A public security patrol route planning method and system

Abstract

The present invention proposes a method and system for planning a public security patrol path, which relates to the field of drone control technology. The method includes: collecting patrol action information of multiple drones during public security patrols, wherein the patrol action information includes the path nodes, real-time motion status, public security risk areas and historical operation data passed by each drone; constructing a particle swarm evolution function according to the patrol path formed between the path nodes, the real-time motion status and the historical operation data; based on the path base point corresponding to the patrol path and the location parameters of the public security risk area, a plurality of candidate public security patrol paths are generated through a path planning fitting function, and the particle swarm evolution function generates a plurality of particles based on the plurality of candidate public security patrol paths, and each candidate public security patrol path corresponds to a particle in the particle swarm evolution function; all particles generated in the particle swarm evolution function are iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal public security patrol path. The present invention is helpful to improve the positioning accuracy of the public security patrol path.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle control, and in particular to a security patrol path planning method and system. Background Art

[0002] With the acceleration of urbanization and the increase in social security needs, drones are increasingly used in urban security patrols. UAV security patrol path planning refers to planning a flight path from the starting point to the target point for a drone that performs a security patrol mission. The path needs to meet constraints such as safe flight, path distance, and energy consumption. It is crucial for drones to complete security patrol missions safely and efficiently.

[0003] The Chinese patent with announcement number CN115752490B discloses a safe travel path optimization method and system based on big data and positioning technology. The method includes obtaining a number of initial walking paths, and constructing a graph structure with intersections and paths. Then, based on the graph structure and the risk factor information corresponding to the actual environment, an improved A* algorithm is used to form the optimal path. Its priority cost function includes movement cost, estimation cost and risk measurement cost. The risk measurement cost is the weighted sum of the dimensionless standard values ​​of each risk factor. The standardized value of the risk factor item in the risk factor is determined based on the positive or negative correlation with the environmental safety situation. However, the above scheme relies on the positioning accuracy of the path points in the modeling, and has little interaction with the actual application scenario, resulting in the problem that the path planning result cannot achieve the expected effect. Therefore, it is very necessary to provide a security patrol path planning method and system to improve the positioning accuracy of the security patrol path. Summary of the invention

[0004] In view of this, the present invention proposes a security patrol path planning method and system. By collecting patrol action information of multiple drones, the actual patrol environment and drone operation status can be fully reflected. Multiple candidate patrol paths that meet safety constraints are generated through path planning fitting functions, which helps to improve the convergence speed of particle swarm evolution functions, thereby improving the positioning accuracy of security patrol paths.

[0005] The present invention provides a method for planning a public security patrol path, the method comprising:

[0006] Collect patrol action information of multiple drones during security patrols, where the patrol action information includes the path nodes passed by each drone, real-time movement status, security risk areas, and historical operation data;

[0007] Constructing a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion state and the historical operation data;

[0008] Based on the path base points corresponding to the patrol path and the location parameters of the security risk area, a plurality of candidate security patrol paths are generated through a path planning fitting function, and the particle swarm evolution function generates a plurality of particles based on the plurality of candidate security patrol paths, and each candidate security patrol path corresponds to a particle in the particle swarm evolution function;

[0009] All particles generated in the particle swarm evolution function are iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

[0010] On the basis of the above technical solution, preferably, the expression of the particle swarm evolution function is:

[0011] x i (t+1)=x i (t)+v i (t+1)

[0012]

[0013] v i (t+1)=τv i (t)+l1·r1[IH(t)-x i (t)]+l2·r2[GH(t)-x i (t)]

[0014] Among them, x i (t+1) represents the particle position at time t+1, v i (t+1) represents the particle velocity at time t+1, τ represents the adaptive inertia weight, τ max represents the maximum value of the adaptive inertia weight, S max represents the upper limit of the domain of the activation function, τ min represents the minimum value of the adaptive inertia weight, T represents the total number of iterations, i represents the i-th iteration, l1 represents the first learning factor, r1 represents the first random factor, IH(t) represents the individual historical information data at time t, r2 represents the second random factor, l2 represents the second learning factor, and GH(t) represents the group historical information data at time t.

[0015] On the basis of the above technical solution, preferably, the expressions of the first learning factor and the second learning factor are respectively:

[0016]

[0017]

[0018] Among them, tanh() represents the hyperbolic tangent function, l max represents the upper limit of the domain of the hyperbolic tangent function, l 1max represents the maximum value of the first learning factor, l 1min represents the minimum value of the first learning factor, l 2max represents the maximum value of the second learning factor, l 2min Represents the minimum value of the second learning factor.

[0019] More preferably, the generating of a plurality of candidate public security patrol paths through a path planning fitting function based on the path base points corresponding to the patrol path and the location parameters of the public security risk area specifically includes:

[0020] The inspection path is divided into intervals to obtain a plurality of path base points, wherein each inspection path includes path base points with random positions and the same number;

[0021] According to the path base point information corresponding to the patrol path and the location parameters of the public security risk area, the risk distances of multiple path base points to the center of the public security risk area are obtained;

[0022] Based on the risk distances from the multiple path base points to the center of the public security risk area, multiple candidate patrol paths are obtained through a path planning fitting function.

[0023] More preferably, the path planning fitting function can be expressed as:

[0024]

[0025]

[0026] Where D represents the path planning fitting function, n represents the number of path nodes, m represents the number of path base points inserted between two adjacent path nodes, i represents the i-th path node, j represents the j-th path base point, and x i,j Indicates the horizontal coordinate of the jth path base point between the i-th path node and the i+1-th path node, y i,j represents the ordinate of the jth path base point between the i-th path node and the i+1-th path node, μ represents the penalty coefficient, G represents the total number of security risk areas, and d ij represents the distance from the jth path base point between the i-th path node and the i+1-th path node to the center of the g-th security risk area, x obs(g) represents the horizontal coordinate of the g-th security risk area, y obs(g) represents the ordinate of the g-th security risk area, r obs(g) Represents the radius of the g-th security risk area.

