A method for optimizing energy consumption of an unmanned aerial vehicle

Abstract

The application discloses a kind of unmanned aerial vehicle energy consumption optimization track planning method, the present application relates to unmanned aerial vehicle energy consumption optimization technical field, including the following steps: step one: multiple sensor fusion system is arranged on unmanned aerial vehicle, multiple sensor fusion system includes vision sensor, IMU inertial measurement unit and GPS backup module.The unmanned aerial vehicle energy consumption optimization track planning method, through the track optimization algorithm based on machine learning is combined with current position information, original energy consumption optimization track and surrounding environmental information, the shortest connection path of connecting path from current offset position is connected back to original track is calculated in real time Energy efficiency is highest;And according to the remaining power of unmanned aerial vehicle, target point distance optimization obtains improved path, ensure that unmanned aerial vehicle can smoothly transition back to original energy consumption optimization track and continue energy-saving flight;It improves the continuity and reliability of unmanned aerial vehicle in complex environment to execute task, reduces the risk that task fails or energy consumption increases significantly due to signal interruption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle energy consumption optimization, in particular to a flight path planning method for unmanned aerial vehicle energy consumption optimization. BACKGROUND

[0002] With the wide application of unmanned aerial vehicle technology in various fields, such as warehouse inventory, geographic mapping, environmental monitoring, etc., the requirements for unmanned aerial vehicle flight path planning are also increasing; in a multi-warehouse environment, especially in the warehouse inventory scene, the unmanned aerial vehicle faces complex geographical environment and strict airspace control; on the one hand, the shelves are densely distributed in the warehouse, and the distribution of warehouse goods is dense and crowded, which brings many challenges to the flight of the unmanned aerial vehicle; on the other hand, the warehouse may block the signal transmission between the unmanned aerial vehicle and the ground control station, resulting in signal interruption; once the signal is interrupted, the unmanned aerial vehicle cannot accurately obtain its own position information and receive instructions from the ground, so it is easy to deviate from the preset energy consumption optimization flight path; moreover, when flying in the internal area of the warehouse, the unmanned aerial vehicle also needs to frequently adjust the flight attitude to avoid the goods stored in the warehouse, which not only increases the complexity of flight, but also leads to an increase in energy consumption.

[0003] That is, in the warehouse goods inventory scene, the unmanned aerial vehicle needs to fly according to the energy consumption optimization flight path planning method in the environment with complex shelf structure and multiple no-fly zones, but due to the signal interruption caused by the warehouse blocking, the unmanned aerial vehicle may deviate from the preset energy consumption optimization flight path, how to ensure that the unmanned aerial vehicle can quickly return to the optimal flight path and continue to fly energy-efficiently after the signal is restored. SUMMARY

[0004] To solve the above technical problems, the present application is implemented by the following technical scheme: a flight path planning method for unmanned aerial vehicle energy consumption optimization, comprising the following steps:

[0005] Step 1: A multi-sensor fusion system is arranged on the unmanned aerial vehicle, which includes a vision sensor, an IMU inertial measurement unit and a GPS backup module;

[0006] Step 2: The unmanned aerial vehicle flies in the warehouse according to the preset original energy consumption optimization flight path, when the signal is interrupted due to the blocking of the warehouse, the vision sensor acquires image data of the surrounding warehouse and environment, and the IMU inertial measurement unit continuously records the motion state parameters of the unmanned aerial vehicle;

[0007] Step 3: A three-dimensional local map of the surroundings of the unmanned aerial vehicle is constructed according to the image data using vision SLAM technology, and the relative displacement and attitude change of the unmanned aerial vehicle during the signal interruption period are calculated in combination with the motion state parameters to obtain the current position information;

[0008] Step four: After the signal is restored, compare the current position information with the original energy consumption optimized flight path, solve the shortest connection path from the current offset position to the original flight path through the path planning algorithm, and get the connection path;

[0009] Step five: optimize the strategy instruction according to the improved path, get the optimized strategy instruction, and execute the maneuvering adjustment based on the optimized strategy instruction to make the UAV smoothly transition back to the original energy consumption optimized flight path and continue to fly.

[0010] Preferably, in the multi-sensor fusion system, the visual sensor is an omnidirectional camera or a camera with wide-angle characteristics, which is used to visually cover the surrounding environment during signal interruption.

[0011] Preferably, the motion state parameters include acceleration, angular velocity and heading information, and are continuously accumulated during signal interruption, which are used to assist in estimating the relative position change of the UAV when the visual SLAM technology cannot provide the current position information.

[0012] Preferably, the three-dimensional local map is a sparse point cloud map or a key feature point map constructed by the visual SLAM technology during signal interruption, so as to quickly correct the position of the UAV and calculate the path access after the signal is restored.

[0013] Preferably, the process of obtaining the current position information is as follows:

[0014] A data synchronization mechanism of the visual sensor and the IMU inertial measurement unit is established, the image data collected by the visual sensor is preprocessed, including image filtering to remove noise interference, image enhancement to improve feature recognition, and the motion state parameters are filtered to remove measurement noise and outliers, improving data quality;

[0015] Feature extraction algorithms are used to extract representative feature points from the preprocessed image data, including corner points and edge points; through the optical flow method and feature descriptor matching technology, the feature points in the continuous frame image are tracked and matched to determine the corresponding relationship of the feature points between different image frames, and the motion trajectory of the feature points is obtained;

[0016] According to the feature point motion trajectory and the camera imaging model, the position change of the feature points in the three-dimensional space is calculated by the triangulation method, the displacement of the UAV in the horizontal, vertical and depth directions is derived, and the displacement estimation result of the UAV is preliminarily estimated.

[0017] According to the acceleration and angular velocity data recorded by the IMU inertial measurement unit, the attitude change of the unmanned aerial vehicle is calculated by using an inertial navigation algorithm; first, the acceleration data is integrated to obtain velocity change data, and then the velocity change data is integrated to obtain displacement change, and the zero speed update ZUPT is used to correct the error; and the angular displacement of the unmanned aerial vehicle is calculated through the gyroscope data, the change data of the heading, pitch and roll attitude angles are determined, and the attitude estimation result is obtained;

[0018] The displacement estimation result based on the visual SLAM technology and the attitude estimation result based on the IMU inertial measurement unit are fused by using the extended Kalman filter EKF or the unscented Kalman filter UKF to obtain fusion information; during the fusion process, according to the uncertainty of the two data sources, the weight is dynamically adjusted through the covariance matrix, so as to improve the accuracy of the fusion result;

[0019] The previously estimated relative displacement and attitude change are optimized and adjusted by using the fusion information, and are iteratively updated, so that the measurement result is closer to the real unmanned aerial vehicle motion state, the deviation caused by single sensor error or environmental interference is reduced, and finally the accurate relative displacement and attitude change data are obtained, so as to determine the current position information of the unmanned aerial vehicle during the signal interruption period.

[0020] Preferably, the fourth step includes a machine learning-based trajectory optimization algorithm, which compares the current position information with the original energy consumption optimized trajectory, calculates the shortest and most energy efficient flight path of the connection path connecting the current offset position back to the original trajectory in real time, and obtains the connection path.

