Unmanned aerial vehicle electric power inspection obstacle avoidance system based on sparse Bayesian inference

Abstract

The invention relates to the technical field of environment perception, in particular to an unmanned aerial vehicle electric power inspection obstacle avoidance system based on sparse Bayesian inference. The system constructs a two-dimensional grid map based on digital map information and distributes an environment probability value for each grid in combination with point cloud data analysis; performing weighting processing on radar data according to the directivity and the target distance of the millimeter wave radar, and endowing different confidence coefficients in different directions and distances; fusing the current measurement grid map with a previously detected probability grid map based on the Bayesian theory to realize dynamic updating of the occupation grid map; by establishing a gravitational field model and applying a vertical virtual force when a resultant force is smaller than a set threshold value, the unmanned aerial vehicle is guided to avoid obstacles. The system also comprises a simulation verification environment which is used for testing the performance and stability of an obstacle avoidance algorithm, and solves the problems of inaccurate route planning and frequent manual intervention in traditional electric power inspection through employing a point cloud segmentation model of deep learning and a multi-sensor fusion scheme.

Description

Technical Field

[0001] The present invention relates to the field of environmental perception technology, and in particular to an unmanned aerial vehicle power inspection and obstacle avoidance system based on sparse Bayesian inference. Background Art

[0002] With the continuous expansion of China's power grid and the significant increase in electricity consumption, the safe and stable operation of the power grid is particularly important. In order to improve the inspection efficiency and accuracy of power equipment, the operation and maintenance department urgently needs an accurate, real-time, and high-frequency inspection method. In this context, the combination of drone technology and intelligent defect recognition algorithms has gradually become a mainstream power line inspection method. Although drone technology has been increasingly widely used in the field of overhead line inspection, the current inspection of distribution lines still faces many challenges. Specifically, there is a high reliance on manual operation, the overall inspection efficiency is low, and the existing technology still has a low degree of automation in inspection route planning and flight control.

[0003] During the route planning stage, professionals usually need to rely on pre-collected laser point cloud data to manually set the location of the drone's photo points and flight routes. This process is not only time-consuming, but also often has problems such as poor route consistency and low planning accuracy. During the inspection flight stage, the operator must continuously monitor the flight status of the drone, especially when the drone encounters buildings, trees or other obstacles, causing the positioning signal to attenuate, and then deviate from the original route and fail to complete the scheduled inspection task. Manual intervention is often required to adjust the flight path. The above problems highlight the low automation and low efficiency of the current inspection process.

[0004] In recent years, although some academic progress has been made in improving the efficiency of drones in the field of power inspection, most of the research has focused on transmission line inspection. However, compared with transmission lines, the environment faced by distribution line inspection is more complex. The poles and wires of the distribution network are usually low and close to obstacles such as buildings and trees. Changes caused by human activities and the natural environment (such as tree growth) can seriously affect the flight safety of drones. Therefore, distribution line inspection faces more complex challenges than transmission line inspection. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to solve the problem that in the route planning stage, professionals usually need to rely on pre-collected laser point cloud data to manually set the location of the drone's photo points and flight routes. This process is not only time-consuming, but also often has problems such as poor route consistency and low planning accuracy. During the inspection flight stage, the operator must continuously monitor the flight status of the drone, especially when the drone encounters buildings, trees or other obstacles, causing the positioning signal to attenuate, and then deviate from the original route and fail to complete the scheduled inspection task, which often requires manual intervention to adjust the flight path.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a UAV power inspection and obstacle avoidance system based on sparse Bayesian inference, which includes constructing a two-dimensional grid map based on digital map information;

[0009] The radar data is weighted according to the directionality of the millimeter-wave radar and the target distance;

[0010] Based on Bayesian theory, the current measurement grid map is combined with the previously detected probability grid map to update the current occupancy grid map;

[0011] Establish a target gravity field model to guide the drone to avoid obstacles.

[0012] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the construction of a two-dimensional grid map based on digital map information includes: constructing a two-dimensional grid map in the flight direction, dividing the flight path into a number of grid units; obtaining channel point cloud data, and analyzing the distribution of point cloud data in space; based on the grid position occupied by the point cloud, combined with the density and feature information of the point cloud, assigning an environmental probability value to each grid.