[0027] Further preferably, the particle swarm evolution function, the first learning factor and the second learning factor are used to iteratively optimize all particles generated in the particle swarm evolution function to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path, specifically including:

[0028] Initializing the original parameters input into the particle swarm evolution function, wherein the original parameters include the first learning factor, the second learning factor, the coordinates of the path nodes, the number of path nodes, the population number of the particle swarm, and the number of iterations;

[0029] Determining an optimal adaptive inertia weight interval of the particle swarm evolution function and a value strategy of the adaptive inertia weight in the optimal adaptive inertia weight interval;

[0030] Using the optimal adaptive inertia weight interval, the adaptive inertia weight value strategy in the optimal adaptive inertia weight interval, and the fitness of the particle swarm evolution function, the fitness of the particle swarm of each iteration is sorted and divided into a first sub-population and a second sub-population, wherein the first sub-population represents a sub-population greater than a population threshold, and the second sub-population represents a sub-population less than the population threshold;

[0031] Performing a copy operation on the first sub-population, and performing a crossover and mutation operation on the second sub-population, so as to update the individual optimal value and the global optimal value of the fitness of the particle swarm;

[0032] If the fitness of the particle swarm evolution function is greater than or equal to the target accuracy, the iteration process of the particle swarm evolution function is terminated, and the optimal security patrol path is obtained.

[0033] More preferably, the method further comprises:

[0034] If the fitness of the particle swarm evolution function is less than the target accuracy, multiple optimized security patrol paths are generated through the path replanning function;

[0035] The optimized security patrol path is iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal security patrol path.

[0036] More preferably, the expression of the path replanning function is:

[0037]

[0038] Among them, D re represents the path replanning function, h represents any path base point between the i-th path node and the i+1-th path node and within the g-th security risk area, C (j,h)represents the first safety penalty value of the jth path base point, G represents the total number of security risk areas, C (j,h,g) represents the second safety penalty value between the line segment between the jth path base point and the hth path base point and the gth public security risk area, S (h,t) represents the dynamic threat penalty value between the UAV and the path base point located in the g-th security risk area at time t, and A(j,j+1,j+2) represents the angle penalty value formed between the j-th path base point, the j+1-th path base point, and the j+2-th path base point.

[0039] More preferably, the expressions of the first safety penalty value, the second safety penalty value, the dynamic threat penalty value and the angle penalty value are respectively:

[0040]

[0041]

[0042]

[0043]

[0044] Among them, d (j,h) represents the distance between the jth path base point and the hth path base point, r obs(g) represents the radius of the g-th security risk area, d (j,h,g) represents the shortest distance between the line segment between the jth path base point and the hth path base point and the midpoint of the gth public security risk area, U t represents the position of the UAV at time t, g h,t represents the position of the path base point h closest to the current drone position in the g-th security risk area at time t, |||| represents the Euclidean distance, min angle represents the minimum turning angle, minD T Indicates the minimum turning distance.

[0045] In a second aspect of the present application, a security patrol path planning system is provided, the security patrol path planning system comprises a data acquisition module, a function construction module, a path simulation module and a path optimization module, wherein:

[0046] The data collection module is used to collect patrol action information of multiple drones during security patrols, wherein the patrol action information includes the path nodes passed by each drone, real-time motion status, security risk areas, and historical operation data;

[0047] The function construction module is used to construct a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion state and the historical operation data;

[0048] The path simulation module is used to generate multiple candidate public security patrol paths through a path planning fitting function based on the path base points corresponding to the patrol path and the location parameters of the public security risk area, and the particle swarm evolution function generates multiple particles based on the multiple candidate public security patrol paths, and each candidate public security patrol path corresponds to a particle in the particle swarm evolution function;

[0049] The path optimization module is used to perform iterative optimization processing on all particles generated in the particle swarm evolution function through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

[0050] The security patrol route planning method provided by the present invention has the following beneficial effects compared with the prior art:

[0051] (1) By collecting patrol action information of multiple drones, the actual patrol environment and drone operation status can be fully reflected. The patrol action information is used as the construction parameter of the particle swarm evolution function, which can improve the reliability of path planning. Through the path planning fitting function, multiple candidate patrol paths that meet safety constraints are generated, which provides a good initial solution for the subsequent particle swarm optimization and helps to improve the convergence speed of the particle swarm evolution function. At the same time, the particle swarm evolution function can dynamically adjust the motion state of particles according to individual historical information and group historical information to achieve adaptive path optimization. In addition, by introducing the first learning factor and the second learning factor to balance the role of individual experience and group experience in the optimization process, the convergence and robustness of the particle swarm evolution function are improved, thereby improving the positioning accuracy of the public security patrol path;

[0052] (2) By dividing the original patrol route into intervals and generating multiple path base points with random positions and the same number, the geometric characteristics of the patrol path can be described in more detail. This can better capture the detailed changes of the path, help generate better candidate security patrol routes, and ensure that the generated candidate paths can contain the most security risk areas by calculating the risk distance, thereby improving the patrol intensity of security risk areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1A schematic diagram of the steps of the security patrol route planning method provided by the present invention;

[0055] Figure 2 A schematic diagram of the route for the UAV security patrol provided by the present invention;

[0056] Figure 3 This is a schematic diagram of the structure of the public security patrol route planning system provided by the present invention.

[0057] Explanation of the accompanying drawings: 1. Public security patrol path planning system; 11. Data acquisition module; 12. Function construction module; 13. Path simulation module; 14. Path optimization module. DETAILED DESCRIPTION

[0058] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The present invention discloses a method for planning a security patrol path, referring to Figure 1 The method includes steps S1 to S4.

[0060] Step S1, collecting patrol action information of multiple drones during security patrols, wherein the patrol action information includes the path nodes passed by each drone, real-time movement status, security risk areas, and historical operation data.