[0021] Preferably, the fourth step further includes optimizing the connection path during the connection path calculation process according to the current remaining power of the unmanned aerial vehicle, the distance to the target point, and the surrounding environment constraints, to obtain an improved path, wherein the surrounding environment constraints include warehouse distribution, no-fly zone position, and weather conditions; so as to ensure that the unmanned aerial vehicle can still maintain the energy consumption optimization characteristics of subsequent navigation after the connection is completed.

[0022] Preferably, the path planning algorithm adopts a multi-objective optimization strategy when calculating the connection path, including shortest distance, minimum turning radius, avoiding no-fly zones, and comprehensive energy consumption indicators, and solves the flight trajectory that meets the predetermined optimization criteria through weight allocation, wherein the predetermined optimization criteria include shortest path length, lowest energy consumption, flight safety, attitude adjustment smoothness, and remaining power utilization efficiency criteria.

[0023] Preferably, the process of obtaining the connection path is as follows:

[0024] The current position information determined by the UAV during the signal interruption is imported, original energy consumption optimized flight path data is acquired, surrounding environment information of the UAV is fused, and a comprehensive data model containing position and environment constraints is constructed; key features are extracted from the current position and the original energy consumption optimized flight path, including key navigation points of the original energy consumption optimized flight path, geometric shape features, and the orientation and distance features of the current position relative to the original energy consumption optimized flight path;

[0025] A graph structure representation of the original energy consumption optimized flight path is constructed, with navigation points as nodes, connection relationships as edges, and edge weights being set, and the current position is represented as a special node related to the graph structure; a deep reinforcement learning algorithm is selected as the machine learning algorithm type, a model structure is constructed, and input layer, output layer and hidden layer parameters are determined, wherein the input layer includes current position features, original energy consumption optimized flight path features and environment features; the output layer includes an action sequence or a path node sequence of the connection path; simulation or historical flight data is collected, the machine learning model is trained according to the set reward function, and the model parameters are optimized to learn the optimal connection strategy; wherein the reward function takes short path length and low energy consumption as high reward basis;

[0026] The current position information, original energy consumption optimized flight path information and environment information are input into the trained machine learning model to generate a candidate path sequence or action sequence for connecting back to the original energy consumption optimized flight path from the current offset position, and a weighted summation method is used to evaluate and screen the candidate paths according to the target weights of path length, energy consumption, flight safety, to determine the final connection path.

[0027] Preferably, the process of obtaining the improved path is as follows:

[0028] Residual power evaluation: according to the flight performance model of the UAV, the estimated energy consumption of each path segment of the connection path is compared to determine the current residual power of the UAV; path segments that cannot be supported by the current residual power are screened out and removed from the connection path to obtain a path subset that can be supported by the residual power;

[0029] Target point distance analysis: the distance from the current position to the target point of the original flight path and the distance from the end points of the path subset that can be supported by the residual power to the target point are calculated, and the path is adjusted according to the relationship between the two distances; to avoid the selection of the path leading to the target point being too far away, which may affect the overall energy consumption and flight efficiency;

[0030] If the path is within the range of the residual power, but will make the UAV deviate from the target point too far, resulting in a significant increase in the subsequent flight distance and thus affecting the overall energy consumption and flight efficiency, the path is modified: by inserting a new navigation point in the middle of the path or changing the direction of the path, the path is made closer to the direction of the target point of the original flight path under the premise of ensuring sufficient residual power;

[0031] Surrounding environment constraint processing: warehouse distribution processing: with the help of image data or pre-stored warehouse distribution data, the relationship between the connection path and the warehouse space is analyzed, and for the path segment close to the warehouse that affects safety or has high energy consumption when flying around, the flight height is adjusted, the direction is changed, and the complex warehouse area is avoided, so as to obtain the path after the warehouse distribution is optimized; that is, for the path segment close to the warehouse and affecting flight safety or causing too high energy consumption due to flying around the warehouse, the flight height is adjusted, the path direction is changed, and the complex warehouse area is avoided, so as to obtain the path after the warehouse distribution is optimized;

[0032] No-fly zone position processing: according to the no-fly zone position information, it is checked whether the connection path crosses the no-fly zone; if there is a crossing situation, a path planning algorithm is used to re-plan a path segment outside the boundary of the no-fly zone, and the original path segment crossing the no-fly zone is replaced by a path segment bypassing the no-fly zone, so as to ensure that the path is legal and the energy consumption increment of bypassing is minimized;

[0033] Weather condition processing: obtain weather condition data, including wind direction and wind speed information, for the path segment flying against the wind, calculate the energy consumption increase of flying against the wind according to the aerodynamic performance model of the unmanned aerial vehicle (such as the resistance coefficient change curve under different wind direction angles); if the energy consumption increase is too large, adjust the path direction to obtain a detour path segment, including: finding a more favorable wind direction area to fly, such as adopting a detour strategy, flying along the tangential direction of the wind for a distance and then turning to the target point direction to reduce the overall energy consumption; at the same time, considering the influence of wind speed on the flight stability of the unmanned aerial vehicle, for the path segment with too large wind speed that may cause the unmanned aerial vehicle to be difficult to control, the flight speed is appropriately reduced or the flight height is adjusted to ensure flight safety, and the path length and flight time are correspondingly optimized to maintain energy consumption optimization;

[0034] Based on the path after the remaining power evaluation, target point distance analysis and surrounding environment constraint processing, the path is optimized again by using the weighted summation method to determine the improved path that meets the current remaining power of the unmanned aerial vehicle, the distance of the target point and the surrounding environment constraint condition, which is the final result obtained in the connection path calculation process and is used for subsequent flight control and navigation operation; when the improved path is determined, the control unit sends adjustment instructions to the flight control system of the unmanned aerial vehicle, including height fine adjustment, heading change and speed control, so that the unmanned aerial vehicle can smoothly complete the route correction without significantly increasing the energy consumption.

[0035] The application provides a flight path planning method for optimizing energy consumption of an unmanned aerial vehicle, which has the following beneficial effects:

[0036] I. The unmanned aerial vehicle energy consumption optimization flight path planning method, through the multi-sensor fusion system arranged on the unmanned aerial vehicle, the relative displacement and attitude change of the unmanned aerial vehicle can be accurately calculated during signal interruption by using visual SLAM technology combined with IMU data, so as to determine the current position information; This makes the unmanned aerial vehicle in the complex warehouse environment, even if the signal is lost, it can also accurately know its own position, which lays a foundation for subsequent flight path recovery;

[0037] II. The unmanned aerial vehicle energy consumption optimization flight path planning method, when the signal is restored, the flight path optimization algorithm based on machine learning combined with the current position information, the original energy consumption optimization flight path and the surrounding environment information, can calculate the shortest and most energy efficient connection path from the current deviation position to the original flight path; And in the connection path calculation process, the improved path can also be further optimized according to the current remaining power of the unmanned aerial vehicle and the target point distance, to ensure that the unmanned aerial vehicle can smoothly transition back to the original energy consumption optimization flight path and continue to fly in an energy-saving manner; It improves the continuity and reliability of the unmanned aerial vehicle in executing tasks in complex environments, reduces the risk of task failure or significant increase in energy consumption caused by signal interruption; It can make the unmanned aerial vehicle fly in the warehouse inventory scene according to the energy consumption optimization flight path planning method in the environment with many shelf structures and multiple no-fly zones, solve the problem that the signal interruption caused by warehouse shielding may make the unmanned aerial vehicle deviate from the preset energy consumption optimization flight path, and ensure that the unmanned aerial vehicle can quickly return to the optimal flight path and continue to fly in an energy-saving manner after the signal is restored. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flowchart of the unmanned aerial vehicle energy consumption optimization flight path planning method of the present application. DETAILED DESCRIPTION

[0039] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments of the present application are given for the purpose of example and description, and are not exhaustive or limit the present application to the disclosed forms. Many modifications and changes will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present application, and to enable those of ordinary skill in the art to understand the present application so as to design various embodiments with various modifications suitable for specific purposes.