[0013] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the weighted processing of radar data according to the directionality of the millimeter-wave radar and the target distance includes: assigning a higher measurement confidence in the forward flight direction; assigning a lower measurement confidence in the lateral or backward direction; assigning a higher confidence to a target that is closer; and assigning a lower confidence to a target that is farther away.

[0014] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the updating of the occupancy grid map at the current moment based on Bayesian theory includes: obtaining the measurement grid map at the current moment; combining the measurement grid map with the previously detected probability grid map; calculating the probability of the grid being occupied according to Bayesian theory; and calculating the probability of each grid being unoccupied.

[0015] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the establishment of the target gravitational field model includes: calculating the gravitational force exerted on the UAV in the gravitational field; setting the gravitational coefficient; determining the maximum effective distance of the artificial potential field method; and calculating the gravity based on the distance from the UAV to the target.

[0016] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, it also includes: when the resultant force acting on the UAV is less than a set threshold, a virtual force is applied in the vertical upward direction; the virtual force is used to help the UAV get rid of the local oscillation state.

[0017] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the triggering conditions for applying virtual force in the vertical upward direction include: detecting the existence of obstacles in the vertical direction; the UAV falls into a local oscillation state; encountering vertical obstacles such as trees or crossed power lines.

[0018] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, it also includes: building a simulation environment for distribution network inspection on the Airsim simulation environment and the UnrealEngine4 platform; simulating high-quality models including trees, wires, insulators and towers; and simulating millimeter-wave radar signals and UAV flight dynamics at the same time.

[0019] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, the simulation environment is used to: verify whether the algorithm has the ability to escape from the local minimum area; test the performance and stability of the obstacle avoidance algorithm in scenarios where multiple obstacles exist; and test by randomly changing the position and number of obstacles multiple times.

[0020] As a preferred solution of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference described in the present invention, it also includes: a point cloud segmentation model based on deep learning, which is used to automatically classify the line channel point cloud; a joint perception solution combining millimeter wave radar and visual sensor; and a time-series-based millimeter wave radar noise filtering algorithm.

[0021] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as described in the first aspect of the present invention are implemented.

[0022] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as described in the first aspect of the present invention are implemented.

[0023] The beneficial effects of the present invention are as follows: by constructing a two-dimensional grid map based on digital map information and combining point cloud data analysis, accurate modeling of the flight environment is achieved, solving the problem of inaccurate traditional manual route planning. Through the directional weighted processing of millimeter wave radar data and the probability update mechanism of Bayesian theory, the system can dynamically perceive environmental changes and perform real-time map updates, improving the accuracy and timeliness of obstacle detection.

[0024] The system adopts an obstacle avoidance strategy that combines the artificial potential field method with vertical virtual force, effectively solving the problem that drones are prone to local oscillation in complex power line environments. At the same time, through the deep learning point cloud segmentation model and multi-sensor fusion solution, accurate identification and classification of the line channel environment are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 This is a flow chart of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference;

[0027] Figure 2 This is a computer equipment diagram of the UAV power inspection and obstacle avoidance system based on sparse Bayesian inference;

[0028] Figure 3 Schematic diagram of the obstacle avoidance path of the UAV in the UAV power inspection obstacle avoidance system based on sparse Bayesian inference. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from the description, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0032] Example 1

[0033] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a UAV power inspection and obstacle avoidance system based on sparse Bayesian inference, comprising:

[0034] S100: constructing a two-dimensional grid map based on digital map information;

[0035] In an embodiment of the present application, a two-dimensional grid map constructed based on digital map information includes the following three parts: constructing a two-dimensional grid map in the flight direction, dividing the flight path into a number of grid units; acquiring channel point cloud data, and analyzing the distribution of the point cloud data in space; and assigning an environmental probability value to each grid based on the grid position occupied by the point cloud, combined with the density and feature information of the point cloud.

[0036] Specifically, when constructing a two-dimensional grid map in the flight direction, the system first divides the entire flight space into uniform grids according to the preset grid size (such as 0.5m×0.5m). Each grid unit represents an area in the actual space, which is used for subsequent environmental perception and path planning. At the same time, the system determines the main flight direction according to the direction of the power line and the location of the tower, and establishes a coordinate system along this direction.

