Path planning method for rescue drone based on support vector machine and genetic algorithm

Abstract

The present invention provides a rescue drone path planning method based on support vector machine and genetic algorithm. The present invention combines support vector machine and genetic algorithm, uses multi-classification support vector machine to quantitatively evaluate and classify the collision risk of drone flight path, and uses genetic algorithm to iteratively optimize support vector machine model, so as to be able to perform collision analysis and evaluation on the simulated rescue path, and the result is scientific and reasonable, which effectively improves the safety and stability of drone rescue, and at the same time improves the accuracy and quality of path planning; in addition, the present invention improves the traditional genetic algorithm, uses the improved genetic algorithm to plan the path, and at the same time combines the trained support vector machine model to identify the risk of the planned path, so as to further improve the path quality, realize high-precision and high-efficiency path planning, and has more advantages than the traditional rescue route constructed by single weight and model.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle control, and more specifically, to a rescue unmanned aerial vehicle path planning method based on support vector machine and genetic algorithm. Background Art

[0002] my country is one of the countries with the most serious natural disasters in the world. Disasters are of many types, distributed over a wide area, and occur frequently, causing great damage to infrastructure, economic development, and the safety of people's lives and property. In recent years, affected by global climate change and the increase in extreme weather in my country, geological disasters such as floods, typhoons, wind and hail, and landslides have occurred frequently, and the economic losses caused by natural disasters have been increasing year by year.

[0003] In recent years, drones have played an increasingly important role in disaster investigation, on-site assessment, monitoring and tracking, precise positioning, auxiliary rescue, auxiliary supervision, and delivery of relief supplies. After extreme disasters, due to the destruction or closure of ground transportation, traditional human resources are difficult to quickly reach the disaster area for rescue and disaster assessment. Drones are an ideal choice due to their portability and autonomous flight capabilities. By using drones for path planning, the rescue needs and priorities of the disaster area can be quickly and accurately determined; at the same time, drones can also overcome the limitations of terrain and environment and easily fly over damaged buildings, obstacles and complex terrain. Drone clusters can cover large areas in a short period of time, greatly improving the efficiency of search and rescue; in addition, drones can also quickly determine closed roads and rescue difficulties, and provide real-time image and video transmission, enabling command centers and rescue personnel to make quick decisions.

[0004] Based on the many advantages of drones in post-disaster rescue missions, some existing technologies have applied drones to this specific scenario. For example, an existing patent document discloses a drone emergency collaborative path planning method and system for extreme post-disaster scenarios. The specific steps include: real-time priority sorting among drones according to the flight indicators and preset rules of each drone; based on the disaster level of each disaster point in the extreme post-disaster scenario, drone flight missions are allocated through a drone hybrid scheduling strategy. First, the drone with the highest real-time priority is selected according to the greedy strategy to match the disaster point with the highest disaster level, and then the remaining drones are allocated through the nearest neighbor allocation strategy. disaster point; use the A* algorithm to plan the path of each drone to the disaster point, and estimate the time it takes for the drone to reach each path point on the flight path based on the attributes of each drone and the distance to the disaster point, and determine whether there is a collision risk; however, this solution focuses more on optimizing the priority of the drone, and the planned drone rescue route is only based on the A* algorithm. The planning model and weights are relatively simple, and the planned path lacks a scientific and reasonable means of risk identification and quality assessment. When drones fly in complex and harsh post-disaster scenarios, there are still risks and hidden dangers of operational failures due to collisions with the environment or other drones. The quality of the path planned by this existing technology is still low. Summary of the invention

[0005] In order to overcome the defects of the above-mentioned prior art that the rescue drone path planning model is single and the planned path lacks risk identification means, the present invention provides a rescue drone path planning method based on support vector machine and genetic algorithm, which uses a multi-classification support vector machine to quantitatively evaluate and classify the collision risk of the drone flight path, and combines the improved genetic algorithm to perform high-precision and high-efficiency path planning. The results are scientific and reasonable, and have good application in the risk identification of rescue routes of emergency rescue drone clusters. In addition, the present invention has more advantages than the rescue route constructed by the traditional single weight method.

[0006] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0007] A rescue drone path planning method based on support vector machine and genetic algorithm includes the following steps:

[0008] S1: Collect historical flight data of several drones in the rescue area, construct a historical flight data set and perform preprocessing;

[0009] S2: Establish a collision risk assessment model based on support vector machine;

[0010] The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area;

[0011] S3: inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model;

[0012] S4: using the improved genetic algorithm to plan the rescue paths of all UAVs in the rescue area, obtaining the simulated rescue paths of each UAV, and extracting the simulated flight data corresponding to each UAV;

[0013] S5: Inputting the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtaining the collision risk level of each UAV when flying in the rescue area;

[0014] S6: Determine whether the collision risk level of each UAV is lower than the preset level. If so, use its simulated rescue path as the final rescue path; otherwise, re-execute steps S4-S5 to re-plan the path and perform collision risk assessment.

[0015] Preferably, the method further comprises: before the rescue, establishing a hierarchical structure model based on the hierarchical analysis method, analyzing the safety assurance capability of the rescue area, and obtaining the analysis results;

[0016] If the analysis result meets the preset conditions, steps S1 to S6 are executed; otherwise, the parameters of each indicator in the hierarchical structure model are adjusted to obtain a new analysis result.

[0017] Preferably, the hierarchical model includes a target layer, a criterion layer and an indicator layer; the target layer is to evaluate the safety assurance capability of the rescue area; the criterion layer includes: personnel safety, equipment safety, operating environment safety and organizational management safety; the indicator layer includes: various evaluation indicators corresponding to personnel safety, equipment safety, operating environment safety and organizational management safety;

[0018] The evaluation indicators of personnel security safety include: rescuer workload, rescuer health status and rescuer professional level;

[0019] The evaluation indicators of equipment safety include: rescue drone operation status, communication equipment operation status, flight control system operation status and navigation system operation status;

[0020] The evaluation indicators of the safety of the operating environment include: airspace environment of the rescue area, geographical environment of the rescue area, weather environment of the rescue area and electromagnetic environment of the rescue area;

[0021] The evaluation indicators of organizational management safety include: rescue manuals and rules and regulations, rescue operation procedures, rescue human resources management and rescue safety management;

[0022] Determine the weight of each evaluation indicator by using the hierarchical analysis method, multiply the indicator score collected in advance for each evaluation indicator and its corresponding weight, and obtain the final score of each evaluation indicator;

[0023] The final scores of all evaluation indicators are added together to obtain a comprehensive safety index as an analysis result. When the comprehensive safety index is greater than a preset threshold, steps S1 to S6 are executed; otherwise, the evaluation indicators are adjusted and the comprehensive safety index is recalculated.