[0040] As Figure 1 shown, the present application provides a technical solution: a flight path planning method for unmanned aerial vehicle energy consumption optimization, comprising the following steps:

[0041] Step 1: Arrange a multi-sensor fusion system on the unmanned aerial vehicle, which includes a visual sensor, an IMU inertial measurement unit and a GPS backup module;

[0042] Step two: the UAV flies along the preset energy consumption optimization flight path in the warehouse, when the signal is interrupted due to warehouse shielding, the visual sensor acquires image data of the surrounding warehouse and environment, and the IMU inertial measurement unit continuously records the UAV motion state parameters;

[0043] Step three: a three-dimensional local map around the UAV is constructed according to the image data using visual SLAM technology, and the relative displacement and attitude change of the UAV during the signal interruption period are calculated in combination with the motion state parameters, to obtain the current position information;

[0044] Step four: after the signal is restored, the current position information is compared with the original energy consumption optimization flight path, and the shortest connection path from the current offset position to the original flight path is solved through path planning algorithm to obtain the connection path;

[0045] Step five: the strategy instruction is optimized according to the improved path to obtain the optimized strategy instruction, and the UAV is executed based on the optimized strategy instruction to make the UAV smoothly transition back to the original energy consumption optimization flight path and continue to fly.

[0046] In the multi-sensor fusion system, the visual sensor is an omnidirectional camera or a camera with wide-angle characteristics, which is used to visually cover the surrounding environment during signal interruption.

[0047] The motion state parameters include acceleration, angular velocity and heading information, and are continuously accumulated during signal interruption, which are used to assist in estimating the relative position change of the UAV when the visual SLAM technology cannot provide the current position information.

[0048] The three-dimensional local map is a sparse point cloud map or key feature point map constructed by visual SLAM technology during signal interruption, so as to quickly correct the position of the UAV and calculate the path access after the signal is restored.

[0049] The process of obtaining the current position information is as follows:

[0050] A data synchronization mechanism of the visual sensor and the IMU inertial measurement unit is established, the image data collected by the visual sensor is preprocessed, including image filtering to remove noise interference and image enhancement to improve feature recognition, and the motion state parameters are filtered to remove measurement noise and outliers, improving data quality;

[0051] Feature extraction algorithms are used to extract representative feature points from the preprocessed image data, including corner points and edge points; through optical flow method and feature descriptor matching technology, the feature points in the continuous frame image are tracked and matched to determine the corresponding relationship of the feature points between different image frames, and the motion trajectory of the feature points is obtained;

[0052] According to the feature point motion trajectory and the camera imaging model, the position change of the feature point in the three-dimensional space is calculated by the triangulation method, and the displacement of the unmanned aerial vehicle in the horizontal, vertical and depth directions is derived, and the displacement estimation result of the unmanned aerial vehicle is preliminarily estimated.

[0053] According to the acceleration and angular velocity data recorded by the IMU inertial measurement unit, the attitude change of the unmanned aerial vehicle is calculated by using the inertial navigation algorithm; first, the acceleration data is integrated to obtain the velocity change data, then the velocity change data is integrated to obtain the displacement change, and the zero speed update ZUPT is used to correct the error; and the angular displacement of the unmanned aerial vehicle is calculated through the gyroscope data, the change data of the heading, pitch and roll attitude angles are determined, and the attitude estimation result is obtained.

[0054] The displacement estimation result based on the visual SLAM technology and the attitude estimation result based on the IMU inertial measurement unit are fused by using the extended Kalman filter EKF or the unscented Kalman filter UKF to obtain the fusion information; in the fusion process, according to the uncertainty of the two data sources, the weight is dynamically adjusted through the covariance matrix, so as to improve the accuracy of the fusion result.

[0055] The relative displacement and attitude change estimated before are adjusted and optimized by using the fusion information, and are iteratively updated, so that the measurement result is closer to the real motion state of the unmanned aerial vehicle, the deviation caused by the error of a single sensor or environmental interference is reduced, and finally the accurate relative displacement and attitude change data are obtained to determine the current position information of the unmanned aerial vehicle during the signal interruption.

[0056] The fourth step includes a machine learning-based trajectory optimization algorithm, which compares the current position information with the original energy consumption optimized trajectory, calculates the shortest and most energy efficient flight path that connects the original trajectory from the current offset position in real time, and obtains the connection path.

[0057] In the fourth step, the connection path is optimized according to the current remaining power of the unmanned aerial vehicle, the distance to the target point, and the surrounding environment constraints during the connection path calculation process, and the improved path is obtained, wherein the surrounding environment constraints include warehouse distribution, no-fly zone position and weather conditions; so as to ensure that the unmanned aerial vehicle can still maintain the energy consumption optimization characteristics of subsequent navigation after the connection is completed.

[0058] The path planning algorithm adopts a multi-objective optimization strategy when calculating the connection path, including the shortest distance, the smallest turning radius, avoiding no-fly zones and comprehensive energy consumption indicators, and solves the flight trajectory that meets the predetermined optimization criteria through weight allocation, wherein the predetermined optimization criteria includes the shortest path length, the lowest energy consumption, flight safety, attitude adjustment smoothness and remaining power utilization efficiency criteria.

[0059] The process of obtaining the connection path is as follows: First, data preparation and model building are carried out, including data import and fusion. First, the current location information of the UAV determined during the signal interruption is imported into the system. At the same time, the pre-planned original energy consumption optimized trajectory data and the surrounding environment information of the UAV obtained through sensors are acquired, including warehouse distribution, no-fly zone location, and weather conditions. Then, these data are fused to build a comprehensive data model that includes location and environmental constraints, reflecting the current status of the UAV and the impact of the surrounding environment on trajectory planning.

[0060] Optimize the trajectory based on the current location and original energy consumption, and extract key features, including image features, trajectory features, and environmental features; acquire image data using an omnidirectional camera or wide-angle camera in the visual sensor, and use feature extraction algorithms such as SIFT and SURF to extract representative feature points in the image to obtain image features, including corner points, edges, or texture-rich areas in the image;

[0061] Based on the original energy consumption optimized trajectory, key navigation points and geometric features of the original energy consumption optimized trajectory are extracted to obtain trajectory features. Among them, by analyzing the UAV energy consumption optimized trajectory, the trajectory is first divided into segments according to flight phases to obtain takeoff segment, cruise segment, and descent segment. Then, through trajectory planning algorithm, key navigation points in the optimized trajectory are identified, including turning points, flight altitude change points, speed change points, acceleration change points, start point, end point, and turning point.

