A node multi-objective optimization deployment method for portable perimeter security

Abstract

The application discloses a node multi-objective optimization deployment method for portable perimeter security, and relates to the technical field of wireless sensor network planning and deployment. The specific implementation scheme is as follows: terrain elevation and surface classification data are collected and rasterized, a probability perception model considering terrain shielding and distance attenuation is established, and the grid invasion threat degree is calculated according to the visibility and accessibility; an optimization function with three sub-targets of coverage rate, energy consumption balance and reliable connectivity and with dynamic weighting in stages is constructed; first, skeleton nodes are selected by marginal gain greediness and a support tree is constructed, then the position, type and orientation of the coding nodes are combined, and a Pareto solution set is obtained through neighborhood distributed evolution; a compromise scheme is selected according to the stage weight, and a deployment list is output after supplementing and compensating nodes through double verification of coverage and connectivity. Through the application, a deployment scheme that is adaptive to terrain and threat distribution, covers no blind area, is reliable in connectivity, balanced in energy consumption and adaptive to stages can be quickly obtained without a central unit.

Description

Technical Field

[0001] This application relates to the field of wireless sensor network planning and deployment technology, and more specifically, to a multi-target optimized deployment method for nodes for portable perimeter surveillance. Background Technology

[0002] Portable perimeter surveillance systems are used in areas such as temporary camps without fixed security facilities, temporary protection of key locations, security for major events, and disaster relief sites. They achieve intrusion detection by rapidly deploying multiple sensor nodes to form a wireless sensor network along the perimeter. In such scenarios, the terrain is complex and cannot be accurately surveyed in advance. The nodes are battery-powered and have limited computing power. Personnel are required to quickly establish a perimeter control zone after arrival and maintain long-term monitoring after the zone is established. The number of nodes that can be deployed is limited, and the node deployment scheme directly determines the perimeter coverage integrity, network connectivity reliability, and continuous monitoring time.

[0003] Existing sensor node deployment methods mainly include: regular deployment methods that arrange nodes on the perimeter at equal intervals or in regular geometric patterns, with the goal of maximizing geometric coverage; optimization deployment methods that use genetic algorithms, particle swarm optimization, or other methods to perform global optimization in continuous coordinate space, with the goal of coverage, number of nodes, or energy consumption at the central computing unit; and coverage connectivity deployment methods that introduce communication connectivity constraints in addition to coverage optimization and organize the network topology with minimum spanning trees or Thiessen polygons.

[0004] However, the aforementioned prior art has at least the following shortcomings: First, existing deployment methods generally adopt an ideal perception disk model and take regular areas as the planning premise. They do not introduce the actual terrain elevation and surface classification data of the deployment area, and cannot characterize the impact of terrain occlusion and distance attenuation on the actual perception ability of nodes. This results in a mismatch between the planned coverage and the actual effective coverage on site, and the formation of actual perception blind spots on the perimeter. Secondly, existing methods aim for maximum geometric coverage, treating all perimeter locations as equally important. They fail to differentiate the degree of intrusion threat at different locations based on terrain visibility and surface accessibility. A limited number of nodes are distributed evenly to low-threat areas, while high-threat road sections that are easy for people to approach covertly are left undefended. Third, existing multi-objective deployment methods mostly rely on centralized computing units to solve problems and the weights of each objective are fixed. The calculation is time-consuming and it is difficult to complete the planning quickly on portable nodes with limited computing power. Furthermore, they cannot adapt to the phased changes in demand, such as prioritizing coverage during rapid containment and prioritizing energy consumption balance during long-term monitoring. At the same time, coverage optimization and connectivity assurance are disconnected from each other, which can easily lead to the problem of achieving coverage targets but breaking the communication topology. Fourth, existing methods mostly assume that nodes are isomorphic and only optimize node coordinates, without taking into account the selection of heterogeneous node types and the orientation of directional sensors in deployment optimization. Mismatch between node types and surface characteristics and deviation in detection orientation further reduce the actual sensing efficiency. Moreover, after the scheme is determined, there is a lack of verification and compensation for actual coverage and connectivity, and the remaining blind spots cannot be eliminated. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a multi-target optimization deployment method for portable perimeter surveillance nodes. This method can achieve the following by establishing a probabilistic perception model that considers terrain occlusion and distance attenuation and introducing intrusion threat-weighted coverage, so that the node deployment matches the actual perception conditions of complex terrain and the distribution of perimeter threats. At the same time, under the constraint of the connectivity skeleton, distributed multi-target optimization is performed based on the local information exchange in the neighborhood, and the output scheme is verified and compensated. Thus, without the need for a central computing unit, a deployment scheme with no blind spots in actual coverage, reliable network connectivity, balanced node energy consumption, and a deployment focus that can be dynamically adjusted according to the stage of the surveillance task, is obtained in an integrated manner, with node position, type and orientation.