[0024] Preferably, in step S1, the historical flight data includes: the airspace volume V of the rescue area, the drone density ρ in the rescue area, the flight time t of each drone in the rescue area, the reliability P of the successful collision avoidance of the man-machine system composed of rescue personnel and equipment, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk.

[0025] Preferably, the preprocessing in step S1 includes: performing data noise reduction and normalization on all collected historical flight data in turn, calculating the collision index CR of each UAV when flying in the rescue area according to the normalized historical flight data, and saving all collision indexes CR together as a preprocessed historical flight data set to complete the preprocessing;

[0026] The data noise reduction includes: data cleaning and outlier removal;

[0027] The various data in each historical flight data are normalized according to the following formula:

[0028]

[0029] in, is the normalized data; is the data before normalization; and They are the maximum and minimum values ​​of the corresponding category data respectively;

[0030] The collision index CR of each drone flying in the rescue area is calculated according to the following formula:

[0031]

[0032]

[0033] Among them, CR1 is the collision risk of drones per unit time when only airspace factors are considered.

[0034] Preferably, the collision risk assessment model in step S2 is specifically: a multi-classification support vector machine model whose kernel function is a Gaussian radial basis kernel function; the parameters of the multi-classification support vector machine model include a penalty factor and a kernel parameter;

[0035] The multi-classification support vector machine model is used to classify the collision risk level of each drone when flying in the rescue area into four levels: low, medium, high and very high according to the collision index CR of each drone when flying in the rescue area.

[0036] Preferably, in step S3, iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model includes:

[0037] S3.1: Population initialization: Encode the parameters of the multi-classification support vector machine model into binary codes, represented by a chromosome; Randomly perform genetic inheritance on the initial population of chromosomes with 2 parameters within a preset range;

[0038] S3.2: Fitness calculation: The preprocessed historical flight data set is used as a training sample to calculate the fitness value of each chromosome, and the optimal parameter value of the initial population is found by searching for the best fitness value;

[0039] S3.3: Termination criteria: determine whether the preset number of iterations has been reached. If so, execute step S3.5; otherwise, execute step S3.4;

[0040] S3.4: Optimization algorithm: select the next generation population parent according to the fitness value and the preset selection operator, generate a new population through crossover and mutation operations, and repeat steps S3.2~S3.3;

[0041] S3.5: Save the optimal parameter value obtained from the last population iteration to obtain the optimal collision risk assessment model.

[0042] Preferably, in step S4, the following improvements are made on the basis of the traditional genetic algorithm to obtain the improved genetic algorithm:

[0043] Increase the penalty factor and motivational factors , thereby optimizing the fitness function, the optimized fitness function Specifically:

[0044]

[0045]

[0046]

[0047] in, is the drone serial number; , and are the first, second and third weight coefficients respectively; is the length of the simulated rescue path of the UAV; D is the Euclidean distance between the UAV and the target point; is the number of obstacles crossed by the drone in the simulated rescue path;

[0048] Improvements to the selection operation: The selection operator uses the roulette selection method, assuming that the population size is , the fitness of an individual is , then the probability of being selected for:

[0049]

[0050] Improvements to the crossover operation: Use the single-point crossover algorithm to select two parent individuals from the current population, randomly select crossover sites for single-point crossover, and generate new offspring individuals;

[0051] For mutation operation: adopt differential evolution algorithm, use the differences between individuals in the population to achieve gene mutation, and randomly select 3 individuals from the current population , and ,right and Perform differential calculation to express, then the differential vector for:

[0052]

[0053] in, is the scaling factor;

[0054] Individual With the difference vector Combine to create new individuals :

[0055]

[0056] At the same time, the mutation probability is introduced to control the mutation. is the number of iterations of the current population, is the maximum number of iterations, then the mutation probability for:

[0057]

[0058] in, is the mutation probability when iterating to the i-th generation; is the initial probability of mutation.

[0059] Preferably, in step S4, the simulated flight data includes: the airspace volume V of the rescue area, the drone density ρ in the rescue area, the flight time t of each drone simulating the rescue path, the reliability P of the man-machine system composed of rescue personnel and equipment for successful collision avoidance, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk.

[0060] The present invention also provides a rescue drone path planning system based on support vector machine and genetic algorithm, which uses the above-mentioned rescue drone path planning method based on support vector machine and genetic algorithm, including:

[0061] Preprocessing unit: used to collect historical flight data of several drones in the rescue area, construct historical flight data sets and perform preprocessing;

[0062] Model building unit: used to build a collision risk assessment model based on support vector machine;

[0063] The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area;

[0064] Model optimization unit: used for inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model;

[0065] Path simulation unit: used to plan the rescue paths of all UAVs in the rescue area using an improved genetic algorithm, obtain the simulated rescue paths of each UAV, and extract the simulated flight data corresponding to each UAV;

[0066] A collision risk assessment unit is used to input the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtain the collision risk level of each UAV when flying in the rescue area;

[0067] Rescue path generation unit: used to determine whether the collision risk level of each drone is lower than the preset level. If so, its simulated rescue path is used as the final rescue path; otherwise, the path simulation unit and collision risk assessment unit are re-executed to re-plan the path and perform collision risk assessment.