[0062] Extract the geometric features of the track: the curvature, turning angle, and rate of change of heading; use spline interpolation and Bézier curves in curve fitting techniques to smooth the optimized track, remove unnecessary oscillations and abrupt changes, and obtain the geometric features of the original energy-optimized track; this helps to analyze the smoothness and stability of the track, and can also effectively reduce energy waste caused by excessive track curvature.

[0063] The drone's surrounding environment information, such as warehouse distribution, no-fly zone location, and weather conditions, is obtained through sensors. Data processing algorithms are then used to extract features related to flight path planning, thus obtaining environmental characteristics.

[0064] The process of obtaining environmental characteristics is as follows: using the barometer, temperature sensor, wind speed sensor, and GPS onboard the UAV, environmental parameters that may affect energy consumption during flight are collected in real time. These environmental parameters include wind speed and direction, temperature, air pressure, humidity, and terrain undulation. Kalman filtering, mean filtering, and low-pass filtering are then used to denoise the collected environmental data to improve data quality and ensure the accuracy of feature extraction. At the same time, the denoised environmental data is normalized or standardized to make data from different sources comparable, facilitating subsequent analysis and obtaining standard environmental data.

[0065] According to the real-time wind speed and wind direction data in the flight process of the unmanned aerial vehicle, a fluid mechanics model or a physical-based wind field modeling algorithm is used to simulate the wind field distribution of the flight area; a CFD computational fluid dynamics simulation model is used to predict the influence of wind on the flight path, combined with real-time climate data, the influence of climate factors on energy consumption is analyzed through regression analysis or support vector machine, random forest, to obtain the climate conditions, and then using the digital elevation model DEM, the flight path is analyzed, the terrain undulation features are extracted, and through the terrain analysis algorithm, the terrain obstacles that may affect the flight path and the trend of flight height change are identified; thereby the geographical environment features are extracted; wherein the climate data includes temperature, humidity and air pressure; the terrain undulation features include mountains and valleys, and the terrain obstacles include mountains and buildings;

[0066] The PCA method is used to reduce the dimensionality of the standard environment data and extract the most representative environment features; K-means is used to cluster the flight path data under different environmental conditions to identify typical flight patterns under specific weather or geographical conditions; ARIMA is used to predict the trend of environmental changes for the wind speed and temperature changes with time series properties in the standard environment data, thereby providing a reference for the dynamic changes of the environment for the flight path planning in advance;

[0067] The random forest regression is used to establish the mapping relationship between the environment features and the energy consumption combined with the environment features, and through the training of the regression model, the energy consumption changes under different environmental conditions can be predicted, thereby providing a basis for the flight path optimization; the influence degree of different environmental features, such as wind speed, temperature and air pressure, on the energy consumption is analyzed by variance analysis, and the key environmental factors affecting the energy consumption of the flight path are found out and obtained;

[0068] The Kalman filter is used to comprehensively process the environment information combined with the data of multiple sensors, including GPS, IMU, anemometer and barometer, to obtain accurate environment features; and combined with external environment information, including topographic map and weather data obtained by remote sensing satellites or unmanned aerial vehicles, image processing and pattern recognition techniques are used to extract the environment features of the flight area, including wind field distribution and weather conditions, to provide external environment reference for the flight path planning; and the relationship between the wind speed, wind direction, temperature and air pressure in the environment features and the flight path changes is analyzed to obtain the interaction between the environmental conditions and the flight path, and the flight path patterns under different environmental conditions are extracted; the flight path patterns include that the flight path needs to be adjusted to reduce the influence of wind on the flight in strong wind environment; the interaction between the environment features and the flight path planning is modeled to optimize the energy consumption of the flight path;

[0069] The environment features include the boundary, height and internal obstacle distribution of the warehouse, and the location and range of the no-fly zone;

[0070] Among them, for the extraction of key navigation points, by analyzing the track data, the key navigation points in the track are identified: the starting point, the ending point, the turning point or the marker point at a specific position of the track; the coordinates, height and speed information of these points are extracted, which can provide basic data for subsequent path planning;

[0071] For the extraction of geometric shape features, line segment fitting and curve fitting in geometric algorithms are used to fit and analyze the track, and the geometric shape features of the track are extracted; the geometric shape features include the length of the straight line segment, the curvature of the curve segment and the turning angle; the geometric shape features can reflect the smoothness, complexity and energy consumption characteristics of the track.

[0072] And the azimuth and distance features of the current position relative to the original energy consumption optimized track;

[0073] Constructing graph structure and determining algorithm type: constructing the graph structure representation of the original energy consumption optimized track, taking the navigation points in the track as nodes and the connection relationship between the navigation points as edges, and setting the edge weight according to the actual situation; at the same time, the current position of the unmanned aerial vehicle is represented as a special node related to the graph structure; then, the deep reinforcement learning algorithm is selected as the machine learning algorithm type; then the model structure is constructed, and the input layer, output layer and hidden layer parameters are determined; the input layer includes the current position features, the original energy consumption optimized track features and the environment features; the output layer includes the action sequence or path node sequence of the connecting path;

[0074] Among them, the deep reinforcement learning algorithm consists of an agent, an environment, a state, an action, a reward function, a deep neural network and a training algorithm Q-Learning; and combined with the energy consumption optimized track planning requirements of the unmanned aerial vehicle, a complete technical architecture is formed;

[0075] The agent refers to the flight control system responsible for selecting actions according to the current environment state; the task of the agent is to select appropriate flight paths or strategies according to the environmental information collected during flight: climate, wind speed, terrain, to maximize energy efficiency and flight performance;

[0076] The environment is the object that the agent interacts with, which includes all external factors and conditions; the flight environment of the unmanned aerial vehicle includes but is not limited to: meteorological data: wind speed, air pressure, temperature, humidity and other climate factors; flight state data: speed, height, heading, attitude, acceleration and other dynamic states of the unmanned aerial vehicle; terrain information: terrain undulation, obstacles, flight area, etc.; task requirements: waypoints, flight targets, predetermined paths, etc.; the environment provides the agent with information about the current state, and returns feedback (reward or punishment) according to the action of the agent;

[0077] The state is a description of the environment at a certain moment, including: the flight state of the UAV: such as current position (coordinates), speed, flight height, heading, acceleration, etc.; environmental characteristics: such as wind speed, wind direction, temperature, air pressure, etc.; task-related information: such as target waypoint position, task progress, etc.; The representation of state is a vector or tensor, which is used as the input of neural network for the agent to make decisions;

[0078] The action is the choice made by the agent in a given state; in UAV path planning, the action space may include:

[0079] Flight path adjustment: change flight direction, speed, heading;

[0080] Flight altitude change: adjust flight altitude to avoid obstacles or optimize energy consumption;