[0006] According to a first aspect of this application, a multi-target optimized deployment method for nodes in portable perimeter surveillance is provided, comprising: Collect terrain elevation and surface classification data and rasterize the perimeter. Establish a probabilistic perception model that considers terrain occlusion and distance attenuation. Calculate the intrusion threat level of each grid based on visibility and surface accessibility. A multi-objective optimization function is constructed with threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability as sub-objectives. The threat-weighted effective coverage is the sum of the comprehensive perception probabilities of each grid weighted by the threat degree, and dynamic weights are set according to the stage. We extract the extreme points of perimeter curvature and the highest points as candidate backbone positions. Under connectivity constraints, we evaluate the marginal gain using an optimization function, prioritize the high-threat grid coverage with the largest gain, greedily select the backbone nodes, and construct a connected support tree. Based on the skeleton, each node exchanges neighborhood information and jointly encodes its position, type and orientation angle. Using the three objectives as fitness, a multi-objective evolutionary algorithm is used for iteration. After non-dominated sorting and crowding, a Pareto optimal deployment scheme set is obtained. The sub-objectives of each scheme in the scheme set are weighted and summed according to the current stage dynamic weights. The compromise deployment scheme with the largest comprehensive objective value is selected. The grid threat weighted perception probability and network connectivity are verified. Compensation nodes are added for areas below the threshold. The deployment list with location, type and orientation is output.

[0007] In some embodiments, a probabilistic perception model considering terrain occlusion and distance attenuation is established, including: Obtain the deployment coordinates, sensor installation height, and effective sensing radius of each node to be deployed, as well as the center coordinates and ground elevation of each grid cell, and calculate the horizontal distance from the node to the center of the grid cell; Elevation profiles are extracted along the line connecting the node and the center of the grid cell at sampling intervals of no more than 1 meter. The actual ground elevation of each sampling point on the line is compared with the line-of-sight elevation of the sampling point on the line connecting the node sensor installation height to the grid ground elevation. If the actual ground elevation of any sampling point is higher than the line-of-sight elevation, it is determined that there is terrain occlusion, and the perception probability of the node for the grid cell is set to 0. If there is no terrain obstruction, the perception probability decreases exponentially with horizontal distance. When the horizontal distance is 0, the perception probability is 1. When the horizontal distance is equal to the effective perception radius, the perception probability decreases to 0.5. When the horizontal distance is greater than twice the effective perception radius or the perception probability decreases to a probability threshold below 0.05, the perception probability is set to 0. For all nodes covering the same grid cell, the overall perception probability of the grid cell is calculated by subtracting the product of the missed detection probabilities of each node for that grid cell from 1. The missed detection probability of a single node is equal to 1 minus the perception probability of that node for that grid cell.