[0061] In this step, the patrol action information can be collected in real time by the sensors carried by the drone (such as GPS, camera, etc.) to collect the drone's location information and motion status data, and combined with data sources such as the Geographic Information System (GIS) to obtain the path node information passed by the drone, and identify and mark the security risk areas through image recognition, hot spot analysis and other technologies. The real-time motion status includes real-time dynamic data such as the speed, heading and altitude of the drone. The security risk area includes the security risk areas marked according to historical event records, real-time monitoring and other information. The historical operation data includes the path, time, event records, etc. of the previous drone patrol.

[0062] Furthermore, public security risk areas refer to areas where public security incidents are prone to occur due to factors such as geographical environment, personnel composition, and historical events. Public security risk areas can be determined through geographical factors, population structure factors, historical event records, and real-time monitoring data. Among them, geographical factors include areas with remote geographical locations and strong concealment, such as back alleys, paths, and abandoned buildings; areas with poor lighting conditions and many monitoring blind spots, such as streets and parks with few people at night. Population structure factors include areas with dense populations and high mobility, such as stations, markets, and entertainment venues. At the same time, the identified public security risk areas should be re-evaluated and graded regularly, and the evaluation indicator system and grading standards should be adjusted in a timely manner in light of changes in actual conditions. An information feedback mechanism should be established to pass the latest risk assessment results to relevant law enforcement departments.

[0063] In one example, the fuzzy comprehensive evaluation method is used to quantify each indicator into 0-10 points. The weight coefficient of each indicator is determined based on expert experience, such as 0.3 for geographical environment factors, 0.4 for population structure factors, 0.2 for historical events, and 0.1 for real-time monitoring. The weighted average score of each area is calculated as the risk level. The grading standard divides the public security risk areas into three levels based on the scoring results: high-risk areas (scores of 7 points or above), medium-risk areas (scores of 4-7 points), and low-risk areas (scores of less than 4 points). Re-evaluate and grade the public security risk areas every quarter. According to the emerging geographical environment changes, population mobility, event records, etc., the evaluation indicator system and weight coefficients are adjusted in a timely manner. For high-risk areas, medium-risk areas, and low-risk areas, the inspection priority of drones is increased respectively.

[0064] Step S2, constructing a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion status and the historical operation data.

[0065] In this step, according to the aforementioned risk area assessment, various nodes such as high-risk areas and medium-risk areas that need to be focused on are determined, and the location information of these nodes is entered into the patrol path planning system. Set the initial position and speed of the particles, determine the constraints according to the actual road conditions and patrol requirements, and construct the fitness function. Reflect the time efficiency, coverage and other indicators of the patrol path. The particle swarm algorithm is used to continuously iterate and optimize the patrol path through the flight of particles and group collaboration. The flight of particles is affected by both the individual optimal solution and the group optimal solution, and finally converges to the global optimal path. At the same time, dynamic factors such as real-time traffic conditions are considered to dynamically adjust the flight speed and direction of the particles. Using GPS and other equipment mounted on the drone, the current position and driving status of the drone are obtained in real time, and the real-time motion data is fed back to the particle swarm model, and the initial state of the particles is dynamically adjusted to achieve dynamic perception of the drone position and real-time optimization of the path. And by analyzing the running trajectory of historical patrol vehicles, high-frequency passing areas and less covered areas are identified. According to the historical trajectory data, the parameter settings of the particle swarm evolution function are adjusted to improve the pertinence.

[0066] In this step, the expression of the particle swarm evolution function is:

[0067] x i (t+1)=x i (t)+v i (t+1)

[0068]

[0069] v i (t+1)=τv i (t)+l1·r1[IH(t)-x i (t)]+l2·r2[GH(t)-x i (t)]

[0070] Among them, x i (t+1) represents the particle position at time t+1, v i (t+1) represents the particle velocity at time t+1, τ represents the adaptive inertia weight, τ max represents the maximum value of the adaptive inertia weight, S max represents the upper limit of the domain of the activation function, τ min represents the minimum value of the adaptive inertia weight, T represents the total number of iterations, i represents the i-th iteration, l1 represents the first learning factor, r1 represents the first random factor, IH(t) represents the individual historical information data at time t, r2 represents the second random factor, l2 represents the second learning factor, and GH(t) represents the group historical information data at time t.

[0071] The expressions of the first learning factor and the second learning factor are:

[0072]

[0073]

[0074] Among them, tanh() represents the hyperbolic tangent function, l max represents the upper limit of the domain of the hyperbolic tangent function, l 1max represents the maximum value of the first learning factor, l 1min represents the minimum value of the first learning factor, l 2max represents the maximum value of the second learning factor, l 2min Represents the minimum value of the second learning factor.

[0075] Step S3, based on the path base points corresponding to the patrol path and the location parameters of the public security risk area, a plurality of candidate public security patrol paths are generated through the path planning fitting function, and a particle swarm evolution function generates a plurality of particles based on the plurality of candidate public security patrol paths, and each candidate public security patrol path corresponds to a particle in the particle swarm evolution function.

[0076] In this step, through the path planning fitting function, multiple candidate paths covering various security risk areas can be generated. The particle swarm evolution function will dynamically optimize these candidate paths according to the distribution of risk areas and real-time road conditions to ensure that high-risk areas are covered. The optimized patrol path can cover the security risk area more comprehensively and accurately, and improve the overall security prevention and control effect. The particle swarm evolution function continuously searches for the global optimal path through group wisdom, which can greatly shorten the time required for patrol. At the same time, it dynamically perceives real-time road condition changes and adjusts the particle flight direction in real time, effectively improving the patrol efficiency.

[0077] This step also includes steps S31 to S33.

[0078] Step S31 , dividing the inspection path into intervals to obtain a plurality of path base points, wherein each inspection path includes path base points with random positions and the same number.

[0079] In this step, if Figure 2 As shown in the figure, the number of path base points that each candidate patrol route should contain is determined based on actual patrol needs and law enforcement resources. The optimal number of base points can be determined through historical patrol experience, personnel allocation and other factors to achieve the goal of both comprehensive coverage and enforceability. In each path interval, the geographical coordinate position of a path base point is randomly generated. The random number generation algorithm is used to ensure the randomness and discreteness of the base point position to avoid excessive concentration. Randomly allocating base point positions can increase the flexibility of the solution and improve adaptability.