[0081] Path correction: fine-tune the flight path according to real-time environmental changes (such as wind speed changes); the action space can be discrete (for example, select a specific path) or continuous (for example, adjust flight angle and speed);

[0082] The reward function defines the return of the agent in a certain state when taking a certain action; in energy optimization path planning, the reward function is usually closely related to flight energy consumption; including: when the UAV chooses a path that can effectively reduce flight energy consumption, give positive reward, that is, reward low energy consumption; when the path leads to increased energy consumption, give negative reward, that is, punish high energy consumption; when the task is successfully completed (for example, reaching the destination or completing a specific flight task), give additional reward; also includes environmental adaptation reward: for example, flying with wind or adjusting flight path according to climate change to optimize energy efficiency;

[0083] Among them, the reward function is the driving force for the agent to learn strategies, ensuring that the agent always optimizes in the direction of reducing energy consumption and improving task execution efficiency during flight;

[0084] In deep reinforcement learning, deep neural networks (DNNs) are used to approximate Q-value functions, policy functions, or value functions to handle complex state spaces. Deep neural network models include: Deep Q-Networks (DQNs) use neural networks to approximate Q-value functions. In discrete action spaces, DQN algorithms optimize Q-value functions to enable agents to select optimal actions. Policy networks are neural networks that output action probability distributions. In policy gradient methods, policy networks directly output probabilities of selecting each action given a state to guide decision-making. Actor-Critic architectures combine policy networks (Actors) and value networks (Critics). Actors generate action policies, while Critics evaluate the value of current policies to provide feedback. By optimizing both networks simultaneously, Actor-Critic methods can effectively learn optimal policies. Based on the models contained in deep neural networks, training optimizes their parameters to enable agents to make effective decisions in complex state spaces, optimizing flight paths and energy consumption.

[0085] The training process of deep reinforcement learning involves agent-environment interaction, experience collection, and neural network parameter updates through backpropagation. Training algorithms include Q-Learning, Policy Gradient, A3C, and DDPG. Q-Learning updates Q-value functions to find optimal policies. DQN is a deep learning extension of Q-learning. Policy Gradient directly optimizes policy functions to enable agents to select optimal actions from current states. A3C uses multiple parallel agents to accelerate the training process, optimizing policy and value functions. DDPG is a deep reinforcement learning algorithm for continuous action spaces, enabling optimization in high-dimensional state spaces and continuous action spaces.

[0086] Deep reinforcement learning algorithms in UAV energy optimization path planning cover components from agent interaction, environment modeling, reward design, to deep neural network training and optimization. Through continuous agent-environment interaction, reward mechanisms, and effective application of deep neural networks, deep reinforcement learning can adaptively optimize flight paths in complex dynamic environments, reduce energy consumption, and improve flight efficiency.

[0087] Flight data is collected through actual flight or simulation flight of the UAV, including the position, speed, acceleration, attitude, and energy consumption of the UAV, as well as environmental information during the flight, such as wind direction, wind speed, and temperature; then, a flight performance model of the UAV is established according to the mechanical structure and power system characteristics of the UAV, which is used to simulate the energy consumption of the UAV under different flight conditions; then, the collected flight data is input into the machine learning model for training, and the model is set to output the optimal connection path according to the current position information, the original energy consumption optimized flight path, and the surrounding environmental information;

[0088] During the training process, different weight distributions are set to optimize different targets: the balance of path length, energy consumption, and flight safety; at the same time, the energy consumption of the candidate path is estimated by using the flight performance model, and the energy consumption is used as an optimization target for model training;

[0089] When the length of the candidate path generated by the model is short and the energy consumption is low, a higher reward value is given, and vice versa; by continuously adjusting the model parameters, the model can learn the optimal connection strategy, that is, it can accurately generate the connection path that meets the requirements under different conditions;

[0090] Generating candidate path sequences or action sequences: input the current position information, original energy consumption optimized flight path information, and environmental information into the trained machine learning model, and the model will generate candidate path sequences or action sequences that connect from the current offset position to the original energy consumption optimized flight path based on the learned strategy; these candidate paths are generated based on the model's comprehensive consideration of various factors, and have different lengths, energy consumptions, and flight safety;

[0091] Each candidate path is comprehensively evaluated in terms of shortest distance, highest energy efficiency, avoidance of no-fly zones, and adaptation to weather conditions; for the shortest distance, the actual length of the candidate path is directly calculated; for the highest energy efficiency, the energy consumption is estimated according to the flight performance model of the UAV and the characteristics of the path; for avoiding no-fly zones, it is checked whether the path crosses the no-fly zone; for adapting to weather conditions, the influence of wind direction and wind speed on flight energy consumption and stability is considered, including the increase in energy consumption during headwind flight and the influence of excessive wind speed on flight attitude;

[0092] During the estimation of energy consumption based on the flight performance model of the UAV and the characteristics of the path, the flight performance model of the UAV includes thrust, lift, drag, and aerodynamic characteristics; through the physical characteristics and flight modes of the UAV, a dynamic model of the UAV is established, which reflects the energy consumption changes of the UAV under different flight states; specifically, the flight performance model includes the following key factors:

[0093] Thrust and speed relationship: the thrust requirement of the UAV is closely related to the flight speed, air density, and aerodynamic drag; the size of the thrust determines the energy consumption of the UAV, which in turn affects the energy efficiency;

[0094] Lift and gravity balance: the lift and gravity balance of the UAV affects the stability and energy efficiency of flight; the relationship between lift and drag at different flight altitudes of the UAV, as well as the change of lift coefficient during flight, will affect energy consumption;

[0095] Aerodynamic drag and speed: the relationship between the drag of the UAV and the flight speed can be modeled by classical aerodynamic theory; in general, the drag increases with the increase of flight speed;

[0096] Flight path and flight stage analysis: the flight path is divided into take-off stage, cruising stage, and landing stage, and the energy consumption of each stage is calculated; among them, the change of flight altitude, the curvature of flight route, and the change of flight speed of the flight path will also affect the energy consumption of the UAV;

[0097] In the take-off stage, the UAV needs to overcome gravity and accelerate to flight speed, so the thrust requirement is large and the energy consumption is also high; the energy consumption of the take-off stage is estimated by calculating the relationship between thrust and take-off speed; in the cruising stage, the UAV usually maintains a relatively constant speed and flight altitude, and the energy consumption is relatively stable; according to the aerodynamic efficiency and flight speed of the UAV, the performance model of the UAV is used to calculate the energy consumption of the cruising stage; in the landing stage, the energy consumption of the landing stage is estimated by analyzing the changes of thrust and speed of the UAV during landing;

[0098] Influence of environmental factors: meteorological conditions and terrain have important influence on energy consumption; including air flow, wind speed, air temperature, air pressure will affect the aerodynamic efficiency of the UAV, and then affect the energy consumption; therefore, when estimating energy consumption, real-time meteorological data of wind speed, wind direction, air pressure, and terrain data of height, slope are added for correction; by combining digital elevation model DEM for terrain compensation, the optimization scheme of flight path is calculated to reduce the additional energy consumption caused by terrain changes; based on the flight performance model and environmental factors, dynamic programming DP and particle swarm optimization PSO are used to optimize the flight path, and the flight path with the minimum energy consumption is selected; this algorithm can dynamically adjust the flight path, flight altitude and speed according to the real-time flight state and environmental changes, and reduce energy loss; MATLAB is used to verify and optimize the energy consumption model; through simulation of energy consumption under different flight conditions, the energy efficiency performance under different paths and flight states is evaluated, so as to verify the accuracy and practicability of the energy consumption estimation method;

[0099] The energy consumption prediction method of this invention can accurately predict the energy consumption of UAVs during the flight mission planning stage, optimize the flight path based on flight performance models and environmental factors, and maximize flight energy efficiency. This method not only improves the execution efficiency of flight missions, but also extends the flight time of UAVs and reduces operating costs.