[0037] In an optional embodiment, the size of the grid can be dynamically adjusted according to the actual application scenario. For example, in a dense urban environment, a smaller grid (such as 0.3m×0.3m) can be set to improve map accuracy; in open areas, a larger grid (such as 1m×1m) can be used to reduce the computational burden. In addition, the system can also implement adaptive grid division, that is, use small grids in areas with dense obstacles and use large grids in open areas, so as to balance the computational efficiency and accuracy requirements.

[0038] In the embodiment of the present application, the method of obtaining channel point cloud data includes: real-time scanning using airborne laser radar; pre-loading high-precision three-dimensional map data; and fusing multi-source sensor data. Among them, laser radar scanning can obtain real-time environmental information, which is suitable for dynamic environments; pre-loaded map data provides basic static environmental information; multi-source sensor fusion can improve the reliability and integrity of data.

[0039] It should be noted that the analysis process of point cloud data includes three key steps: data preprocessing, feature extraction and semantic segmentation. In the preprocessing stage, the system will perform noise reduction, spatial downsampling and other operations to improve the efficiency of subsequent processing. The feature extraction stage mainly calculates the geometric features of the point cloud, such as normal vectors, curvature, etc. Semantic segmentation classifies point cloud data into different categories such as ground, vegetation, buildings, etc., to provide support for subsequent environmental modeling.

[0040] S101: Constructing a two-dimensional grid map based on digital map information includes: constructing a two-dimensional grid map in the flight direction, dividing the flight path into a number of grid units; acquiring channel point cloud data, and analyzing the distribution of the point cloud data in space; and assigning an environmental probability value to each grid based on the grid position occupied by the point cloud and combining the density and feature information of the point cloud.

[0041] S200: Weighted processing of radar data based on the directionality of the millimeter-wave radar and the target distance;

[0042] In an embodiment of the present application, the weighted processing of radar data mainly considers four aspects: assigning a higher measurement confidence in the forward flight direction; assigning a lower measurement confidence in the lateral or backward direction; assigning a higher confidence to targets at a closer distance; and assigning a lower confidence to targets at a farther distance.

[0043] Specifically, the system uses the cosine function to model the directionality, so that the weight of the forward direction (0°) is the largest, the side direction (90° or -90°) is the second largest, and the rear direction (180°) is the smallest. At the same time, an exponential decay function is used to describe the effect of distance on confidence, that is, the greater the distance, the lower the confidence. The final weight is a weighted combination of the direction weight and the distance weight.

[0044] In an optional embodiment, the weight calculation formula can be expressed as:

[0045]

[0046] Among them, θ is the angle of the target relative to the flight direction, d is the target distance, d0 is the characteristic distance (usually 1 / 3 of the effective detection range of the radar), and α and β are adjustable weight coefficients.

[0047] It should be noted that this weighted processing mechanism can effectively cope with the performance differences of millimeter-wave radar in different directions and distances, and provide more reliable environmental perception results. At the same time, the system also implements an adaptive weight adjustment function, which can dynamically adjust the weight parameters according to environmental conditions (such as weather, occlusion, etc.).

[0048] S201: weighted processing of radar data according to the directivity of the millimeter-wave radar and the target distance includes: assigning a higher measurement confidence in the forward flight direction; assigning a lower measurement confidence in the lateral or backward direction; assigning a higher confidence to a target at a closer distance; and assigning a lower confidence to a target at a farther distance.

[0049] S300: Based on the Bayesian theory, the current measurement grid map is combined with the previously detected probability grid map to update the current occupancy grid map;

[0050] In the embodiment of the present application, the occupancy grid map update process mainly includes four steps: obtaining the measurement grid map at the current moment; combining the measurement grid map with the previously detected probability grid map; calculating the probability of the grid being occupied according to Bayesian theory; and calculating the probability of each grid being unoccupied.

[0051] Specifically, the system first uses the measurement data of the millimeter-wave radar to generate the measurement grid map at the current moment. For each grid, its occupancy status is preliminarily determined based on whether it contains obstacle reflection points. Then, this preliminary determination result is fused with the historical observation data (i.e., the previous probability grid map), and the updated occupancy probability is calculated using the Bayesian formula.

[0052] In an optional embodiment, in order to improve the robustness of map updates, the system adopts a progressive update strategy. Specifically, instead of directly replacing the old probability value with the new observation, a smoothing factor (such as 0.3) is used to control the magnitude of the update. This can reduce the impact of a single erroneous observation on the map and improve the stability of the system. For example, when a sudden change in the occupancy state of a grid is detected, the system will continuously observe multiple time steps, and only when this change persists will the probability value of the grid be significantly changed.