[0080] In one example, the setting of path base points needs to take into account the drone's endurance, maneuverability, perception, and drone formation coordination. It is necessary to determine the reasonable path length and number of path base points based on the actual endurance parameters of the drone. Too many path base points may exceed the drone's endurance range and affect the overall inspection effect. The number of path base points of the drone can be appropriately selected to use its ability to quickly switch to improve coverage efficiency, but it is also necessary to weigh the drone's precise positioning and formation coordination capabilities to ensure seamless connection between path base points. It is possible to consider multiple drones for coordinated inspections to form a formation for inspections. Each drone in the formation can be responsible for different base points to improve overall efficiency.

[0081] Furthermore, the setting of path base points needs to consider the path length and coverage range. According to the endurance and flight speed of the drone, the maximum path length that can be covered is evaluated, and the total path length is compared with the maximum flight time of the drone to determine whether the number of base points needs to be increased. For long paths that exceed the endurance of the drone, it is necessary to reasonably divide them into multiple sub-paths and increase the number of base points. Combined with the performance of the monitoring equipment carried by the drone, the monitoring radius of its single base point is determined. According to the base point monitoring radius and path length, the actual coverage area of ​​a single drone is calculated, and the coverage area is compared with the entire public security control area to determine whether the number of base points needs to be increased. For areas with insufficient coverage, the number of base points can be increased, or the distribution of base point locations can be adjusted.

[0082] Furthermore, the setting of path base points needs to consider the distribution of public security risk areas, and combine the previous public security risk area assessment to understand the distribution density and location characteristics of high-risk areas. Identify large high-risk areas and scattered small high-risk points; determine whether high-risk areas are concentrated or scattered to determine the base point distribution strategy; evaluate the dynamic characteristics of high-risk areas over time, and reserve adjustment space for base point layout. According to the distribution characteristics of high-risk areas, the number and location of base points should be appropriately adjusted. For example, for large high-risk areas, the number of path base points in the public security risk area can be appropriately increased to improve the coverage density. For scattered small high-risk points, the number of path base points in the public security risk area can be increased to reasonably distribute the coverage. For dynamically changing high-risk areas, the number of path base points near the public security risk area can be appropriately increased to cope with changes.

[0083] Furthermore, the setting of path base points needs to consider historical inspection experience, analyze the previous drone inspection paths and base point layout to understand the best practices. By comparing and analyzing historical data, we can gain an in-depth understanding of the following path planning schemes, base point distribution, and inspection effect evaluation. Among them, the path planning scheme includes the drone inspection path schemes used in history, such as path length, number of path nodes, number of path base points, etc. The base point distribution includes the location distribution of the previously arranged path base points, and determines whether the key areas are covered. The inspection effect evaluation includes comparing the actual implementation effect of the historical scheme to understand the advantages and disadvantages of the historical scheme. Drawing on the effect evaluation of the previous drone inspection, we can judge whether the current number of base points is reasonable. If the historical scheme works well, we can make appropriate adjustments and optimizations on this basis. If there are problems with the historical scheme, we need to make more substantial adjustments based on the new law enforcement needs. According to the changes in the new security risk areas, we can appropriately adjust the number of base points and continuously optimize the inspection plan. For the newly added high-risk areas, we can refer to historical experience to increase the number of base points. For the original low-risk areas, we can refer to historical experience to appropriately reduce the number of base points.

[0084] Step S32, obtaining the risk distances from multiple path base points to the center of the public security risk area based on the path base point information corresponding to the patrol path and the location parameters of the public security risk area.

[0085] In this step, the geographic coordinate information of each path node and path base point is obtained according to the pre-planned drone patrol path. The path base point location data is sorted into a path data set. Combined with the previous public security risk assessment work, the geographic location coordinates of each high-risk area are obtained. The center point coordinates of these high-risk areas are sorted into a risk data set. Use appropriate spatial distance calculation methods (such as Euclidean distance, Manhattan distance, etc.) to calculate the distance from each path base point to the center point of each public security risk area. Record these distance values ​​to form a distance matrix between the base point and the center of the risk area. Determine key distance indicators such as minimum distance, average distance, and maximum distance according to public security patrol needs. Associate the calculated distance indicators with the corresponding path base point information to form a complete data set.

[0086] Step S33, based on the risk distances from multiple path base points to the center of the public security risk area, multiple candidate inspection paths are obtained through a path planning fitting function.

[0087] The path planning fitting function can be expressed as:

[0088]

[0089]

[0090] Where D represents the path planning fitting function, n represents the number of path nodes, m represents the number of path base points inserted between two adjacent path nodes, i represents the i-th path node, j represents the j-th path base point, and x i,j Indicates the horizontal coordinate of the jth path base point between the i-th path node and the i+1-th path node, y i,j represents the ordinate of the jth path base point between the i-th path node and the i+1-th path node, μ represents the penalty coefficient, G represents the total number of security risk areas, and d ij represents the distance from the jth path base point between the i-th path node and the i+1-th path node to the center of the g-th security risk area, x obs(g) represents the horizontal coordinate of the g-th security risk area, y obs(g) represents the ordinate of the g-th security risk area, r obs(g) Represents the radius of the g-th security risk area.

[0091] By dividing the original patrol route into intervals and generating multiple path base points with random positions and the same number, the geometric characteristics of the patrol path can be described in more detail. This can better capture the detailed changes of the path, help generate higher-quality candidate security patrol routes, and by calculating the risk distance, ensure that the generated candidate paths can contain the most security risk areas, thereby improving the patrol intensity of security risk areas.

[0092] Step S4, performing iterative optimization processing on all particles generated in the particle swarm evolution function through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

[0093] This step also includes steps S41 to S45.

[0094] Step S41, initializing the original parameters of the input particle swarm evolution function, wherein the original parameters include a first learning factor, a second learning factor, the coordinates of the path nodes, the number of path nodes, the population size of the particle swarm, and the number of iterations.