[0100] To avoid no-fly zones, the process of checking whether a route crosses a no-fly zone is as follows: No-fly zone data is obtained via API interfaces from information released by air traffic control agencies and the Civil Aviation Administration of China. This data includes geographical information about the no-fly zones. The no-fly zone data includes polygonal regions, circular regions, and composite regions. Polygonal regions represent no-fly zones with complex boundaries, such as urban airspace or no-fly zones in specific areas. Circular regions are commonly used to represent no-fly zones around airports, represented by a center and radius. Composite regions are no-fly zones composed of various geometric shapes, processed by combining different region shapes. The ray casting method is used to determine whether the route crosses a polygonal region within the no-fly zone. When the no-fly zone is circular, the intersection of the route with the circular region is calculated to determine if the route enters the no-fly zone. For no-fly zones composed of multiple geometric shapes, Boolean operations such as intersection and union are used to determine if the route crosses the no-fly zone. Finally, a comprehensive check is performed to determine whether the route crosses the no-fly zone, yielding the check result.

[0101] Based on the pre-set target weights of path length, energy consumption, and flight safety, a weighted summation method is used to comprehensively evaluate candidate paths;

[0102] If the path length weight is 0.4, the energy consumption weight is 0.3, and the flight safety weight is 0.3, for candidate path X, calculate its weighted score (path length score × 0.4 + energy consumption score × 0.3 + flight safety score × 0.3); by comparing the weighted scores of each candidate path, select the path with the highest score as the final connecting path; this path, while meeting the requirements of shortest distance, highest energy efficiency, avoiding no-fly zones, and adapting to weather conditions, achieves a balanced optimization among the objectives, ensuring that the UAV can smoothly transition back to the original energy-optimized flight path and continue to fly energy-savingly.

[0103] The implementation process for constructing the comprehensive data model is as follows:

[0104] Obtaining current location information: During signal interruption, the current location information of the drone is obtained by combining visual SLAM technology with an IMU inertial measurement unit; among which, visual SLAM technology uses visual sensors: omnidirectional cameras to collect image data of the surrounding warehouse and environment, and construct a three-dimensional local map around the drone;

[0105] Meanwhile, the IMU inertial measurement unit continuously records the motion state parameters of the UAV, including acceleration, angular velocity and heading information; by establishing a data synchronization mechanism between the visual sensor and the IMU inertial measurement unit, the collected data is subjected to image filtering, image enhancement, motion state parameter filtering processing, and then the feature extraction algorithm, optical flow method, feature descriptor matching technology, triangulation method and inertial navigation algorithm are used to calculate the motion trajectory of the feature points, the displacement of the UAV in the three-dimensional space and the attitude change, and finally the accurate current position information is obtained.

[0106] Obtain original energy consumption optimized track data: obtain original energy consumption optimized track data from pre-stored track planning data, which contains flight path information planned by the UAV under normal flight conditions according to the energy consumption optimization principle, including: coordinates of each navigation point in the track, flight direction, speed setting;

[0107] Collect surrounding environment information: collect warehouse distribution information through pre-stored warehouse layout data, obtain no-fly zone position information from geographic information system or airspace management regulations, and obtain weather condition information including wind direction and wind speed data through weather sensor or connection with weather data service platform;

[0108] Data fusion and model construction:

[0109] Data fusion processing: fuse the obtained current position information, original energy consumption optimized track data and surrounding environment information; in the fusion process, according to the characteristics and mutual relationship of the data, determine the appropriate fusion method; including: for position information and track data, coordinate transformation and spatial alignment are carried out to ensure that they are processed in the same reference system; for environmental information, associate it with position and track information, and analyze the influence of environmental factors on the flight path;

[0110] Construct comprehensive data model: based on the fused data, construct a comprehensive data model containing position and environmental constraints; adopt a graph structure to take the navigation points in the original energy consumption optimized track as nodes, the connection relationship between nodes as edges, and set the edge weight according to the track characteristic distance and energy consumption; at the same time, represent the current position as a special node related to the graph structure, and integrate the surrounding environment information: warehouse, no-fly zone, weather condition; as the attributes of nodes or edges into the model; the comprehensive data model constructed in this way can comprehensively reflect the current state of the UAV, the target track and the constraint conditions of the surrounding environment, providing a complete data basis for the subsequent path planning algorithm, so that the algorithm can solve the optimal connection path under the consideration of multiple factors.

[0111] The process of obtaining the improved path is as follows:

[0112] According to the unmanned aerial vehicle flight performance model, the residual power of the unmanned aerial vehicle is determined by comparing the estimated energy consumption of each path segment of the connection path; the path segments that cannot be supported by the current residual power are filtered out from the connection path, and a path subset that can be supported by the current residual power is obtained;

[0113] The distance from the current position to the original track target point and the distance from the end point of each path in the residual power feasible path subset to the target point are calculated, and the path is adjusted according to the relationship between the two distances; the straight line distance from the current position to the original track target point and the straight line distance from the end point of each path in the residual power feasible path subset to the target point are calculated by using a geometric algorithm, and distance information is obtained. These distance information can reflect the relative position relationship between the unmanned aerial vehicle and the target point and the advantages and disadvantages of different paths; according to the calculated distance information, a path that meets the residual power requirement and is as close to the target point as possible is selected as the final path. If there are multiple paths that meet the conditions, further comprehensive evaluation and selection are made considering the path length, energy consumption, and flight safety; after the final path is selected, if the path is within the range of the residual power but will make the unmanned aerial vehicle deviate from the target point too far, the path is adjusted by inserting a new navigation point in the middle of the path or changing the direction of the path. The adjusted path should be closer to the direction of the original track target point under the premise of sufficient residual power; avoid deviating from the target point too far due to path selection, which affects the overall energy consumption and flight efficiency; the distance from the current position to the original track target point is D1, and the distance from the end point of each path in the residual power feasible path subset to the target point is D2; the current position information of the unmanned aerial vehicle (obtained by fusing visual SLAM technology and IMU data) and the position information of the original track target point are used to calculate D1; in the residual power feasible path subset, the distance between the position information of each path end point and the position information of the original track target point is calculated to obtain the D2 set; if D1 is less than all D2, that is, the distance from the current position to the original track target point is shorter than the distance through any residual power feasible path subset, then the unmanned aerial vehicle should return to the original track along the shortest path as the direct return strategy; if D1 is greater than some D2, that is, there is a shorter path through the residual power feasible path subset to return to the original track target point, then the path is analyzed, and the path with the smallest D2 and meeting the energy consumption optimization requirement is selected as the adjusted path as the path optimization strategy; according to the above analysis, the adjusted path is determined, and adjustment instructions including height fine-tuning, heading change, and speed control are sent to the flight control system of the unmanned aerial vehicle through the control unit to complete path planning and implementation;