[0053] It should be noted that the update of the probability grid map is a dynamic process that needs to take into account both the uncertainty of the measurement and the dynamic changes of the environment. The system better expresses the uncertainty of the environment by maintaining the probability distribution of each grid instead of a simple binary state (occupied / unoccupied). In addition, the system also implements a regional probability joint update mechanism, that is, when the state of a grid is updated, the influence of the adjacent grids will be considered, thereby improving the accuracy of the map update.

[0054] S301: Updating the occupancy grid map at the current moment based on Bayesian theory includes: obtaining the measurement grid map at the current moment; combining the measurement grid map with the previously detected probability grid map; calculating the probability of the grid being occupied according to Bayesian theory; and calculating the probability of each grid being unoccupied.

[0055] S400: Establish a target gravity field model to guide the drone to avoid obstacles.

[0056] In the embodiment of the present application, the target gravitational field model mainly includes four core components: calculating the gravitational force on the drone in the gravitational field; setting the gravitational coefficient; determining the maximum range of the artificial potential field method; and calculating the gravitational force based on the distance from the drone to the target. This method based on the artificial potential field can provide a smooth motion trajectory for the drone while effectively avoiding obstacles.

[0057] Specifically, the system first establishes a gravitational field based on the location of the target point (such as the next checkpoint). The magnitude of gravity decreases as the distance from the target increases, usually in the form of a quadratic function or exponential function. At the same time, for detected obstacles, the system will establish a corresponding repulsive field so that the drone can naturally avoid these areas. The setting of the gravity coefficient needs to balance the target attraction and obstacle repulsion to ensure that the drone can reach the target position safely and efficiently.

[0058] In an optional embodiment, the system can also implement dynamic gravitational field adjustment. For example, when multiple obstacles are detected ahead, the system will appropriately reduce the gravitational coefficient and increase the repulsive force, so that the drone has more maneuvering space to avoid obstacles. At the same time, the system will also dynamically adjust the range of action according to the drone's motion state (such as speed, acceleration), expand the potential field range during high-speed flight, and ensure sufficient reaction time.

[0059] In the embodiment of the present application, when the drone is trapped in a local minimum point (i.e., the gravitational force and the repulsive force are approximately balanced), the system will activate a special escape mechanism. Specifically, it includes:

[0060] Virtual force application: Apply an upward virtual force in the vertical direction to help the drone get rid of the shock state;

[0061] Potential field reconstruction: temporarily adjust the repulsive force distribution of surrounding obstacles to create a force field environment that is conducive to escaping;

[0062] Route memory: records the escape directions that have been tried to avoid repeating the same escape strategy.

[0063] It should be noted that the implementation of the entire obstacle avoidance system requires fine parameter tuning. The key parameters include:

[0064] Gravity coefficient Katt: controls the strength of the drone's movement toward the target point, usually set between 0.5-2.0;

[0065] Repulsion coefficient Krep: controls the repulsion strength of obstacles, usually set to 1.5-3 times the gravity coefficient;

[0066] Maximum range dmax: determines the range of influence of the force field, usually set to 2-3 times the maximum speed of the drone;

[0067] Safety threshold dsafe: defines the minimum safe distance between the drone and obstacles, usually set to 2-3 times the size of the drone.

[0068] S401: Establishing a target gravitational field model includes: calculating the gravitational force exerted on the UAV in the gravitational field; setting the gravitational coefficient; determining the maximum action distance of the artificial potential field method; and calculating the gravitational force based on the distance from the UAV to the target.

[0069] S402: Also includes: When the resultant force acting on the drone is less than a set threshold, applying a virtual force in the vertical upward direction; the virtual force is used to help the drone get rid of the local oscillation state.

[0070] S403: The triggering conditions for applying the virtual force in the vertical upward direction include: detecting the presence of an obstacle in the vertical direction; the drone falls into a local oscillation state; encountering vertical obstacles such as trees or crossed power lines.

[0071] S404: Also includes: building a simulation environment for distribution network inspection on the Airsim simulation environment and UnrealEngine4 platform; simulating high-quality models including trees, wires, insulators and towers; and simulating millimeter-wave radar signals and drone flight dynamics at the same time.