[0095] In this step, based on the multiple candidate inspection paths obtained in the previous step, the coordinate information of the path nodes on each path is extracted. The coordinate information of these path nodes is used as the initialization input to provide the particle swarm algorithm with a search space range. Combined with the candidate path solutions obtained above, the number of path base points on each path is determined. Select a suitable number of base points as the initialization input parameter to provide a benchmark for subsequent path optimization. Determine the population size of the particle swarm algorithm, that is, how many particles need to be generated during initialization. The selection of the population size needs to balance between computational efficiency and solution quality, and is usually set between 50-100. Set the maximum number of iterations of the particle swarm algorithm, that is, how many rounds of iterations the optimization process will perform. The number of iterations should be determined based on factors such as the complexity of the problem and the required accuracy, and is usually set to 200-500 times.

[0096] Step S42, determining the optimal adaptive inertia weight interval of the particle swarm evolution function and the value strategy of the adaptive inertia weight in the optimal adaptive inertia weight interval.

[0097] In this step, the optimal range of the adaptive inertia weight interval is [0.4, 0.9]. Therefore, in most applications based on the particle swarm algorithm, the inertia weight interval is [0.4, 0.9] as the upper and lower limits of the inertia weight value. However, for different optimization problems, the range of the inertia weight interval is not fixed, so the improved particle swarm algorithm is used to optimize the inertia weight interval.

[0098] Furthermore, according to the disclosed inertia weight reference range in the classic PSO algorithm, multiple inertia weight reference values ​​are obtained with a fixed value as the interval interval; the classic PSO algorithm corresponding to the multiple inertia weight reference values ​​is operated with a group number of T1 and an iteration number of T2; and the classic PSO algorithm corresponding to each inertia weight reference value is run T3 times and the average value is taken to obtain the reference fitness corresponding to each inertia weight reference value.

[0099] In this embodiment, the inertia weight reference range is [0, 1.4], and the inertia weight reference range is [0, 1.4] with an interval of 0.1, and the values ​​are 0, 0.1, 0.2, 0.3, ..., 1.4, a total of 15 values, and the classic PSO algorithm corresponding to each inertia weight reference value is operated with a population size of 50 and an iteration number of 50. In order to avoid the influence of random numbers in the algorithm, the PSO algorithm corresponding to each inertia weight reference value is run 10 times and the average value is taken to obtain the corresponding fitness f1, f2, ..., f 15 .

[0100] Compare the sizes of multiple reference fitnesses, take the two smallest reference fitnesses, use the inertia weight reference value corresponding to the small reference fitness as the lower limit weight, and use the inertia weight reference value corresponding to the large reference fitness as the upper limit weight to obtain the optimal inertia weight interval.

[0101] In this embodiment, the fitness f1, f2, ..., f 15 The size of the inertia weight interval [w min ,w max ]. Among them, w min is the fitness f1, f2, ..., f 15 The inertia weight reference value corresponding to the minimum fitness in w. max is the fitness f1, f2, ..., f 15 The inertia weight reference value corresponding to the maximum fitness.

[0102] In this embodiment, two strategies are adopted. The first strategy is that the inertia weight in the optimal inertia weight interval gradually increases with the increase of the number of iterations; the second strategy is that the inertia weight in the optimal inertia weight interval gradually decreases with the increase of the number of iterations.

[0103] The expression of the inertia weight in the optimal inertia weight interval gradually increasing with the increase of the number of iterations is:

[0104] w=w min +(w max -w min )*(t / T) 2

[0105] Where: w represents the inertia weight of the current iteration, w min Indicates the inertia weight reference value corresponding to the minimum fitness, w max It represents the inertia weight reference value corresponding to the maximum fitness, t represents the current number of iterations, and T represents the maximum number of iterations.

[0106] The expression of the inertia weight in the optimal inertia weight interval gradually decreasing with the increase of the number of iterations is:

[0107] w=w max -(w max -w min )*(t / T) 2

[0108] The fitness results obtained under the two strategies are compared, and the smaller fitness result is taken as the inertia weight value strategy in the optimal inertia weight interval of the improved Retinanet deep learning model.

[0109] Step S43, using the optimal adaptive inertia weight interval, the adaptive inertia weight value strategy in the optimal adaptive inertia weight interval and the fitness of the particle swarm evolution function, sort the fitness of the particle swarm of each iteration and divide it into a first sub-population and a second sub-population, wherein the first sub-population represents a sub-population greater than the population threshold, and the second sub-population represents a sub-population less than the population threshold.

[0110] In this step, the optimal inertia weight interval [w min ,w max ] will be used in this step to calculate the inertia weight w of the current iteration according to the selected adaptive inertia weight update strategy (increasing or decreasing). Record the fitness value of each particle (i.e. the quality of the patrol path). Sort these fitness values ​​in ascending order to obtain excellent individuals with high rankings and inferior individuals with low rankings. Set a population threshold as the basis for dividing the first sub-population and the second sub-population. Particles ranked higher than the population threshold are divided into the first sub-population, and these individuals have higher fitness. Particles ranked lower than the population threshold are divided into the second sub-population, and these individuals have lower fitness.

[0111] In one example, the initial population size is 100 particles, each particle represents a drone inspection path, and the fitness value is calculated based on the total distance of each inspection path, the estimated inspection time and other indicators. The fitness values ​​of the 100 particles are sorted in ascending order, and the population threshold is initially set to 50%, that is, the first 50 particles are divided into the first sub-population, and the last 50 particles are divided into the second sub-population. The fitness distribution of the first sub-population and the second sub-population is observed, and it is found that the average fitness of the first sub-population is much higher than that of the second sub-population, indicating that the division effect is good, otherwise, the division effect is poor. As the number of iterations increases, the population threshold is appropriately increased, such as increasing to 60%. If the fitness of the first sub-population is significantly improved at this time, it means that the algorithm focuses more on utilization, but the fitness of the second sub-population decreases more, indicating that the algorithm's exploration ability has decreased. Taking into account the exploration and utilization capabilities of the algorithm, it is decided to adopt an adaptive population threshold strategy. The population threshold increases linearly with the number of iterations, from 50% to 80%. This way, a strong exploration capability can be maintained in the early iterations and the utilization capability can be enhanced in the later iterations.