[0114] It needs to be further explained that in the specific implementation process, if the path is in the remaining power range, but will make the UAV deviate from the target point too far, resulting in a large increase in subsequent flight distance, which may affect the overall energy consumption and flight efficiency, the path is corrected: by inserting a new navigation point in the middle of the path or changing the path direction, it is closer to the target point direction of the original flight path under the premise of ensuring sufficient remaining power;

[0115] Perimeter environment constraint processing: warehouse distribution processing: with the help of image data or pre-stored warehouse distribution data, analyze the relationship between the connection path and the warehouse space, adjust the flight height and change the direction for the path segment close to the warehouse that affects safety or has high energy consumption for flying around, avoid complex warehouse areas, and get the path after optimizing the warehouse distribution; It needs to be further explained that in the specific implementation process, for the path segment close to the warehouse and affecting flight safety (such as the flight height being lower than the warehouse height and being too close) or causing too high energy consumption (such as frequent turns) due to flying around the warehouse, by adjusting the flight height and changing the path direction to avoid complex warehouse areas, the path after optimizing the warehouse distribution is obtained;

[0116] No-fly zone position processing: according to the no-fly zone position information, check whether the connection path crosses the no-fly zone; if there is a crossing situation, use the path planning algorithm to re-plan the path segment outside the no-fly zone boundary, replace the original path segment crossing the no-fly zone with the path segment bypassing the no-fly zone, and ensure that the path is legal and the energy consumption increment of bypassing is minimized;

[0117] Weather condition processing: obtain weather condition data, including wind direction and wind speed information, for the path segment flying against the wind, calculate the energy consumption increase of flying against the wind according to the aerodynamic performance model of the UAV (such as the resistance coefficient change curve under different wind direction angles); if the energy consumption increase is too large, adjust the path direction to get a detour path segment, it needs to be further explained that in the specific implementation process, find a more favorable wind direction area to fly, use the detour strategy, fly along the tangent direction of the wind for a distance and then turn to the target point direction to reduce the overall energy consumption; At the same time, considering the influence of wind speed on the flight stability of the UAV, for the path segment with too large wind speed that may cause the UAV attitude difficult to control, appropriately reduce the flight speed or adjust the flight height to ensure flight safety, and correspondingly optimize the path length and flight time to maintain energy consumption optimization;

[0118] Based on the path after the remaining power evaluation, target point distance analysis and surrounding environment constraint processing; again, the weighted summation method is used to optimize the path, and the improved path that meets the current remaining power of the unmanned aerial vehicle, the distance of the target point and the constraint conditions of the surrounding environment is determined. The path is the final result obtained in the process of connecting the path calculation, which is used for subsequent flight control and navigation operation; when the improved path is determined, the control unit sends adjustment instructions to the flight control system of the unmanned aerial vehicle, including height fine tuning, heading change, speed control, so that the unmanned aerial vehicle can smoothly complete the route correction without significantly increasing energy consumption.

[0119] It needs to be further explained that in the specific implementation process, by arranging a multi-sensor fusion system on the unmanned aerial vehicle, the relative displacement and attitude change of the unmanned aerial vehicle can be accurately measured and calculated during signal interruption by using visual SLAM technology combined with IMU data, so as to determine the current position information; this enables the unmanned aerial vehicle to accurately know its own position in a complex warehouse environment even if the signal is lost, laying a foundation for subsequent track recovery;

[0120] When the signal is restored, the track optimization algorithm based on machine learning combines the current position information, the original energy consumption optimized track and the surrounding environment information, and can calculate the shortest and most energy-efficient connection path that connects the original track from the current offset position in real time; and in the process of connection path calculation, the improved path can also be further optimized according to the current remaining power of the unmanned aerial vehicle and the distance of the target point, to ensure that the unmanned aerial vehicle can smoothly transition back to the original energy consumption optimized track and continue to fly in an energy-saving manner; the continuity and reliability of the unmanned aerial vehicle in executing tasks in a complex environment are improved, and the risk of task failure or significant increase in energy consumption caused by signal interruption is reduced;

[0121] In the track planning process, a multi-objective optimization strategy is adopted, and factors such as shortest distance, minimum turning radius, avoidance of no-fly zones and comprehensive energy consumption indicators are considered, and the flight trajectory that meets the predetermined optimization criteria is solved through weight distribution; this enables the unmanned aerial vehicle to select a better path during flight, reducing unnecessary flight distance and maneuvering actions, thereby reducing energy consumption; for example, by reasonably planning the path to avoid windward areas or reduce the number of turns, the energy consumption increase caused by air resistance and attitude adjustment is effectively reduced;

[0122] Whether during signal interruption or during normal flight, this method can fully consider the impact of surrounding environmental constraints on energy consumption; when facing changes in weather conditions, the path direction can be adjusted in time to find a more favorable wind direction area for flight, maintaining energy consumption optimization; at the same time, the path is dynamically adjusted according to the current remaining power of the unmanned aerial vehicle, ensuring that the optimal energy consumption utilization can be achieved under different power states, improving the endurance of the unmanned aerial vehicle and enabling it to complete more tasks with limited power;

[0123] Through the technical scheme, the unmanned aerial vehicle can fly according to the energy consumption optimization flight path planning method in the environment where the shelf structures are densely arranged and multiple no-fly zones exist, the problem that the signal interruption caused by the warehouse shielding may make the unmanned aerial vehicle deviate from the preset energy consumption optimization flight path is solved, and it is ensured that the unmanned aerial vehicle can quickly return to the optimal flight path and continue to fly in an energy-saving manner after the signal is recovered.

[0124] Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art and related fields without creative labor shall belong to the protection scope of the present application. The structures, devices and operation methods not specifically described and explained in the present application are implemented according to the conventional means in the art, unless otherwise specified and limited.