[0072] S405: The simulation environment is used to: verify whether the algorithm has the ability to escape from the local minimum area; test the performance and stability of the obstacle avoidance algorithm in scenarios with multiple obstacles; and test by randomly changing the position and number of obstacles multiple times.

[0073] The next moment's position update of the UAV is calculated based on the combined force.

[0074] In the embodiment of the present application, the position update calculation needs to comprehensively consider the following factors: the current motion state (position, speed, posture); the acceleration generated by the combined force; dynamic constraints; and environmental factors (such as wind speed). The system adopts a hierarchical control strategy, which decomposes the position update into a path planning layer and a trajectory tracking layer.

[0075] Specifically, the system first determines the desired direction of motion based on the direction of the resultant force, and then calculates the feasible control input through the dynamic model. The position update formula uses a discrete time model:

[0076] x(t+1)=x(t)+v(t)dt+0.5a(t)×dt 2

[0077] Where x is the position vector, v is the velocity vector, a is the acceleration vector, and dt is the control period (usually set to 20ms).

[0078] In an optional embodiment, the system implements an adaptive location update strategy:

[0079] In areas with sparse obstacles, use larger speed increments to improve flight efficiency;

[0080] In areas with dense obstacles, the speed increment is automatically reduced to ensure control accuracy;

[0081] When a sudden obstacle is detected, the robot can quickly decelerate or hover to allow sufficient reaction time.

[0082] In the embodiment of the present application, the location update also includes multiple protection mechanisms:

[0083] Speed ​​limit: By combining soft constraints and hard constraints, the speed of the drone is always within a safe range;

[0084] Acceleration limit: Considering the dynamic characteristics of the drone, the maximum acceleration and angular acceleration are limited;

[0085] Attitude protection: Prevent the drone from experiencing large attitude angle changes and maintain a stable flight state.

[0086] It should be noted that the calculation of location updates needs to take into account the performance limitations of actual hardware. The system adopts the following optimization measures:

[0087] Use quaternions to represent attitude, reduce calculation and avoid gimbal deadlock;

[0088] Realize the prediction function of state estimation and compensate for the impact of control delay;

[0089] The piecewise linearization method is used to simplify the dynamic model to improve the computational efficiency while ensuring accuracy.

[0090] S406: Also includes: a point cloud segmentation model based on deep learning for automatically classifying line channel point clouds; a joint perception solution combining millimeter wave radar and visual sensors; and a millimeter wave radar noise filtering algorithm based on timing.

[0091] In summary, by constructing a two-dimensional grid map based on digital map information, the automatic segmentation of the flight path and the dynamic allocation of environmental probability values ​​are achieved, thereby solving the problems of poor consistency and low planning accuracy in the traditional manual setting of routes, enabling drones to perceive and plan flight paths more accurately.

[0092] Through the directionality and distance-weighted processing mechanism of millimeter-wave radar data, the system can assign differentiated confidence levels based on the different characteristics of the target location, effectively improving the accuracy of obstacle detection, especially the perception capabilities in complex power line environments, and reducing detection errors caused by sensor performance limitations.

[0093] The scheme of updating the grid map through Bayesian theory integrates the current measurement with the historical observation data, which not only improves the timeliness of environmental perception, but also improves the robustness of map updates through probabilistic inference, effectively avoiding the impact of a single erroneous observation on the system.

[0094] By establishing a target gravitational field model and combining it with the design of vertical virtual force, the system can adaptively adjust its flight strategy when encountering vertical obstacles such as wires and trees, effectively solving the problem that traditional obstacle avoidance algorithms are prone to oscillation at local minimum points.

[0095] By introducing a simulation verification environment, the system can verify algorithm performance in a virtual scene, especially the obstacle avoidance capability in a multi-obstacle environment, which greatly reduces the risk and cost of actual testing while improving the reliability and adaptability of the algorithm.

[0096] Through the deep learning point cloud segmentation model and multi-sensor fusion solution, the system can accurately identify and classify the line channel environment. At the same time, it improves the quality of perception data through time series noise filtering, thereby ensuring stable operation in complex environments.

[0097] Example 2

[0098] Reference Figure 2 - Figure 3 , which is the second embodiment of the present invention, and this embodiment provides a UAV power inspection and obstacle avoidance system based on sparse Bayesian inference.