[0068] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0069] The present invention provides a rescue drone path planning method based on a support vector machine and a genetic algorithm. First, historical flight data of several drones in a rescue area are collected, and a historical flight data set is constructed and preprocessed; then a collision risk assessment model is established based on a support vector machine; the preprocessed historical flight data set is input into the collision risk assessment model for training, and the parameters of the collision risk assessment model are iteratively optimized using a genetic algorithm to obtain an optimal collision risk assessment model; then, the rescue paths of all drones in the rescue area are planned using an improved genetic algorithm, a simulated rescue path of each drone is obtained, and simulated flight data corresponding to each drone is extracted; the simulated flight data corresponding to each drone is input into the optimal collision risk assessment model for collision risk assessment, and a collision risk level of each drone when flying in the rescue area is obtained; finally, it is determined whether the collision risk level of each drone is lower than a preset level, and if so, its simulated rescue path is used as the final rescue path; otherwise, the path is replanned and a collision risk assessment is performed;

[0070] The present invention combines support vector machines and genetic algorithms, uses multi-classification support vector machines to quantitatively evaluate and classify the collision risk of UAV flight paths, and uses genetic algorithms to iteratively optimize the support vector machine model, which can perform collision analysis and evaluation on the simulated rescue paths. The results are scientific and reasonable, effectively improving the safety and stability of UAV rescue, and at the same time improving the accuracy and quality of path planning.

[0071] In addition, the present invention improves the traditional genetic algorithm, so that the screening efficiency of the fitness function is improved, the algorithm iteration speed is faster, and the new path generated by mutation is of higher quality. The improved genetic algorithm is used to plan the path, and the trained support vector machine model is combined to identify the risks of the planned path, which further improves the path quality and realizes high-precision and high-efficiency path planning, which is more advantageous than the traditional rescue route constructed with a single weight and model. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of a rescue drone path planning method based on support vector machine and genetic algorithm provided in Example 1.

[0073] Figure 2 This is a schematic diagram of the hierarchical model provided in Example 2.

[0074] Figure 3 This is a flow chart for iteratively optimizing the collision risk assessment model using a genetic algorithm provided in Example 2.

[0075] Figure 4 This is a structural diagram of a rescue drone path planning system based on support vector machine and genetic algorithm provided in Example 3. DETAILED DESCRIPTION

[0076] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0077] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0078] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0079] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0080] Example 1

[0081] like Figure 1 As shown, this embodiment provides a rescue drone path planning method based on support vector machine and genetic algorithm, comprising the following steps:

[0082] S1: Collect historical flight data of several drones in the rescue area, construct a historical flight data set and perform preprocessing;

[0083] S2: Establish a collision risk assessment model based on support vector machine;

[0084] The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area;

[0085] S3: inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model;

[0086] S4: using the improved genetic algorithm to plan the rescue paths of all UAVs in the rescue area, obtaining the simulated rescue paths of each UAV, and extracting the simulated flight data corresponding to each UAV;

[0087] S5: Inputting the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtaining the collision risk level of each UAV when flying in the rescue area;

[0088] S6: Determine whether the collision risk level of each UAV is lower than the preset level. If so, use its simulated rescue path as the final rescue path; otherwise, re-execute steps S4-S5 to re-plan the path and perform collision risk assessment.

[0089] In the specific implementation process, we first collect historical flight data of several drones in the rescue area, construct a historical flight data set and perform preprocessing;

[0090] Then, a collision risk assessment model was established based on support vector machine;

[0091] The preprocessed historical flight data set is input into the collision risk assessment model for training, and the parameters of the collision risk assessment model are iteratively optimized using a genetic algorithm to obtain the optimal collision risk assessment model;

[0092] Then, the improved genetic algorithm is used to plan the rescue paths of all UAVs in the rescue area, obtain the simulated rescue paths of each UAV, and extract the simulated flight data corresponding to each UAV;

[0093] The simulated flight data corresponding to each UAV is input into the optimal collision risk assessment model to perform collision risk assessment, and obtain the collision risk level of each UAV when flying in the rescue area;

[0094] Finally, it is determined whether the collision risk level of each drone is lower than the preset level. If so, its simulated rescue path is used as the final rescue path; otherwise, the path is replanned and the collision risk assessment is performed;

[0095] This method combines support vector machines and genetic algorithms, uses multi-classification support vector machines to quantitatively evaluate and classify the collision risk of UAV flight paths, and uses genetic algorithms to iteratively optimize the support vector machine model. It can perform collision analysis and evaluation on the simulated rescue path. The results are scientific and reasonable, effectively improving the safety and stability of UAV rescue, while improving the accuracy and quality of path planning.

[0096] In addition, this method improves the traditional genetic algorithm and uses the improved genetic algorithm to plan the path. At the same time, it combines the trained support vector machine model to identify risks of the planned path, further improving the path quality and achieving high-precision and high-efficiency path planning. It has more advantages than the traditional rescue route constructed by single weight and model.

[0097] Example 2

[0098] This embodiment provides a rescue drone path planning method based on support vector machine and genetic algorithm, comprising the following steps:

[0099] S1: Collect historical flight data of several drones in the rescue area, construct a historical flight data set and perform preprocessing;

[0100] S2: Establish a collision risk assessment model based on support vector machine;

[0101] The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area;

[0102] S3: inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model;

[0103] S4: using the improved genetic algorithm to plan the rescue paths of all UAVs in the rescue area, obtaining the simulated rescue paths of each UAV, and extracting the simulated flight data corresponding to each UAV;

[0104] S5: Inputting the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtaining the collision risk level of each UAV when flying in the rescue area;

[0105] S6: Determine whether the collision risk level of each drone is lower than a preset level. If so, use its simulated rescue path as the final rescue path; otherwise, re-execute steps S4 to S5 to re-plan the path and perform collision risk assessment;

[0106] The method further comprises: before the rescue, establishing a hierarchical structure model based on the hierarchical analysis method, analyzing the safety guarantee capability of the rescue area, and obtaining the analysis result;

[0107] If the analysis result meets the preset conditions, steps S1 to S6 are executed; otherwise, the parameters of each indicator in the hierarchical structure model are adjusted to obtain a new analysis result;

[0108] The hierarchical model includes a target layer, a criterion layer and an indicator layer; the target layer is to evaluate the safety assurance capability of the rescue area; the criterion layer includes: personnel safety, equipment safety, operating environment safety and organizational management safety; the indicator layer includes: various evaluation indicators corresponding to personnel safety, equipment safety, operating environment safety and organizational management safety;

[0109] The evaluation indicators of personnel security safety include: rescuer workload, rescuer health status and rescuer professional level;

[0110] The evaluation indicators of equipment safety include: rescue drone operation status, communication equipment operation status, flight control system operation status and navigation system operation status;

[0111] The evaluation indicators of the safety of the operating environment include: airspace environment of the rescue area, geographical environment of the rescue area, weather environment of the rescue area and electromagnetic environment of the rescue area;

[0112] The evaluation indicators of organizational management safety include: rescue manuals and rules and regulations, rescue operation procedures, rescue human resources management and rescue safety management;

[0113] Determine the weight of each evaluation indicator by using the hierarchical analysis method, multiply the indicator score collected in advance for each evaluation indicator and its corresponding weight, and obtain the final score of each evaluation indicator;

[0114] The final scores of all evaluation indicators are added together to obtain a comprehensive safety index as the analysis result. When the comprehensive safety index is greater than a preset threshold, steps S1 to S6 are executed; otherwise, the evaluation indicators are adjusted and the comprehensive safety index is recalculated.