[0008] In some embodiments, the intrusion threat level of each grid is calculated based on visibility and surface accessibility, including: Starting from a grid cell, three lines of sight are emitted along the outer perimeter normal direction and at 15-degree intervals on both sides of the normal, forming a total of seven lines of sight. Within a 50-meter warning depth range from the perimeter, elevation visibility is assessed point-by-point along each line of sight. The proportion of the unobstructed segment from the grid cell to the first obstruction point for each line of sight is calculated to the total length of that line of sight within the warning depth range. If the entire line of sight is unobstructed, this proportion is set to 1. The arithmetic mean of this proportion for all seven lines of sight is taken as the visibility factor for that grid cell. The passage cost is calculated based on the slope value and surface type at each point within the warning depth range outside the grid cell. The slope value is calculated from the elevation gradient of the terrain elevation data, and the slope is less than 15 degrees. The slope access cost is set to 0 when the slope is greater than 30 degrees, and to 1 when the slope is greater than 30 degrees. For slopes between 15 and 30 degrees, linear interpolation is used to determine the access cost. The surface access cost is determined based on surface classification data. The surface access cost is obtained from a preset surface access cost mapping table and normalized to the interval between 0 and 1. The slope access cost is weighted with a weight of 0.6 and the surface type access cost is weighted with a weight of 0.4 to obtain the comprehensive access cost. The accessibility factor of the grid cell is obtained by subtracting the comprehensive access cost from 1. The concealment access factor is obtained by subtracting the visibility factor from 1. The concealment access factor and the accessibility factor are weighted with a weight of 0.5 and normalized to the interval between 0 and 1 to obtain the intrusion threat level of the grid cell.

[0009] In some embodiments, a multi-objective optimization function is constructed with threat-weighted effective coverage, node residual energy balance, and network connectivity reliability as sub-objectives, including: The threat-weighted effective coverage rate is the sum of the comprehensive perception probabilities of each grid cell, weighted by their intrusion threat levels, and then divided by the sum of the intrusion threat levels of all grid cells, resulting in a value normalized to the range of 0 to 1. If the sum of the intrusion threat levels of all grid cells is 0, the threat-weighted effective coverage rate is 0. The node remaining energy balance is determined as follows: the predicted energy consumption of a node within a set duty period is the sum of its perception energy consumption, communication energy consumption, and standby energy consumption. Perception energy consumption is the product of the node's perception power and duty duration. Communication energy consumption is determined along the connected support tree according to the communication distance between the node and its parent node and increases with increasing communication distance. Standby energy consumption is the product of the node's standby power and duty duration. The remaining energy after the node's duty period is obtained by subtracting the predicted energy consumption from the initial energy, and its proportion of the initial energy is calculated. The remaining energy is then calculated by subtracting the total energy consumption from the initial energy. The standard deviation of the remaining energy ratio of a node is used as the node's remaining energy balance. First, the standard deviation of the remaining energy ratio of a node is calculated, and then normalized according to the preset maximum standard deviation. The inverse of the normalized value is used as the node's remaining energy balance. The network connectivity reliability is determined in the following way: the single-hop link reliability is determined based on the distance between adjacent nodes and the terrain occlusion. When the distance between adjacent nodes is less than 50 meters (the reliable communication distance), the single-hop link reliability is 1; when it is greater than 100 meters (the maximum communication distance), it is 0; when it is in between, it decreases linearly with the increase of distance. When there is terrain occlusion in the connection between adjacent nodes, it is reduced according to the proportion of the occlusion segment. The product of the reliability of all single-hop links on the path from each node to the sink node is calculated along the connectivity support tree as the end-to-end connectivity reliability of that node. The arithmetic mean of the end-to-end connectivity reliability of all nodes is taken as the network connectivity reliability. The sub-objectives, which are all normalized to the range of 0 to 1 and whose larger values ​​are preferred, are multiplied by their corresponding weight coefficients and then summed to obtain the comprehensive optimization objective function. The deployment optimization direction is to obtain the maximum value of the comprehensive optimization objective function. The weight coefficients are dynamically set according to the stage of the alert task.