[0112] Step S44, performing a copy operation on the first sub-population, and performing a crossover and mutation operation on the second sub-population, so as to update the individual optimal value and the global optimal value of the particle swarm fitness.

[0113] In this step, individuals with higher fitness are selected from the first subpopulation for replication. The replication operation can directly copy excellent individuals to the next generation population to retain high-quality solutions. The replication operation can perform probabilistic selection based on the size of individual fitness. Individuals with higher fitness have a greater probability of being selected. Through the replication operation, the excellent features in the first subpopulation can be effectively retained and utilized. Individuals with lower fitness are selected from the second subpopulation for crossover and mutation operations. The crossover operation can exchange some decision variables between two individuals to generate new solutions. The mutation operation can randomly perturb the decision variables of individuals to increase the diversity of the population. Through the crossover and mutation operations, the solution space in the second subpopulation can be effectively explored to discover potential high-quality solutions. After completing the replication operation of the first subpopulation and the crossover and mutation operations of the second subpopulation, the fitness value of each particle needs to be recalculated. According to the new fitness value, the individual optimal value (pbest) of each particle is updated. At the same time, the global optimal value (gbest) of the entire particle swarm also needs to be updated. The updated individual optimal value and global optimal value will provide a reference for subsequent particle swarm updates.

[0114] Step S45: if the fitness of the particle swarm evolution function is greater than or equal to the target accuracy, the iteration process of the particle swarm evolution function is terminated, and the optimal security patrol path is obtained.

[0115] Furthermore, if the fitness of the particle swarm evolution function is less than the target accuracy, multiple optimized security patrol paths are generated through the path replanning function; the optimized security patrol path is iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal security patrol path.

[0116] The expression of the path replanning function is:

[0117]

[0118] Among them, D re represents the path replanning function, h represents any path base point between the i-th path node and the i+1-th path node and within the g-th security risk area, C (j,h) represents the first safety penalty value of the jth path base point, G represents the total number of security risk areas, C (j,h,g) represents the second safety penalty value between the line segment between the jth path base point and the hth path base point and the gth public security risk area, S (h,t) represents the dynamic threat penalty value between the UAV and the path base point located in the g-th security risk area at time t, and A(j,j+1,j+2) represents the angle penalty value formed between the j-th path base point, the j+1-th path base point, and the j+2-th path base point.

[0119] The expressions for the first safety penalty value, the second safety penalty value, the dynamic threat penalty value, and the angle penalty value are:

[0120]

[0121]

[0122]

[0123]

[0124] Among them, d (j,h) represents the distance between the jth path base point and the hth path base point, r obs(g) represents the radius of the g-th security risk area, d (j,h,g) represents the shortest distance between the line segment between the jth path base point and the hth path base point and the midpoint of the gth public security risk area, U t represents the position of the UAV at time t, g h,t represents the position of the path base point h closest to the current drone position in the g-th security risk area at time t, |||| represents the Euclidean distance, min angle represents the minimum turning angle, minD T Indicates the minimum turning distance.

[0125] By collecting patrol action information of multiple drones, the actual patrol environment and drone operation status can be fully reflected. Taking the patrol action information as the construction parameter of the particle swarm evolution function can improve the reliability of path planning. Through the path planning fitting function, multiple candidate patrol paths that meet the safety constraints are generated, which provides a good initial solution for the subsequent particle swarm optimization and helps to improve the convergence speed of the particle swarm evolution function. At the same time, the particle swarm evolution function can dynamically adjust the motion state of particles according to individual historical information and group historical information to achieve adaptive path optimization. By introducing the first learning factor and the second learning factor to balance the role of individual experience and group experience in the optimization process, the convergence and robustness of the particle swarm evolution function are improved, thereby improving the positioning accuracy of the public security patrol path.

[0126] Based on the above method, the present application embodiment discloses a security patrol path planning system, referring to Figure 3 The security patrol path planning system 1 includes a data collection module 11, a function construction module 12, a path simulation module 13 and a path optimization module 14, wherein:

[0127] The data collection module 11 is used to collect patrol action information of multiple drones during security patrols, wherein the patrol action information includes the path nodes passed by each drone, real-time motion status, security risk areas, and historical operation data;

[0128] The function construction module 12 is used to construct a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion state and the historical operation data;

[0129] The path simulation module 13 is used to generate multiple candidate public security patrol paths through a path planning fitting function based on the path base points corresponding to the patrol path and the location parameters of the public security risk area. The particle swarm evolution function generates multiple particles based on the multiple candidate public security patrol paths, and each candidate public security patrol path corresponds to a particle in the particle swarm evolution function;

[0130] The path optimization module 14 is used to perform iterative optimization processing on all particles generated in the particle swarm evolution function through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

[0131] In one example, the expression of the particle swarm evolution function is:

[0132] x i (t+1)=x i (t)+v i (t+1)

[0133]

[0134] v i (t+1)=τv i (t)+l1·r1[IH(t)-x i (t)]+l2·r2[GH(t)-x i (t)]

[0135] Among them, x i (t+1) represents the particle position at time t+1, v i (t+1) represents the particle velocity at time t+1, τ represents the adaptive inertia weight, τ max represents the maximum value of the adaptive inertia weight, S max represents the upper limit of the domain of the activation function, τ min represents the minimum value of the adaptive inertia weight, T represents the total number of iterations, i represents the i-th iteration, l1 represents the first learning factor, r1 represents the first random factor, IH(t) represents the individual historical information data at time t, r2 represents the second random factor, l2 represents the second learning factor, and GH(t) represents the group historical information data at time t.