[0099] First, a deep learning-based point cloud segmentation model is used to automatically classify and process the point cloud data in the line channel, and a digital three-dimensional model of the inspection target is constructed. Through this model, the system can automatically generate the photo points and flight routes of the drone. Secondly, a joint perception solution of millimeter-wave radar and visual sensor is combined, and a time-series millimeter-wave radar noise filtering algorithm is used to design an efficient real-time obstacle avoidance system. This system enables the drone to fly autonomously in complex environments and accurately avoid obstacles.

[0100] Based on the information of the digital map, a two-dimensional grid map is constructed in the flight direction. In this map, the flight path will be divided into several small grid units, each grid represents a certain spatial area, which is convenient for subsequent environmental perception and path planning. Then, by acquiring the point cloud data of the channel, the distribution of the point cloud data in space is analyzed. According to the grid position occupied by the point cloud, combined with the density and feature information of the point cloud, an environmental probability value is assigned to each grid. This value reflects whether there may be obstacles or other environmental elements in the area. In this way, the initialization of the grid map can be completed.

[0101] Considering that the signal strength reflected by millimeter-wave radar in the forward direction is usually higher than that in the lateral direction, and its detection capability is stronger for closer targets, we can weight the radar data according to the radar's directivity and target distance. In this case, different confidence values ​​should be assigned to the radar's measurement data in different directions and at different distances. Specifically, in the forward flight direction, the reflection intensity of the radar signal is greater, which means that the confidence of its measurement results is higher; while in the lateral or backward direction, the reflection intensity of the radar signal is weaker, and the corresponding confidence is lower. In addition, the reflection signal of closer targets is stronger, and the radar's detection capability for these targets is also stronger. Therefore, the measurement values ​​at closer distances will be assigned higher confidence; conversely, for targets at a longer distance, the radar's detection capability is weaker, and the corresponding measurement values ​​have lower confidence.

[0102] Based on this principle, in the grid map of Airsim at that moment, the measurement value Zt of each grid will be adjusted according to the strength, direction and target distance of the millimeter-wave radar reflection signal. This confidence-based method can effectively reflect environmental information in different directions and distances, thereby providing more accurate obstacle information when constructing the grid map and optimizing subsequent path planning and obstacle avoidance decisions.

[0103] After obtaining the measurement grid map at the moment, we need to combine it with the previously detected probability grid map to update the occupancy grid map at the current moment. This process can be achieved through Bayesian theory, and Bayesian updating provides an effective method to correct the previous probability distribution based on new observation data. After obtaining the measurement grid map at the moment, combine it with the previously detected probability grid map, and use Bayesian theory to calculate the occupancy grid map at the current moment. According to Bayesian theory, the relationship between the measurement grid map and the occupancy grid map is:

[0104]

[0105] The update process is:

[0106]

[0107] get:

[0108]

[0109] Where t is the time; Zt is the measurement grid map; Z1:t- 1 is the probability grid map; mt is the placeholder grid map; Z 1∶ t is the set of observed data from the initial time to time t; m is the intermediate grid state used for state transfer or probability update; p is the probability.

[0110] Using the same method, we can calculate the probability of each grid being unoccupied. Specifically, through the Bayesian update process, we can not only update the probability of the grid being occupied, but also calculate the probability of each grid being unoccupied based on the current measurement data and previous information.

[0111]

[0112] At this point, the probability value of the updated probability grid map depends only on the probability of the current grid, the probability of the previous moment, and the prior probability at the initial moment. Finally, the grid map after Bayesian update is:

[0113]

[0114] Among them, m1:t is the grid state from t=1 to the current moment; L(1:t) is the logarithmic probability, as follows:

[0115]

[0116] First, the gravitational field model of the target is established. The gravitational force on the drone in this gravitational field can be expressed as:

[0117]

[0118] Where ξ is the gravitational coefficient; ρ is the distance from the UAV to the target, and D is the maximum action distance of the artificial potential field method.

[0119] When the combined force on the drone is less than the set threshold Fthreshold, it may fall into a local oscillation state. To avoid this problem, we will apply a non-zero virtual force to help the drone get rid of the oscillation. Considering that the drone will mainly encounter vertical obstacles during flight, such as trees or crossed power lines, our strategy is to prioritize safety, so we will apply a virtual force Fup in the vertical upward direction to guide the drone to avoid obstacles upward.