[0010] In some embodiments, the extreme points of perimeter curvature and the highest points are extracted as candidate backbone locations, and the marginal gain is evaluated using an optimization function under connectivity constraints, including: The perimeter centerline is discretized into an ordered sequence of boundary points at 5-meter intervals. For each boundary point, five adjacent points before and after it are taken, and the directional deflection angle of the line connecting the adjacent points before and after it is calculated. Boundary points with a deflection angle greater than 30 degrees are marked as geometric curvature extreme points. Within the warning buffer zone on both sides of the perimeter, a sliding window with a radius of 20 meters is used to traverse the grid grid by grid. Grids with an elevation higher than all other grids in the window and with an elevation difference greater than 2 meters from the lowest point of the window edge are marked as terrain high points. The extreme points of geometric curvature are merged with the high points of the terrain. For two candidate locations with a distance of less than 10 meters, the one with the higher elevation is retained. After deduplication, a set of candidate backbone locations is formed. The initial skeleton node set contains only the sink node. A greedy selection is performed round by round. For each candidate backbone location, a node is assumed to be placed and the marginal gain of the three sub-objectives of the multi-objective optimization function after its access is calculated. The coverage gain is the difference between the threat-weighted effective coverage before and after the node is placed, which is the intrusion threat-weighted increment of the newly added grid that reaches the coverage determination threshold. The connectivity gain is the end-to-end connectivity of the entire network after the candidate location is connected to the support tree with a reliable link. The improvement of the reliability arithmetic mean is the energy consumption balancing gain, which is the reduction of the standard deviation of the proportion of remaining energy of all nodes after access. The three marginal gains are normalized and then weighted and summed according to the weight coefficient of the current warning task stage to obtain the comprehensive marginal gain. Only when the distance between the candidate position and the selected skeleton node is less than 100 meters of the maximum communication distance and the single-hop link reliability is higher than 0.5, the candidate position can participate in the selection. In each round, the candidate position with the largest comprehensive marginal gain is selected and added to the skeleton node set, and the selected skeleton node with the largest end-to-end connectivity reliability is used as the parent node to access the connectivity support tree. The greedy selection process continues until the comprehensive perception probability of all high-threat grids with an intrusion threat level higher than 0.6 reaches the coverage judgment threshold of 0.7, or all candidate backbone positions are selected, or the number of skeleton nodes reaches the set upper limit, thus obtaining the skeleton node set and the connected support tree.

[0011] In some embodiments, the grid with the highest coverage gain is preferentially selected, and a skeleton node is greedily selected and a connected support tree is constructed, including: All grid cells are sorted from highest to lowest intrusion threat level. Grid cells with an intrusion threat level higher than 0.6 are marked as high-threat grid cells and form a priority coverage queue. In each round of greedy selection, the number of newly covered high-threat grid cells after assuming node placement at each candidate backbone location is counted, and the sum of their intrusion threat levels is used as the high-threat coverage gain. The candidate location with the largest high-threat coverage gain is selected first. When two or more candidate locations have the same high-threat coverage gain, their connectivity gain and energy balance gain are compared sequentially, and the best one is selected. After each backbone node is selected, grid cells whose comprehensive sensing probability has reached the coverage determination threshold of 0.7 are removed from the priority coverage queue. In subsequent marginal gain calculations, the candidate nodes are removed, and based on the updated grid coverage status and the current connected support tree structure, the marginal gain of all remaining candidate backbone locations is recalculated. When a candidate location is connected to the support tree, if it meets the communication distance and link reliability constraints with two or more selected backbone nodes, the node with the highest end-to-end connectivity reliability is used as the parent node to establish the primary link, and the remaining links that meet the constraints are recorded as backup links. When the single-hop link reliability of the primary link is lower than 0.5, the system switches to the backup link with the highest single-hop reliability. This process is repeated until all high-threat grids in the priority coverage queue reach the coverage determination threshold.