[0136] In an example, the expressions of the first learning factor and the second learning factor are respectively:

[0137]

[0138]

[0139] Among them, tanh() represents the hyperbolic tangent function, l max represents the upper limit of the domain of the hyperbolic tangent function, l 1max represents the maximum value of the first learning factor, l 1min represents the minimum value of the first learning factor, l 2max represents the maximum value of the second learning factor, l 2min Represents the minimum value of the second learning factor.

[0140] In one example, the path simulation module 13 is used to divide the patrol path into intervals to obtain multiple path base points, wherein each patrol path includes path base points with random positions and the same number; based on the path base point information corresponding to the patrol path and the location parameters of the public security risk area, the risk distances from the multiple path base points to the center of the public security risk area are obtained; based on the risk distances from the multiple path base points to the center of the public security risk area, multiple candidate patrol paths are obtained through a path planning fitting function.

[0141] In one example, the path planning fitting function can be expressed as:

[0142]

[0143]

[0144] Where D represents the path planning fitting function, n represents the number of path nodes, m represents the number of path base points inserted between two adjacent path nodes, i represents the i-th path node, j represents the j-th path base point, and x i,j Indicates the horizontal coordinate of the jth path base point between the i-th path node and the i+1-th path node, y i,j represents the ordinate of the jth path base point between the i-th path node and the i+1-th path node, μ represents the penalty coefficient, G represents the total number of security risk areas, and d ij represents the distance from the jth path base point between the i-th path node and the i+1-th path node to the center of the g-th security risk area, x obs(g) represents the horizontal coordinate of the g-th security risk area, y obs(g) represents the ordinate of the g-th security risk area, r obs(g) Represents the radius of the g-th security risk area.

[0145] In one example, the path optimization module 14 is used to initialize the original parameters of the input particle swarm evolution function, wherein the original parameters include a first learning factor, a second learning factor, the coordinates of the path nodes, the number of path nodes, the population number of the particle swarm, and the number of iterations; determine the optimal adaptive inertia weight interval of the particle swarm evolution function and the value strategy of the adaptive inertia weight in the optimal adaptive inertia weight interval; use the optimal adaptive inertia weight interval, the value strategy of the adaptive inertia weight in the optimal adaptive inertia weight interval, and the fitness of the particle swarm evolution function to sort the fitness of the particle swarm of each iteration and divide it into a first sub-population and a second sub-population, wherein the first sub-population represents a sub-population greater than a population threshold, and the second sub-population represents a sub-population less than a population threshold; perform a copy operation on the first sub-population, and perform a crossover and mutation operation on the second sub-population to update the individual optimal value and the global optimal value of the particle swarm fitness; if the fitness of the particle swarm evolution function is greater than or equal to the target accuracy, then end the iterative process of the particle swarm evolution function and obtain the optimal security patrol path.

[0146] In one example, the path optimization module 14 is used to generate multiple optimized security patrol paths through the path replanning function if the fitness of the particle swarm evolution function is less than the target accuracy; the optimized security patrol path is iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal security patrol path.

[0147] In one example, the expression of the path replanning function is:

[0148]

[0149] Among them, D re represents the path replanning function, h represents any path base point between the i-th path node and the i+1-th path node and within the g-th security risk area, C (j,h) represents the first safety penalty value of the jth path base point, G represents the total number of security risk areas, C (j,h,g) represents the second safety penalty value between the line segment between the jth path base point and the hth path base point and the gth public security risk area, S (h,t) represents the dynamic threat penalty value between the UAV and the path base point located in the g-th security risk area at time t, and A(j,j+1,j+2) represents the angle penalty value formed between the j-th path base point, the j+1-th path base point, and the j+2-th path base point.

[0150] In one example, the expressions of the first safety penalty value, the second safety penalty value, the dynamic threat penalty value, and the angle penalty value are respectively:

[0151]

[0152]

[0153]

[0154]

[0155] Among them, d (j,h) represents the distance between the jth path base point and the hth path base point, r obs(g) represents the radius of the g-th security risk area, d (j,h,g) represents the shortest distance between the line segment between the jth path base point and the hth path base point and the midpoint of the gth public security risk area, U t represents the position of the UAV at time t, g h,t represents the position of the path base point h closest to the current drone position in the g-th security risk area at time t, |||| represents the Euclidean distance, min angle represents the minimum turning angle, minD T Indicates the minimum turning distance.

[0156] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A security patrol route planning method, characterized in that: The method comprises: Collect patrol action information of multiple drones during security patrols, wherein the patrol action information includes the path nodes passed by each drone, real-time movement status, security risk areas, and historical operation data; Constructing a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion state and the historical operation data; The expression of the particle swarm evolution function is: x i (t+1)=x i (t)+v i (t+1) v i (t+1)=τv i (t)+l1·r1[IH(t)-x i (t)]+l2·r2[GH(t)-x i (t)] Among them, x i (t+1) represents the particle position at time t+1, v i (t+1) represents the particle velocity at time t+1, τ represents the adaptive inertia weight, τ max represents the maximum value of the adaptive inertia weight, S max represents the upper limit of the domain of the activation function, τ min represents the minimum value of the adaptive inertia weight, T represents the total number of iterations, i represents the i-th iteration, l1 represents the first learning factor, r1 represents the first random factor, IH(t) represents the individual historical information data at time t, r2 represents the second random factor, l2 represents the second learning factor, and GH(t) represents the group historical information data at time t; Based on the path base points corresponding to the patrol path and the location parameters of the security risk area, a plurality of candidate security patrol paths are generated through a path planning fitting function, and the particle swarm evolution function generates a plurality of particles based on the plurality of candidate security patrol paths, and each candidate security patrol path corresponds to a particle in the particle swarm evolution function; All particles generated in the particle swarm evolution function are iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

2. The method according to claim 1, characterized in that The expressions of the first learning factor and the second learning factor are respectively: Among them, tanh() represents the hyperbolic tangent function, l max represents the upper limit of the domain of the hyperbolic tangent function, l 1max represents the maximum value of the first learning factor, l 1min represents the minimum value of the first learning factor, l 2max represents the maximum value of the second learning factor, l 2min Represents the minimum value of the second learning factor.