[0115] In step S1, the historical flight data includes: the airspace volume V of the rescue area, the density of drones in the rescue area ρ, the flight time t of each drone in the rescue area, the reliability P of the successful collision avoidance of the human-machine system composed of rescue personnel and equipment, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk;

[0116] The preprocessing in step S1 includes: performing data noise reduction and normalization on all the collected historical flight data in turn, calculating the collision index CR of each UAV when flying in the rescue area according to the normalized historical flight data, and saving all the collision indexes CR together as a preprocessed historical flight data set to complete the preprocessing;

[0117] The data noise reduction includes: data cleaning and outlier removal;

[0118] The various data in each historical flight data are normalized according to the following formula:

[0119]

[0120] in, is the normalized data; is the data before normalization; and They are the maximum and minimum values ​​of the corresponding category data respectively;

[0121] The collision index CR of each drone flying in the rescue area is calculated according to the following formula:

[0122]

[0123]

[0124] Among them, CR1 is the collision risk of drones per unit time when only airspace factors are considered;

[0125] The collision risk assessment model in step S2 is specifically: a multi-classification support vector machine model whose kernel function is a Gaussian radial basis kernel function; the parameters of the multi-classification support vector machine model include a penalty factor and a kernel parameter;

[0126] The multi-classification support vector machine model is used to classify the collision risk level of each drone when flying in the rescue area into four levels: low, medium, high and very high according to the collision index CR of each drone when flying in the rescue area;

[0127] In step S3, the parameters of the collision risk assessment model are iteratively optimized using a genetic algorithm to obtain an optimal collision risk assessment model, including:

[0128] S3.1: Population initialization: Encode the parameters of the multi-classification support vector machine model into binary codes, represented by a chromosome; Randomly perform genetic inheritance on the initial population of chromosomes with 2 parameters within a preset range;

[0129] S3.2: Fitness calculation: The preprocessed historical flight data set is used as a training sample to calculate the fitness value of each chromosome, and the optimal parameter value of the initial population is found by searching for the best fitness value;

[0130] S3.3: Termination criteria: determine whether the preset number of iterations has been reached. If so, execute step S3.5; otherwise, execute step S3.4;

[0131] S3.4: Optimization algorithm: select the next generation population parent according to the fitness value and the preset selection operator, generate a new population through crossover and mutation operations, and repeat steps S3.2~S3.3;

[0132] S3.5: Save the optimal parameter value obtained from the last population iteration to obtain the optimal collision risk assessment model;

[0133] In step S4, the following improvements are made on the basis of the traditional genetic algorithm to obtain the improved genetic algorithm:

[0134] Increase the penalty factor and motivational factors , thereby optimizing the fitness function, the optimized fitness function Specifically:

[0135]

[0136]

[0137]

[0138] in, is the drone serial number; , and are the first, second and third weight coefficients respectively; is the length of the simulated rescue path of the UAV; D is the Euclidean distance between the UAV and the target point; is the number of obstacles crossed by the drone in the simulated rescue path;

[0139] Improvements to the selection operation: The selection operator uses the roulette selection method, assuming that the population size is , the fitness of an individual is , then the probability of being selected for:

[0140]

[0141] Improvements to the crossover operation: Use the single-point crossover algorithm to select two parent individuals from the current population, randomly select crossover sites for single-point crossover, and generate new offspring individuals;

[0142] For mutation operation: adopt differential evolution algorithm, use the differences between individuals in the population to achieve gene mutation, and randomly select 3 individuals from the current population , and ,right and Perform differential calculation to express, then the differential vector for:

[0143]

[0144] in, is the scaling factor;

[0145] Individual With the difference vector Combine to create new individuals :

[0146]

[0147] At the same time, the mutation probability is introduced to control the mutation. is the number of iterations of the current population, is the maximum number of iterations, then the mutation probability for:

[0148]

[0149] in, is the mutation probability when iterating to the i-th generation; is the initial probability of mutation;

[0150] In step S4, the simulated flight data includes: the airspace volume V of the rescue area, the drone density ρ in the rescue area, the flight time t of each drone simulating the rescue path, the reliability P of the man-machine system composed of rescue personnel and equipment for successful collision avoidance, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk.

[0151] In the specific implementation process, in order to ensure the operation safety of the rescue drone during the rescue process, this embodiment comprehensively considers the impact of the flight characteristics of the rescue drone, human factors, environment and management factors on the rescue risk, and evaluates the safety guarantee capability of the rescue work to be carried out in the rescue airspace through the Analytic Hierarchy Process (AHP) according to the four indicators of "people, machine, environment and management" and the current standards and relevant specifications, so as to provide a simple, effective and fast evaluation for the safe operation and path planning of the rescue drone, and improve the efficiency, effect and capability of the rescue;

[0152] As for the influencing factors, "people" refers to personnel security. Rescue personnel from multiple positions will participate in rescue and air traffic control. Rescuers are the key to the success of rescue work, including rescue personnel's workload, health status and professional level; "machine" refers to the safety of equipment such as drones, which are all kinds of software and hardware in the whole process of rescue and air traffic management, including drones, communication equipment, navigation equipment, simulation equipment and radar equipment, including factors such as equipment availability, maintenance level, training level and equipment advancement; "environment" refers to the safety of the operating environment, which is environmental factors including internal environmental factors and external environmental factors, such as airspace conditions, geographical environment, weather conditions, rescue work environment conditions, etc.; "management" refers to organizational management safety. Management is a means of controlling the qualifications of personnel, equipment and environment and the connection between various factors. The focus is on analyzing the rescue operation procedures, the system of control-related work, and the implementation of safety production work;