[0012] Please see Figure 3 , Figure 3 This image displays the intrusion threat distribution of each grid cell within a simulated camp scene. The distribution is represented by the color of the grid cells and vertical background bands: the cell colors indicate four levels of intrusion threat; green cells represent low-threat grids (approximately 0 to 0.35), distributed on the northern cliff and eastern open grassland; pink cells represent low-to-medium threat grids (approximately 0.35 to 0.50), mainly distributed in the western gully; blue cells represent medium-to-high threat grids (approximately 0.50 to 0.70), distributed on the southern edge of the shrub slope and at the high-threat end of the gully; and red cells represent high-threat grids (approximately 0.70 to 1.0), concentrated in the core section of the southern shrub slope. The light green, light yellow, light pink, and light cyan vertical background bands only indicate the four terrain zones: the northern cliff, the eastern open grassland, the southern shrub slope, and the western gully. The grid threat level is determined solely by the cell color and height.

[0013] Figure 3 The threshold value for medium-high threat is 0.6. Grids that cross this line constitute the priority coverage queue in step S103. The lower limit for weak point judgment is 0.5, which serves as the boundary for coverage verification and compensation in step S105. The column colors progress from green to pink to blue to red, corresponding to threat levels from low to high. This visually shows that high-threat grids are concentrated on the gentle shrub slope on the south side, thereby guiding limited node resources to automatically tilt towards this section. This confirms the effectiveness of this application in quantifying threat levels and implementing threat-weighted deployment by integrating visibility and accessibility.

[0014] In some embodiments, based on the skeleton nodes, each node to be deployed jointly encodes its location, type, and orientation angle based on neighborhood local information exchange, using three sub-objectives of a multi-objective optimization function as the fitness evaluation criteria, including: The total number of available nodes is the preset value of the nodes carried before deployment. 10% to 20% of the available nodes are reserved as spare nodes and rounded down. The remainder are planned nodes. After the skeleton node set is determined, the number of filling nodes is equal to the number of planned nodes minus the number of skeleton nodes. Each individual in the population corresponds to a chromosome of fixed length. The total number of chromosome gene loci is equal to the number of planned nodes. Spare nodes do not enter the population or chromosome. The position, type, and orientation of the gene loci corresponding to the skeleton nodes remain fixed during initialization and the entire iterative optimization process and do not participate in crossover or mutation. Only the gene loci corresponding to the filling nodes participate in evolution. The supplementary compensation nodes are post-processing nodes after the compromise scheme is determined. They do not enter the population or chromosome and do not participate in crossover or mutation. Each filled node corresponds to a gene locus, and each gene locus includes three segments: location gene, type gene, and orientation gene. The location gene is represented by the raster cell number, and its value range is the deployable raster after excluding water bodies and slopes greater than 30 degrees. The type gene is represented by the node type number, and its value range is the set of node types allowed by the raster surface classification corresponding to that location gene. Among them, vibration-type nodes are excluded for soft soil raster, and infrared-type nodes are excluded for dense vegetation raster. The orientation gene is discretized into 24 orientation levels from 0 to 345 degrees at 15-degree intervals. An additional omnidirectional identifier level is set. When the type gene is an omnidirectional sensing node type, the orientation gene takes the omnidirectional identifier level value. When the type gene is a directional sensor type, the orientation gene takes one of 24 orientation levels. An initial population is generated according to a set number. The skeleton gene of each individual in the population is fixed. The position gene of the filling node is randomly selected from the deployable grid. The type gene is randomly selected from the set of allowed types in the grid. The orientation gene of the type gene is directional and is randomly selected from 24 levels. The orientation gene of the type gene is omnidirectional and directly takes the omnidirectional identifier level value. In crossover operations, if the orientation gene of the corresponding gene position in the parent generation has an omnidirectional identifier value, the offspring directly inherit the omnidirectional identifier value without orientation crossover. In mutation operations, orientation genes with omnidirectional identifier values ​​do not participate in orientation mutation, and type gene mutation causes the orientation gene to be linked when a node switches between omnidirectional and directional types. Orientation genes mutated to omnidirectional types are forced to take the omnidirectional identifier value, while orientation genes mutated to directional types randomly select one of the 24 identifiers. During the optimization process, each filling node only communicates with its counterparts within a 100-meter radius. Neighboring nodes exchange local information, including the node's current gene, the comprehensive perception probability of the covered grid, and the three local sub-target components. The boundary grids with overlapping coverage areas are summarized by the nearest skeleton node. Each filling node combines the local information obtained from the neighborhood exchange to form the individual's fitness evaluation data, without needing to upload global information to the central node. For each individual, the threat-weighted effective coverage rate, node remaining energy balance, and network connectivity reliability are calculated separately. The three sub-target values ​​together constitute the individual's fitness vector.

