, The flow field solving module includes: The equation-building unit is used to construct the local Poisson equation with the local spatial impedance tensor field as the boundary condition and the initial virtual gravitational potential energy field as the driving source. The iterative solution unit is used to perform numerical iterative solution of the local Poisson equation to obtain the local potential function distribution, and to obtain the velocity vector of each coordinate point in space by calculating the negative gradient of the local potential function, and to combine the velocity vectors of all the coordinate points to generate the continuous velocity vector field.

5. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 1, characterized in that, The servo control module includes: The state extraction unit is used to substitute the current physical coordinates of the mobile platform into the continuous velocity vector field, extract the velocity vector of the corresponding coordinate point as the local streamline tangent vector, and calculate the local streamline curvature. The instruction decoding unit is used to convert the local streamline tangent vector into the longitudinal linear velocity control signal of the chassis motor and the local streamline curvature into the steering angular velocity control signal of the chassis motor through a preset kinematic model.

6. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 5, characterized in that, The servo control module also includes: The closed-loop execution unit is used to input the longitudinal linear velocity control signal and the steering angular velocity control signal into the proportional-integral-derivative controller, and output the motor drive current to the chassis motor through the proportional-integral-derivative controller.

7. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 1, characterized in that, Also includes: An anomaly recording module is used to extract the Frobenius norm of the impedance tensor matrix in the local spatial impedance tensor field, and to extract the coordinates of points where the Frobenius norm is greater than a preset impedance threshold. The heat map generation module is used to record the coordinates of the points where the Frobenius norm of the impedance tensor matrix is ​​greater than the preset impedance threshold and aggregate them to generate a micro-environment dynamic interference heat map, and report the micro-environment dynamic interference heat map as an early warning parameter. For coordinate points where the Frobenius norm of the impedance tensor matrix is ​​less than or equal to the preset impedance threshold, they are determined to be safe regions and the records are discarded.

8. The substation autonomous patrol and inspection system based on three-dimensional laser point cloud and path planning according to claim 1, characterized in that, The mobile platform is an inspection robot configured at the substation.

Images