[0120] Finally, the position of the drone at the next moment is calculated based on the combined force:

[0121]

[0122] Among them, xt is the position of the UAV at time t; w is the total mass of the UAV; s is the step length; xt-1 is the position of the UAV at time t-1; Fsum is the resultant force on the UAV;

[0123] Considering that the signal strength reflected by millimeter-wave radar in the forward direction is usually higher than that in the lateral direction, and its detection capability is stronger for closer targets, we can weight the radar data according to the radar's directivity and target distance. In this case, different confidence values ​​should be assigned to the radar's measurement data in different directions and at different distances. Specifically, in the forward flight direction, the reflection intensity of the radar signal is greater, which means that the confidence of its measurement results is higher; while in the lateral or backward direction, the reflection intensity of the radar signal is weaker, and the corresponding confidence is lower. In addition, the reflection signal of closer targets is stronger, and the radar's detection capability for these targets is also stronger. Therefore, the measurement values ​​at closer distances will be assigned higher confidence; conversely, for targets at a longer distance, the radar's detection capability is weaker, and the corresponding measurement values ​​have lower confidence.

[0124] Based on this principle, in the grid map of Airsim at that moment, the measurement value Zt of each grid will be adjusted according to the strength, direction and target distance of the millimeter-wave radar reflection signal. This confidence-based method can effectively reflect environmental information in different directions and distances, thereby providing more accurate obstacle information when constructing the grid map and optimizing subsequent path planning and obstacle avoidance decisions.

[0125] By locating the towers in the digital map and measuring the distance between the poles, the grid positions and sizes of the occupancy probability grid map are initialized. For the grid containing the laser point, the occupancy probability is set to 100%, and for the unoccupied area, the occupancy probability is set to 50%.

[0126] According to the position and attitude information of the sensor at time t, the fused obstacle position coordinates are converted into the grid map coordinate system, and the occupancy probability of the corresponding grid is set to 70%.

[0127] In order to train the point cloud classification model, this paper used the distribution network line channel data collected by the operation and maintenance team of a provincial power supply bureau through drones. Under the guidance of professionals, these data were annotated and multiple valid files were generated. Specifically, 5578 valid files were generated, of which 1353 were used for training sets, 2789 for validation sets, and 1436 for test sets. The final accuracy rate reached 91.23%.

[0128] A simulation environment for distribution network inspection was built on the Airsim simulation environment and UnrealEngine4 platform. The environment contains high-quality models of trees, wires, insulators, and towers, and simultaneously simulates millimeter-wave radar signals and drone flight dynamics, enabling systematic joint testing of millimeter-wave radar perception and obstacle avoidance algorithms.

[0129] In the embodiment of the present application, the simulation verification environment is constructed based on the Airsim and UnrealEngine4 platforms, and mainly includes the following components: a high-fidelity scene model (including trees, wires, insulators and towers); a sensor simulation module (simulating the signal characteristics of millimeter-wave radar and other sensors); and a complete drone dynamics model.

[0130] Specifically, the simulation environment supports the following functions:

[0131] Scene editing: You can customize the location, size and type of obstacles;

[0132] Weather effects: simulate the impact of different weather conditions on sensors;

[0133] Real-time data logging: records flight trajectories, control inputs, and sensor data;

[0134] Visual analysis: Provides three-dimensional view and data curve display.

[0135] In an optional embodiment, the simulation environment also implements the following advanced functions:

[0136] Hardware-in-the-loop simulation: can be connected to the actual flight control system for testing;

[0137] Multi-machine collaborative simulation: supports scenario testing of multiple drones running simultaneously;

[0138] Fault injection: Simulate various sensor failures and control system abnormalities.

[0139] It should be noted that the following points should be noted when building a simulation verification environment:

[0140] Model accuracy: In particular, the signal propagation model of millimeter-wave radar needs to be calibrated with actual data;

[0141] Real-time performance: ensure that the simulation system can respond in real time and avoid delay accumulation;

[0142] Interface compatibility: Uses the same data interface as the actual system to facilitate code porting.