[0015] In some embodiments, a multi-objective evolutionary algorithm is used iteratively to obtain a set of Pareto optimal deployment schemes through non-dominated sorting and crowding, including: The selection, crossover and mutation operations are performed sequentially on individuals in the population. The selection operation adopts a tournament-style approach, with priority given to non-dominated hierarchies and those with greater crowding distances within the same hierarchical level. The crossover operation first performs single-point crossover on the position genes according to the raster number, and then determines the type genes. That is, it calculates the intersection of the set of type gene values ​​of the two parents at the same gene locus with the set of allowed surface classification types of the raster of the gene locus after the crossover of the offspring. If the intersection contains two valid values, a discrete uniform random selection is made within it. If the intersection contains only one valid value, the offspring directly inherits the valid value. If the intersection is empty, the offspring discretely and uniformly randomly selects a type again within the set of allowed types of the raster of the gene locus. After the type gene of the offspring gene locus is determined, the orientation gene is then determined. If the type gene is of the omnidirectional sensing node type, the orientation gene is directly taken as the omnidirectional identifier gear value. If the type gene is of the orientation sensor type and both parents have orientation gear values ​​for this gene locus, the orientation gene is taken as the middle gear of the orientation gears of the two parents along the shortest angular distance direction. If the type gene is of the orientation sensor type but there is an omnidirectional identifier gear value in the orientation gene of this gene locus of the parents, the orientation gene is randomly selected from 24 orientation gears, and the crossover probability is 0.8. In mutation operations, the position gene mutates to an adjacent reachable grid with a probability of 0.1, the type gene mutates to another type in the set of allowed types for the position grid with a probability of 0.1, and the orientation gene only shifts randomly by one position with a probability of 0.1 when the type gene is a directional sensor type. The omnidirectional identifier position value does not participate in orientation mutation, and the orientation gene is linked when the type gene mutation causes the node to switch between omnidirectional and directional types. The orientation gene mutated to omnidirectional is forced to take the omnidirectional identifier position value, and the orientation gene mutated to directional type randomly selects one position from 24 orientation positions. If, after crossover or mutation, the position falls into an undeployable grid or the type does not match the surface classification, it is repaired to the nearest legal value. Non-dominated sorting is performed as follows: for any two individuals, if the values ​​of all three sub-targets of one individual are not inferior to those of the other individual and at least one sub-target is superior to that of the other individual, then the former dominates the latter, and the individuals in the current population that are not dominated by any other individual are classified as the first non-dominated layer. After removing the layer, repeat the above division for the remaining individuals to obtain each non-dominated layer until all individuals are layered. Within the same non-dominated layer, calculate the crowding distance. Sort the individuals within the layer according to the three sub-target values. For each sub-target dimension, the crowding distance of the first and last individuals in the sort is the maximum value of that layer. For the remaining individuals in that dimension, the crowding component is the difference between the sub-target values ​​of their two adjacent individuals divided by the difference between the maximum and minimum values ​​within that sub-target layer. Sum the three dimensional components to obtain the individual crowding distance. In each generation, merge the parent and offspring populations and perform a unified non-dominated sort. Select individuals equal to the population size in order of increasing non-dominated level and decreasing crowding distance within the same level to retain them for the next generation. Stop iteration when the first non-dominated layer no longer changes after 100 generations or 10 consecutive generations. Output all individuals in the first non-dominated layer of the final population as a Pareto optimal deployment scheme set, with each individual corresponding to a deployment scheme containing the positions, types, and orientations of all filling nodes.