[0153] In this embodiment, a hierarchical structure model of the rescue airspace is constructed by hierarchical decomposition of the research objectives, and the mutual relationship and degree of subordination between the various influencing factors are determined, such as Figure 2 A schematic diagram of the hierarchical model shown, which includes a goal layer, a criterion layer, and an indicator layer;

[0154] Combined with the expert opinions, the indicator positive and negative matrix is ​​constructed. The indicators are divided into two categories: quantitative indicators and qualitative indicators. Since qualitative indicators are highly subjective and difficult to quantify, this method will compare and discriminate each indicator pairwise based on historical relevant data and the opinions of the on-site rescue expert group, and use the comparison results to construct the judgment matrix A;

[0155]

[0156] Among them, due to the relationship between the two factors, a ij =1 / a ji , the resulting judgment matrix is ​​a positive reciprocal matrix;

[0157] Then, the weight of each evaluation index is calculated by using the sum method. First, the elements of each column in the judgment matrix A are normalized, and then the judgment matrix is ​​added row by row to obtain the initial weight. The weight is normalized again to obtain the weight of each evaluation index.

[0158] When comparing and judging the importance of indicators, it is often the case that the results of the previous and subsequent judgments are not very consistent. Therefore, it is very necessary to use a consistency test judgment matrix. The test results are directly related to the validity of the weight results. The consistency test steps are as follows:

[0159] Calculate the maximum eigenvalue of the judgment matrix ,use Calculate the consistency index CI, then find the random consistency index RI according to the random consistency index table, and calculate the randomness ratio CR, CR=CI / RI. If CR≥0.1, the constructed judgment matrix does not pass the consistency test and needs to be adjusted; if CR<0.1, it can be considered that the judgment matrix has satisfactory consistency;

[0160] When the judgment matrix passes the consistency test, the pre-collected index scores and their corresponding weights of each evaluation index are multiplied to obtain the final score of each evaluation index, and the final scores of all evaluation indexes are added to obtain the comprehensive safety index C as the analysis result; this embodiment stipulates that when 1≤C<4, the indicators of the rescue area are not ideal, there are major hidden dangers, and the rescue work should be suspended; 4≤C<7, the indicators of the rescue area are average, there are certain hidden dangers, and the rescue work should be carried out cautiously; 7≤C≤10, the indicators of the rescue area are in good condition, and the rescue work should be carried out in an orderly manner;

[0161] As the first insurance for a successful rescue, before the formal rescue, the safety and security capabilities of the rescue area are analyzed first, which can prevent risks before the rescue starts, effectively ensuring the safety and reliability of the rescue;

[0162] After the safety assurance capability is qualified, the formal rescue work begins. First, it is necessary to collect historical flight data of several drones in the rescue area, build a historical flight data set and perform preprocessing;

[0163] In this embodiment, historical flight data is mainly used to train the collision detection model. In this embodiment, the data is collected at multiple path points in the rescue area, and the distance between the path points is 1 km. The historical flight data collected at each path point includes: the airspace volume V of the rescue area, the drone density ρ in the rescue area, the flight time t of each drone in the rescue area, the reliability P of the successful collision avoidance of the human-machine system composed of rescue personnel and equipment, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk;

[0164] 1) The airspace volume V of the rescue area, the drone density ρ in the rescue area, and the flight time t of each drone in the rescue area are objective and quantifiable factors. The smaller the airspace volume V of the rescue area, the larger the drone density ρ, and the longer the flight time t of each drone in the rescue area, the greater the probability of the drone cluster being collided during flight. If only the airspace factor is considered, the collision risk of drones per unit time can be expressed as:

[0165]

[0166] Among them, CR1 is the collision risk of drones per unit time when only airspace factors are considered;

[0167] 2) When multiple drones share an airspace, managers and drone teams should work together to ensure the safe flight of each drone. At the same time, each drone must have normal flight capabilities and adjust the drone status in time after receiving the warning signal to actively avoid collisions. Managers should send dangerous situations to each drone in a timely manner and take measures according to the actual situation when necessary to command the drone to avoid collisions.

[0168] Each drone with normal flight capability and the management personnel are independent systems. They can form a hybrid system together. Assuming the reliability of the human-machine system to successfully avoid collision is P, we have:

[0169]

[0170] Among them, P1 is the reliability of collision avoidance after the manager mediates the conflict; P2 is the reliability of the UAV successfully avoiding the collision after receiving the collision avoidance signal; P3 is the reliability of the UAV's normal flight;

[0171] 3) The meteorological environment has a great impact on flight safety. Drone clusters generally collide at low altitude or medium-low altitude. The clouds at medium-low altitude have a great impact on flight safety. The cloud base of these medium-low altitude clouds is low, and the visibility under the clouds is low. Some clouds often appear all day long, and there will be continuous rain and snow. Flying in the clouds for a long time may cause ice accumulation, which has a serious impact on flight safety. In addition, weather phenomena that affect flight also include low-altitude wind shear. When the flight altitude is very low, the drone may encounter low-altitude wind shear, which is extremely unfavorable to flight. For example, vertical wind shear changes the lift, thereby changing the flight altitude, affecting the altitude, airspeed, pitch attitude, etc. of the drone.

[0172] Therefore, severe weather phenomena bring adverse flight conditions to UAV flight and also affect the flight performance of UAV, resulting in an increased probability of collision risk. Since meteorological conditions change all the time, it is difficult to quantify them specifically, and different weather phenomena have different effects on flight. In order to facilitate calculation, this method introduces coefficient k1, the size of which represents the impact of meteorological conditions on flight. When the UAV flight does not encounter weather that affects flight, coefficient k1=1; when the UAV flight encounters weather that affects flight, coefficient k1>1.