3. The method according to claim 1, characterized in that The generating of a plurality of candidate public security patrol paths through a path planning fitting function based on the path base points corresponding to the patrol path and the location parameters of the public security risk area specifically includes: The inspection path is divided into intervals to obtain a plurality of path base points, wherein each inspection path includes path base points with random positions and the same number; According to the path base point information corresponding to the patrol path and the location parameters of the public security risk area, the risk distances of multiple path base points to the center of the public security risk area are obtained; Based on the risk distances from the multiple path base points to the center of the public security risk area, multiple candidate patrol paths are obtained through a path planning fitting function.

4. The method according to claim 1, characterized in that The path planning fitting function can be expressed as: Where D represents the path planning fitting function, n represents the number of path nodes, m represents the number of path base points inserted between two adjacent path nodes, i represents the i-th path node, j represents the j-th path base point, and x i,j Indicates the horizontal coordinate of the jth path base point between the i-th path node and the i+1-th path node, y i,j represents the ordinate of the jth path base point between the i-th path node and the i+1-th path node, μ represents the penalty coefficient, G represents the total number of security risk areas, and d ij represents the distance from the jth path base point between the i-th path node and the i+1-th path node to the center of the g-th security risk area, x obs(g) represents the horizontal coordinate of the g-th security risk area, y obs(g) represents the ordinate of the g-th security risk area, r obs(g) Represents the radius of the g-th security risk area.

5. The method according to claim 1, characterized in that The iterative optimization process is performed on all particles generated in the particle swarm evolution function through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path, specifically including: Initializing the original parameters input into the particle swarm evolution function, wherein the original parameters include the first learning factor, the second learning factor, the coordinates of the path nodes, the number of path nodes, the population number of the particle swarm, and the number of iterations; Determining an optimal adaptive inertia weight interval of the particle swarm evolution function and a value strategy of the adaptive inertia weight in the optimal adaptive inertia weight interval; Using the optimal adaptive inertia weight interval, the adaptive inertia weight value strategy in the optimal adaptive inertia weight interval, and the fitness of the particle swarm evolution function, the fitness of the particle swarm of each iteration is sorted and divided into a first sub-population and a second sub-population, wherein the first sub-population represents a sub-population greater than a population threshold, and the second sub-population represents a sub-population less than the population threshold; Performing a copy operation on the first sub-population, and performing a crossover and mutation operation on the second sub-population, so as to update the individual optimal value and the global optimal value of the fitness of the particle swarm; If the fitness of the particle swarm evolution function is greater than or equal to the target accuracy, the iteration process of the particle swarm evolution function is terminated, and the optimal security patrol path is obtained.

6. The method according to claim 5, characterized in that The method further comprises: If the fitness of the particle swarm evolution function is less than the target accuracy, multiple optimized security patrol paths are generated through the path replanning function; The optimized security patrol path is iteratively optimized through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal security patrol path.

7. The method according to claim 6, characterized in that The expression of the path replanning function is: Among them, D re represents the path replanning function, h represents any path base point between the i-th path node and the i+1-th path node and within the g-th security risk area, C (j,h) represents the first safety penalty value of the jth path base point, G represents the total number of security risk areas, C (j,h,g) represents the second safety penalty value between the line segment between the jth path base point and the hth path base point and the gth public security risk area, S (h,t) represents the dynamic threat penalty value between the UAV and the path base point located in the g-th security risk area at time t, and A(j,j+1,j+2) represents the angle penalty value formed between the j-th path base point, the j+1-th path base point, and the j+2-th path base point.

8. The method according to claim 7, characterized in that The expressions of the first safety penalty value, the second safety penalty value, the dynamic threat penalty value and the angle penalty value are respectively: Among them, d (j,h) represents the distance between the jth path base point and the hth path base point, r obs(g) represents the radius of the g-th security risk area, d (j,h,g) represents the shortest distance between the line segment between the jth path base point and the hth path base point and the midpoint of the gth security risk area, U t represents the position of the UAV at time t, g h,t represents the position of the path base point h closest to the current drone position in the g-th security risk area at time t, |||| represents the Euclidean distance, minangle represents the minimum turning angle, minD T Indicates the minimum turning distance.

9. A security patrol route planning system, characterized in that: The public security patrol path planning system (1) comprises a data collection module (11), a function construction module (12), a path simulation module (13) and a path optimization module (14), wherein: The data collection module (11) is used to collect patrol action information of multiple drones during security patrols, wherein the patrol action information includes the path nodes passed by each drone, real-time movement status, security risk areas and historical operation data; The function construction module (12) is used to construct a particle swarm evolution function according to the inspection path formed between the path nodes, the real-time motion state and the historical operation data; The expression of the particle swarm evolution function is: x i (t+1)=x i (t)+v i (t+1) v i (t+1)=τv i (t)+l1·r1[IH(t)-x i (t)]+l2·r2[GH(t)-x i (t)] Among them, x i (t+1) represents the particle position at time t+1, v i (t+1) represents the particle velocity at time t+1, τ represents the adaptive inertia weight, τ max represents the maximum value of the adaptive inertia weight, S max represents the upper limit of the domain of the activation function, τ min represents the minimum value of the adaptive inertia weight, T represents the total number of iterations, i represents the i-th iteration, l1 represents the first learning factor, r1 represents the first random factor, IH(t) represents the individual historical information data at time t, r2 represents the second random factor, l2 represents the second learning factor, and GH(t) represents the group historical information data at time t; The path simulation module (13) is used to generate a plurality of candidate public security patrol paths through a path planning fitting function based on the path base points corresponding to the patrol path and the location parameters of the public security risk area, and the particle swarm evolution function generates a plurality of particles based on the plurality of candidate public security patrol paths, and each candidate public security patrol path corresponds to a particle in the particle swarm evolution function; The path optimization module (14) is used to perform iterative optimization processing on all particles generated in the particle swarm evolution function through the particle swarm evolution function, the first learning factor and the second learning factor to obtain the optimal particle, wherein the optimal particle represents the optimal security patrol path.

Images