Claims

1. A method for optimizing energy consumption of an unmanned aerial vehicle (UAV), characterized in that, The method comprises the following steps: Step 1: A multi-sensor fusion system is arranged on the UAV, and the multi-sensor fusion system comprises a visual sensor, an IMU inertial measurement unit and a GPS backup module; Step 2: The UAV flies along a preset energy consumption optimized flight path in the warehouse, and when the signal is interrupted, the visual sensor acquires image data of the surrounding warehouse and environment, and the IMU inertial measurement unit continuously records the motion state parameters of the UAV; Step 3: A three-dimensional local map around the UAV is constructed according to the image data by using a visual SLAM technology, and the relative displacement and attitude change of the UAV during the signal interruption period are calculated in combination with the motion state parameters to obtain current position information; Step 4: After the signal is restored, the current position information is compared with the original energy consumption optimized flight path, a shortest connection path connecting the current offset position to the original flight path is solved by a path planning algorithm, and a connection path is obtained; Step 5: The strategy instruction is optimized according to the improved path to obtain an optimized strategy instruction, and the UAV is executed by the optimized strategy instruction to make a smooth transition back to the original energy consumption optimized flight path and continue to fly; In the process of obtaining the current position information, first, a feature extraction algorithm is used to extract representative feature points from the image data, the feature points in the continuous frame images are tracked and matched by using an optical flow method and a feature descriptor matching technology, the corresponding relationship between the feature points in different image frames is determined, and the motion trajectory of the feature points is acquired; then, the position change of the feature points in the three-dimensional space is calculated by using a triangulation method according to the feature point motion trajectory and a camera imaging model, the displacement of the UAV in the horizontal, vertical and depth directions is deduced, and a displacement estimation result of the UAV is obtained; Subsequently, the attitude change of the UAV is calculated by using an inertial navigation algorithm according to the acceleration and angular velocity data recorded by the IMU inertial measurement unit; the angular displacement of the UAV is calculated by using a gyroscope data, the change data of the heading, pitch and roll attitude angles are determined, and an attitude estimation result is obtained; In the process of obtaining the connection path, first, the current position information determined during the signal interruption period of the UAV is imported, the original energy consumption optimized flight path data is acquired, the surrounding environment information of the UAV is fused, and a comprehensive data model containing position and environment constraints is constructed; key features are extracted from the current position and the original energy consumption optimized flight path, the key features including key navigation points, geometric shape features of the original energy consumption optimized flight path and the direction and distance features of the current position relative to the original energy consumption optimized flight path; wherein, by analyzing the energy consumption optimized flight path of the UAV, the flight path is segmented according to flight stages to obtain a take-off section, a cruising section and a descending section; then, key navigation points in the optimized flight path are identified by using a flight path planning algorithm, including turning points, flight height change points, speed change points, acceleration mutation points, starting points, ending points and turning points. The geometric features of the flight path are extracted, including the curvature, turning angle, and heading rate of change of the flight path; based on the geometric features of the flight path, the optimized flight path is smoothed using spline interpolation and Bezier curve in the curve fitting technique to remove unnecessary oscillation and sharp changes, and the geometric shape features of the original energy consumption optimized flight path are obtained. 2.The method of claim 1, wherein: The visual sensor in the multi-sensor fusion system is an omnidirectional camera or a camera with wide-angle characteristics, which is used to visually cover the surrounding environment during signal interruption. 3.The method of claim 2, wherein: The motion state parameters include acceleration, angular velocity, and heading information, and are continuously accumulated during signal interruption to assist in estimating the relative position change of the UAV when the visual SLAM technology cannot provide current position information. 4.The method of claim 3, wherein: The three-dimensional local map is a sparse point cloud map or a key feature point map constructed by the visual SLAM technology during signal interruption.

5. The method of claim 4, wherein: The displacement estimation result based on the visual SLAM technology and the attitude estimation result based on the IMU inertial measurement unit are fused by using extended Kalman filter (EKF) or unscented Kalman filter (UKF) to obtain fusion information; the previously estimated relative displacement and attitude change are optimized and adjusted using the fusion information, and are iteratively updated to finally obtain accurate relative displacement and attitude change data, which are used to determine the current position information of the UAV during signal interruption. 6.The method of claim 5, wherein: In step four, a machine learning-based flight path optimization algorithm is used to compare the current position information with the original energy consumption optimized flight path, and to calculate the shortest and most energy-efficient flight path that connects the original flight path from the current offset position in real time, thereby obtaining the connection path.

7. The method of claim 6, wherein the method further comprises: In step four, the connection path is also optimized according to the current remaining power of the UAV, the distance to the target point, and the surrounding environment constraints during the connection path calculation process, thereby obtaining an improved path, wherein the surrounding environment constraints include warehouse distribution, no-fly zone location, and weather conditions. 8.The method of claim 7, wherein: The path planning algorithm uses a multi-objective optimization strategy when calculating the connection path, including shortest distance, minimum turning radius, avoidance of no-fly zones, and comprehensive energy consumption indicators, and solves the flight trajectory that meets the predetermined optimization criteria through weight allocation, wherein the predetermined optimization criteria include shortest path length, lowest energy consumption, flight safety, attitude adjustment smoothness, and remaining power utilization efficiency criteria. 9.The method of claim 8, wherein: The original energy consumption optimized flight path is constructed by graph structure, with navigation points as nodes, connection relationships as edges, and edge weights set, and the current position is represented as a special node related to the graph structure; a deep reinforcement learning algorithm is selected as the machine learning algorithm type, a model structure is constructed, input layer, output layer, and hidden layer parameters are determined, simulation or historical flight data is collected, the machine learning model is trained according to the set reward function, and the model parameters are optimized to learn the optimal connection strategy; wherein the reward function takes short path length and low energy consumption as high reward basis. The current position information, the original energy consumption optimized flight path information and the environmental information are input into the trained machine learning model to generate a candidate path sequence or action sequence that connects from the current offset position to the original energy consumption optimized flight path. A weighted summation method is used to evaluate and screen the candidate paths according to the target weights of path length, energy consumption and flight safety, and to determine the final connection path. 10.The method of claim 9, wherein: The process of obtaining the improved path is as follows: Residual power evaluation: according to the unmanned aerial vehicle flight performance model, the estimated energy consumption of each path segment of the connection path is compared to determine the current residual power of the unmanned aerial vehicle; the path segments that are difficult to support by the current residual power are screened out and removed from the connection path to obtain a path subset that is feasible for the residual power; Target point distance analysis: the distance from the current position to the target point of the original flight path and the distance from the end point of each path in the residual power feasible path subset to the target point are calculated, and the path is adjusted according to the relationship between the two distances; Surrounding environment constraint processing: warehouse distribution processing: with the help of image data or pre-stored warehouse distribution data, the relationship between the connection path and the warehouse space is analyzed, and for the path segments that are close to the warehouse and affect safety or have high energy consumption for circumnavigation, the flight height is adjusted and the direction is changed to avoid complex warehouse areas to obtain a path optimized for warehouse distribution; No-fly zone position processing: according to the no-fly zone position information, it is checked whether the connection path crosses the no-fly zone; if there is a crossing, a path planning algorithm is used to re-plan the path segment outside the no-fly zone boundary, and the original path segment crossing the no-fly zone is replaced by a path segment bypassing the no-fly zone; Weather condition processing: obtain weather condition data including wind direction and wind speed information, for the path segment flying against the wind, according to the aerodynamic performance model of the unmanned aerial vehicle, the energy consumption increase of flying against the wind is calculated; if the energy consumption increase is too large, the path direction is adjusted to obtain a detour path segment; Based on the path after residual power evaluation, target point distance analysis and surrounding environment constraint processing, the weighted summation method is used again to optimize the path to determine the improved path that meets the current residual power of the unmanned aerial vehicle, the target point distance and the surrounding environment constraint conditions; when the improved path is determined, the control unit sends adjustment instructions to the flight control system of the unmanned aerial vehicle, including height fine-tuning, heading change and speed control, so that the unmanned aerial vehicle can smoothly complete the route correction without significantly increasing the energy consumption.

Images