[0173] 4) The quality of management of drone swarms directly determines whether the flight is safe. The management factors related to the safe flight of drones in the airspace can be divided into three parts: drone operation training, flight organization of drone swarms, and monitoring of drone flight activities. Since the impact of management factors on the risk of drone collision is similar to that of the meteorological environment, it is difficult to quantify it specifically. In order to facilitate calculation, this method introduces the coefficient k2, whose size represents the impact of management factors on flight. When management factors can reduce the risk of drone collision, the coefficient k2 of the impact of management factors on the collision risk is determined according to the actual situation, ranging from (0,1]; when there are problems with management, the risk of drone collision increases due to management, and the coefficient k2 is determined according to the actual situation, ranging from (1,+∞);

[0174] Then, all the collected historical flight data are subjected to data denoising and normalization in turn. Data denoising includes: data cleaning and outlier removal, which can effectively remove data noise and irrelevant data, improve data quality and model training accuracy; each type of data in each historical flight data is normalized according to the following formula:

[0175]

[0176] in, is the normalized data; is the data before normalization; and They are the maximum and minimum values ​​of the corresponding category data respectively;

[0177] Using the normalized data, the collision index CR of each drone flying in the rescue area is calculated according to the following formula:

[0178]

[0179] All collision indexes CR are saved together as a preprocessed historical flight data set to complete the preprocessing;

[0180] After completing the collection and data preprocessing of historical flight data (calculating the collision index CR) according to the five influencing factors of "people, machines, environment, management, and airspace", a collision risk assessment model is established based on a support vector machine; in this embodiment, a Gaussian radial basis (RBF) kernel function is selected as the kernel function g of the support vector machine, and the penalty factor is initialized at the same time; the Gaussian radial basis kernel function can map the original features to infinite dimensions, thereby making the indivisible data separable in a high-dimensional space and avoiding high-dimensional operations. By determining a suitable kernel function characterization value, the distribution characteristics of the sample vector in the entire high-dimensional space can be more effectively described; after selecting the radial basis kernel function as the kernel function, the parameter of its penalty factor is selected between [0.1, 10], and training is performed through simulation experiments. As the number of iterations increases, the loss values ​​of training and testing gradually approach 0, and the number of penalty factors suitable for the rescue area is obtained, thereby reducing the rescue risk and improving the accuracy of the UAV flight route;

[0181] Then, the preprocessed historical flight data set is input into the collision risk assessment model for training, and the parameters of the collision risk assessment model are iteratively optimized using a genetic algorithm to obtain the optimal collision risk assessment model;

[0182] As a very mature machine algorithm, genetic algorithm is a global search optimization method that imitates the natural law of survival of the fittest in the natural biological world and seeks the optimal solution through continuous selection and iteration. By using operation methods such as crossover and mutation, coupled with iterative operations, the population is continuously eliminated and developed and evolved, and finally the optimal solution is obtained;

[0183] like Figure 3 As shown in the figure, the basic process of model training is as follows:

[0184] Population initialization: encoding the parameters of the multi-classification support vector machine model into binary codes, represented by a chromosome; randomly inheriting the initial population of chromosomes with two parameters (kernel function parameters and penalty factors) within a preset range;

[0185] Fitness calculation: The preprocessed historical flight data set is used as a training sample to calculate the fitness value of each chromosome, and the optimal parameter value of the initial population is found by searching for the best fitness value;

[0186] Termination criteria: determine whether the preset number of iterations has been reached. If so, save the optimal parameter value obtained from the last population iteration to obtain the optimal collision risk assessment model.

[0187] Otherwise, the optimization algorithm is executed: the next generation population parent is selected according to the fitness value and the preset selection operator, a new population is generated through crossover and mutation operations, the fitness is calculated repeatedly, and it is determined whether the termination criterion is reached;

[0188] Finally, the penalty factor was determined to be 10.5, the parameter of the kernel function g was 0.5, and the optimal collision risk assessment model was output, which can be used for subsequent real-time collision risk assessment of the planned paths. The collision risk of the simulated flight paths of each UAV was divided into four levels: low, medium, high, and extremely high, and statistical analysis was performed on paths of different risk levels.

[0189] In order to further improve the path planning quality of the drone cluster, this embodiment adopts an improved genetic algorithm to schedule multiple drones. The specific improvements are as follows:

[0190] In traditional genetic algorithms, the fitness function mostly only has a penalty factor. In order to improve the algorithm's screening of population quality, this embodiment adds a penalty factor and motivational factors , thereby optimizing the fitness function, the optimized fitness function Specifically:

[0191]

[0192]

[0193]

[0194] in, is the drone serial number; , and are the first, second and third weight coefficients respectively; is the length of the simulated rescue path of the UAV; D is the Euclidean distance between the UAV and the target point; is the number of obstacles crossed by the UAV in the simulated rescue path; the simulation results show that after 20 simulation iterations, the fitness function becomes increasingly stable as the number of iterations increases, solving the optimization problem of the continuous planning space path;

[0195] In addition to the improvement of the fitness function, this method also improves the genetic operations (selection, crossover and mutation). For the improvement of the selection operation: the selection operator adopts the roulette selection method, assuming that the population size is , the fitness of an individual is , then the probability of being selected for:

[0196]

[0197] in, is the total fitness value. Individuals with larger fitness values ​​have a higher probability of being selected and passing their genetic material to the next generation. On the contrary, individuals with lower fitness values ​​have a lower probability of being selected, but their genetic material will not be completely abandoned. This operator simplifies the design of routes and rescue processes, avoids the negative impact caused by inappropriate selection of weighted matrices due to insufficient rescuer experience, and improves flight and rescue efficiency.

[0198] In order to improve the algorithm speed, this method adopts a single-point crossover algorithm, selects two parent individuals from the current population, randomly selects crossover sites for single-point crossover, and generates new offspring individuals; this improvement shortens the optimal flight path in a more complex environment, thus providing better conditions for optimizing the length of the continuous space path;

[0199] For the mutation operation, in order to generate new individuals that are different from their parents, the genes of the individuals are randomly perturbed. The differential evolution algorithm is used to achieve gene mutation using the differences between individuals in the population. Three individuals are randomly selected from the current population. , and ,right and Perform differential calculation to express, then the differential vector for:

[0200]

[0201] in, is the scaling factor, used to control the size of the difference;

[0202] Individual With the difference vector Combine to create new individuals :

[0203]

[0204] When the algorithm iterates to the later stage, the individual differences in the population continue to decrease. If each generation of individuals mutates again, the new individuals generated are similar to the parent generation, which is not a good effect. Therefore, this method introduces the mutation probability to control the mutation. is the number of iterations of the current population, is the maximum number of iterations, then the mutation probability for:

[0205]