[0143] The first experiment aims to verify whether the proposed algorithm has the ability to escape the local minimum area. In the distribution network simulation environment, a single tree obstacle is set at a horizontal distance of 35m from the starting tower, and the height of the tree is 16m. The test results are shown in the attached Figure 3 When the drone is 8 meters away from the tree, the gravitational force is equal to the repulsive force, causing it to fall into local oscillation. At this time, the improved artificial potential field method guides the drone to fly upward by applying an upward gravitational force, thereby maintaining a safe distance and avoiding the tree.

[0144] In the second experiment, the performance and stability of the obstacle avoidance algorithm in the presence of multiple obstacles were tested. In the experiment, the traditional artificial potential field algorithm and the improved artificial potential field algorithm proposed in this paper were run by randomly changing the position and number of obstacles many times, and the obstacle avoidance success rates of the two algorithms were counted. In 20 repeated tests, the average success rate of the obstacle avoidance algorithm based on the improved artificial potential field method exceeded 90%, which was significantly higher than the success rate of the traditional method.

[0145] Example 3

[0146] This embodiment also provides a computer device, which is suitable for a UAV power inspection and obstacle avoidance system based on sparse Bayesian inference, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a forced oscillation detection and positioning method for a distribution network as proposed in the above embodiment.

[0147] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0148] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0149] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0150] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, apparatus, or device and execute instructions), or in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device.

[0151] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0152] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A UAV power inspection and obstacle avoidance system based on sparse Bayesian inference, characterized by: include, Construct a two-dimensional grid map based on digital map information; The radar data is weighted according to the directionality of the millimeter-wave radar and the target distance; Based on Bayesian theory, the current measurement grid map is combined with the previously detected probability grid map to update the current occupancy grid map; Establish a target gravity field model to guide the drone to avoid obstacles.

2. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 1, characterized in that: The method of constructing a two-dimensional grid map based on digital map information includes: constructing a two-dimensional grid map in the flight direction, dividing the flight path into a number of grid units; acquiring channel point cloud data, and analyzing the distribution of the point cloud data in space; and assigning an environmental probability value to each grid based on the grid position occupied by the point cloud and combining the density and feature information of the point cloud.

3. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 2, characterized in that: The weighted processing of radar data according to the directionality of the millimeter-wave radar and the target distance includes: assigning a higher measurement confidence in the forward flight direction; assigning a lower measurement confidence in the lateral or backward direction; assigning a higher confidence to a target at a closer distance; and assigning a lower confidence to a target at a farther distance.

4. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 3, characterized in that: The updating of the current occupancy grid map based on Bayesian theory includes: obtaining the current measurement grid map; combining the measurement grid map with the previously detected probability grid map; calculating the probability of the grid being occupied according to Bayesian theory; and calculating the probability of each grid being unoccupied.

5. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 4, characterized in that: The establishment of the target gravitational field model includes: calculating the gravitational force exerted on the UAV in the gravitational field; setting the gravitational coefficient; determining the maximum action distance of the artificial potential field method; and calculating the gravitational force based on the distance from the UAV to the target.

6. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 5, characterized in that: Also includes: When the resultant force on the drone is less than the set threshold, a virtual force is applied in the vertical upward direction; The virtual force is used to help the drone get rid of the local shock state.

7. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 6, characterized in that: The triggering conditions for applying the virtual force in the vertical upward direction include: detecting the presence of an obstacle in the vertical direction; the drone falls into a local oscillation state; encountering vertical obstacles such as trees or crossed power lines.

8. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 7, characterized in that: Also includes: Build a simulation environment for distribution network inspection on the Airsim simulation environment and UnrealEngine4 platform; Simulate high-quality models of trees, conductors, insulators, and towers; simultaneously simulate millimeter-wave radar signatures and drone flight dynamics.

9. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 8, characterized in that: The simulation environment is used to: verify whether the algorithm has the ability to escape from the local minimum area; test the performance and stability of the obstacle avoidance algorithm in a scenario with multiple obstacles; and perform the test by randomly changing the position and number of obstacles multiple times.

10. The UAV power inspection and obstacle avoidance system based on sparse Bayesian inference as claimed in claim 9, characterized in that: Also includes: A deep learning-based point cloud segmentation model for automatic classification of line channel point clouds; a joint perception solution combining millimeter wave radar and visual sensors; A time-series based millimeter wave radar noise filtering algorithm is used.

Images