[0206] in, is the mutation probability when iterating to the i-th generation; is the initial probability of mutation; as the number of algorithm iterations increases, the probability of mutation will continue to decrease, that is, as the differences between individuals in the population become smaller and smaller, the possibility of mutation will also become smaller and smaller, thereby optimizing the population diversity and being able to obtain the global optimal solution for rescue work better and faster;

[0207] After simulating the flight path of the UAV cluster using the improved genetic algorithm, the simulated flight data corresponding to each UAV is extracted; the simulated flight data includes: the airspace volume V of the rescue area, the density of UAVs in the rescue area ρ, the flight time t of each UAV's simulated rescue path, the reliability P of the successful collision avoidance of the man-machine system composed of rescue personnel and equipment, the influence coefficient k1 of environmental factors on the collision risk and the influence coefficient k2 of management factors on the collision risk. The collision index CR is further calculated based on the simulated flight data to conduct collision risk evaluation;

[0208] The simulated flight data corresponding to each UAV is input into the optimal collision risk assessment model to perform collision risk assessment, and obtain the collision risk level of each UAV when flying in the rescue area;

[0209] Finally, it is determined whether the collision risk level of each drone is lower than the preset level. If so, its simulated rescue path is used as the final rescue path; otherwise, the path is replanned and the collision risk assessment is performed;

[0210] This method combines support vector machines and genetic algorithms, uses multi-classification support vector machines to quantitatively evaluate and classify the collision risk of UAV flight paths, and uses genetic algorithms to iteratively optimize the support vector machine model. It can perform collision analysis and evaluation on the simulated rescue path. The results are scientific and reasonable, effectively improving the safety and stability of UAV rescue, while improving the accuracy and quality of path planning.

[0211] In addition, this method improves the traditional genetic algorithm, which improves the screening efficiency of the fitness function, speeds up the algorithm iteration, and generates new paths with higher quality. The improved genetic algorithm is used to plan the path, and the trained support vector machine model is used to identify risks on the planned path, which further improves the path quality and achieves high-precision and high-efficiency path planning, which is more advantageous than the traditional rescue route constructed with a single weight and model.

[0212] Example 3

[0213] like Figure 4As shown, this embodiment provides a rescue drone path planning system based on a support vector machine and a genetic algorithm, and applies a rescue drone path planning method based on a support vector machine and a genetic algorithm described in Embodiment 1 or 2, including:

[0214] Preprocessing unit 301: used to collect historical flight data of several UAVs in the rescue area, construct a historical flight data set and perform preprocessing;

[0215] Model building unit 302: used to build a collision risk assessment model based on a support vector machine;

[0216] The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area;

[0217] Model optimization unit 303: used for inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model;

[0218] Path simulation unit 304: used to plan the rescue paths of all UAVs in the rescue area using an improved genetic algorithm, obtain the simulated rescue paths of each UAV, and extract the simulated flight data corresponding to each UAV;

[0219] The collision risk assessment unit 305 is used to input the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtain the collision risk level of each UAV when flying in the rescue area;

[0220] The rescue path generation unit 306 is used to determine whether the collision risk level of each UAV is lower than the preset level. If so, the simulated rescue path is used as the final rescue path; otherwise, the path simulation unit and the collision risk assessment unit are re-executed to re-plan the path and perform collision risk assessment.

[0221] In the specific implementation process, first, the preprocessing unit 301 collects historical flight data of several drones in the rescue area, constructs a historical flight data set and performs preprocessing;

[0222] Then the model building unit 302 builds a collision risk assessment model based on the support vector machine;

[0223] The model optimization unit 303 inputs the preprocessed historical flight data set into the collision risk assessment model for training, and uses a genetic algorithm to iteratively optimize the parameters of the collision risk assessment model to obtain an optimal collision risk assessment model;

[0224] Then, the path simulation unit 304 plans the rescue paths of all UAVs in the rescue area using the improved genetic algorithm, obtains the simulated rescue paths of each UAV, and extracts the simulated flight data corresponding to each UAV;

[0225] The collision risk assessment unit 305 inputs the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtains the collision risk level of each UAV when flying in the rescue area;

[0226] Finally, the rescue path generation unit 306 determines whether the collision risk level of each drone is lower than the preset level. If so, the simulated rescue path is used as the final rescue path; otherwise, the path is replanned and the collision risk assessment is performed;

[0227] This system combines support vector machines and genetic algorithms, uses multi-classification support vector machines to quantitatively evaluate and classify the collision risk of UAV flight paths, and uses genetic algorithms to iteratively optimize the support vector machine model. It can perform collision analysis and evaluation on the simulated rescue paths. The results are scientific and reasonable, effectively improving the safety and stability of UAV rescue, while improving the accuracy and quality of path planning.

[0228] In addition, this system improves the traditional genetic algorithm and uses the improved genetic algorithm to plan the path. At the same time, it combines the trained support vector machine model to identify risks of the planned path, further improving the path quality and achieving high-precision and high-efficiency path planning. It has more advantages than the traditional rescue route constructed by single weight and model.

[0229] The same or similar reference numerals correspond to the same or similar components;

[0230] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0231] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A rescue drone path planning method based on support vector machine and genetic algorithm, characterized in that: The following steps are involved: Before the rescue, a hierarchical model based on the hierarchical analysis method is established to analyze the safety assurance capabilities of the rescue area and obtain the analysis results; The hierarchical structure model includes a target layer, a criterion layer and an indicator layer; the target layer is to evaluate the safety guarantee capability of the rescue area; The criteria layer includes: personnel security, equipment safety, operating environment safety and organizational management safety; the indicator layer includes: various evaluation indicators corresponding to personnel security, equipment safety, operating environment safety and organizational management safety; The evaluation indicators of personnel security safety include: rescuer workload, rescuer health status and rescuer professional level; The evaluation indicators of equipment safety include: rescue drone operation status, communication equipment operation status, flight control system operation status and navigation system operation status; The evaluation indicators of the safety of the operating environment include: airspace environment of the rescue area, geographical environment of the rescue area, weather environment of the rescue area and electromagnetic environment of the rescue area; The evaluation indicators of organizational management safety include: rescue manuals and rules and regulations, rescue operation procedures, rescue human resources management and rescue safety management; Determine the weight of each evaluation indicator by using the hierarchical analysis method, multiply the indicator score collected in advance for each evaluation indicator and its corresponding weight, and obtain the final score of each evaluation indicator; The final scores of all evaluation indicators are added together to obtain a comprehensive safety index as the analysis result. When the comprehensive safety index is greater than a preset threshold, steps S1 to S6 are executed; otherwise, the evaluation indicators are adjusted and the comprehensive safety index is recalculated. S1: Collect historical flight data of several drones in the rescue area, construct a historical flight data set and perform preprocessing; The historical flight data includes: the airspace volume V of the rescue area, the density of drones in the rescue area ρ, the flight time t of each drone in the rescue area, the reliability P of the successful collision avoidance of the human-machine system composed of rescue personnel and equipment, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk; The preprocessing includes: performing data noise reduction and normalization on all the collected historical flight data in turn, calculating the collision index CR of each UAV when flying in the rescue area according to the normalized historical flight data, and saving all the collision indexes CR together as a preprocessed historical flight data set to complete the preprocessing; The data noise reduction includes: data cleaning and outlier removal; The various data in each historical flight data are normalized according to the following formula: in, is the normalized data; is the data before normalization; and They are the maximum and minimum values ​​of the corresponding category data respectively; The collision index CR of each drone flying in the rescue area is calculated according to the following formula: Among them, CR1 is the collision risk of drones per unit time when only airspace factors are considered; P is the reliability of the human-machine system to successfully avoid collisions, which satisfies: Among them, P1 is the reliability of collision avoidance after the manager mediates the conflict; P2 is the reliability of the UAV successfully avoiding the collision after receiving the collision avoidance signal; P3 is the reliability of the UAV's normal flight; S2: Establish a collision risk assessment model based on support vector machine; The collision risk assessment model is used to assess the collision risk level of the UAV when it is flying in the rescue area; the collision risk assessment model is specifically: a multi-classification support vector machine model whose kernel function is a Gaussian radial basis kernel function; the parameters of the multi-classification support vector machine model include a penalty factor and a kernel parameter; The multi-classification support vector machine model is used to classify the collision risk level of each drone when flying in the rescue area into four levels: low, medium, high and very high according to the collision index CR of each drone when flying in the rescue area; S3: inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model, including: S3.1: Population initialization: Encode the parameters of the multi-classification support vector machine model into binary codes, represented by a chromosome; Randomly perform genetic inheritance on the initial population of chromosomes with 2 parameters within a preset range; S3.2: Fitness calculation: The preprocessed historical flight data set is used as a training sample to calculate the fitness value of each chromosome, and the optimal parameter value of the initial population is found by searching for the best fitness value; S3.3: Termination criteria: determine whether the preset number of iterations has been reached. If so, execute step S3.5; otherwise, execute step S3.4; S3.4: Optimization algorithm: select the next generation population parent according to the fitness value and the preset selection operator, generate a new population through crossover and mutation operations, and repeat steps S3.2~S3.3; S3.5: Save the optimal parameter value obtained from the last population iteration to obtain the optimal collision risk assessment model; S4: Based on the traditional genetic algorithm, the following improvements are made to obtain the improved genetic algorithm: Increase the penalty factor and motivational factors , thereby optimizing the fitness function, the optimized fitness function Specifically: in, is the drone serial number; , and are the first, second and third weight coefficients respectively; is the length of the simulated rescue path of the UAV; D is the Euclidean distance between the UAV and the target point; is the number of obstacles crossed by the drone in the simulated rescue path; Improvements to the selection operation: The selection operator uses the roulette selection method, assuming that the population size is , the fitness of an individual is , then the probability of being selected for: Improvements to the crossover operation: Use the single-point crossover algorithm to select two parent individuals from the current population, randomly select crossover sites for single-point crossover, and generate new offspring individuals; For mutation operation: adopt differential evolution algorithm, use the differences between individuals in the population to achieve gene mutation, and randomly select 3 individuals from the current population , and ,right and Perform differential calculation to express, then the differential vector for: in, is the scaling factor; Individual With the difference vector Combine to create new individuals : At the same time, the mutation probability is introduced to control the mutation. is the number of iterations of the current population, is the maximum number of iterations, then the mutation probability for: in, is the mutation probability when iterating to the i-th generation; is the initial probability of mutation; The improved genetic algorithm is used to plan the rescue paths of all UAVs in the rescue area, obtain the simulated rescue paths of each UAV, and extract the simulated flight data corresponding to each UAV; The simulated flight data includes: the airspace volume V of the rescue area, the density of drones in the rescue area ρ, the flight time t of each drone simulating the rescue path, the reliability P of the man-machine system composed of rescue personnel and equipment for successful collision avoidance, the influence coefficient k1 of environmental factors on the collision risk, and the influence coefficient k2 of management factors on the collision risk; S5: Inputting the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtaining the collision risk level of each UAV when flying in the rescue area; S6: Determine whether the collision risk level of each UAV is lower than the preset level. If so, use its simulated rescue path as the final rescue path; otherwise, re-execute steps S4-S5 to re-plan the path and perform collision risk assessment.

2. A rescue drone path planning system based on support vector machine and genetic algorithm, applying the rescue drone path planning method based on support vector machine and genetic algorithm described in claim 1, characterized in that: include: Preprocessing unit: used to collect historical flight data of several drones in the rescue area, construct historical flight data sets and perform preprocessing; Model building unit: used to build a collision risk assessment model based on support vector machine; The collision risk assessment model is used to assess the collision risk level of the UAV when flying in the rescue area; Model optimization unit: used for inputting the preprocessed historical flight data set into the collision risk assessment model for training, and iteratively optimizing the parameters of the collision risk assessment model using a genetic algorithm to obtain an optimal collision risk assessment model; Path simulation unit: used to plan the rescue paths of all UAVs in the rescue area using an improved genetic algorithm, obtain the simulated rescue paths of each UAV, and extract the simulated flight data corresponding to each UAV; A collision risk assessment unit is used to input the simulated flight data corresponding to each UAV into the optimal collision risk assessment model to perform collision risk assessment, and obtain the collision risk level of each UAV when flying in the rescue area; Rescue path generation unit: used to determine whether the collision risk level of each drone is lower than the preset level. If so, the simulated rescue path is used as the final rescue path; Otherwise, the path simulation unit and the collision risk assessment unit are re-executed to re-plan the path and perform the collision risk assessment.

Images