[0016] In some embodiments, the sub-objectives of each scheme in the scheme set are weighted and summed using the current stage dynamic weights, and the compromise deployment scheme with the largest comprehensive objective value is selected. The threat-weighted perception probability and network connectivity of each grid are verified, and compensation nodes are added to areas below a threshold. A deployment list containing node location, type, and orientation is output, including: All nodes carried before deployment are available nodes. 10% to 20% of the available nodes are reserved as spare nodes and rounded down. The rest are planned nodes. The planned nodes are determined as skeleton nodes and fill nodes through skeleton selection and evolutionary optimization. Spare nodes do not participate in skeleton selection and evolutionary coding. Spare nodes are composed of directional sensing nodes and omnidirectional relay nodes and are only used for compensation and supplementation after verification. Their initial number is equal to the total number of available nodes minus the number of planned nodes. For each scheme in the Pareto optimal deployment scheme set, its three sub-objective values ​​are multiplied by the weight coefficient corresponding to the current task stage and then summed to obtain the comprehensive optimization objective value of the scheme. The scheme with the largest comprehensive optimization objective value is selected as the compromise deployment scheme. When the comprehensive optimization objective values ​​are the same, the one with the higher threat-weighted effective coverage is selected. For the compromise deployment scheme, the product of the comprehensive perception probability and the intrusion threat level is calculated grid-by-grid as the threat-weighted perception probability. Grids with an intrusion threat level higher than 0.5 and a threat-weighted perception probability lower than 0.5 are marked as weak grids. Grids with an intrusion threat level not higher than 0.5 are not included in the weak grid determination. Adjacent weak grids are connected to form a compensation area. At the same time, the end-to-end connectivity reliability is calculated node-by-node along the deployment topology. Nodes with an end-to-end connectivity reliability lower than 0.6 or without adjacent nodes within 100 meters are marked as weak nodes. For each compensation area, compensation locations are selected from the grid with the highest intrusion threat level and its adjacent deployable grids. The type of compensation node is determined according to the surface classification adaptation of that grid. The orientation of the compensation node is aligned with the center of the compensation area. For nodes with weak connectivity, omnidirectional relay nodes can be added to the grid near the midpoint of the link between the node and the nearest skeleton node. After each compensation node is added, the coverage and connectivity verification is re-executed and the weak area is updated. The process is iteratively added until the threat-weighted perception probability of all grids to be verified is not less than 0.5, the end-to-end connectivity reliability of all nodes is not less than 0.6, or the backup nodes are exhausted. Areas that still do not meet the standards when the backup nodes are exhausted are marked as manual patrol areas. Finally, the deployment list containing the deployment coordinates, node type, orientation angle, access parent node, and deployment order of each node is output to the portable terminal in the order of skeleton nodes first, then filling nodes, and finally compensation nodes.

[0017] One embodiment of the above application has the following advantages or beneficial effects: To address the technical problems of existing deployment methods that employ an ideal perception disk model, ignore terrain occlusion and distance attenuation, and apply equal-weighted coverage to all perimeter locations, resulting in a mismatch between planned coverage and actual terrain, and missed defenses on high-threat concealed road sections, this application addresses these issues. Step one involves collecting terrain elevation and surface classification data, establishing a probabilistic perception model that considers terrain occlusion and distance attenuation, and quantifying intrusion threat levels based on grid visibility and surface accessibility. This is coordinated with Step two's threat-weighted effective coverage target, Step three's greedy selection of high-threat grid coverage priority skeletons, and Step five's threat-weighted perception probability verification. This ensures that node coverage evaluation is based on real terrain perception conditions, automatically prioritizing a limited number of node resources towards high-threat road sections that are easily accessible by concealed personnel. The planning results are consistent with the actual effective coverage on-site, thereby eliminating the actual perception blind spots caused by ideal model assumptions and terrain-averaged coverage. To address the technical problems of existing multi-objective deployments that rely on a central computing unit, have fixed target weights, and exhibit fragmented coverage optimization and connectivity assurance, while also assuming isomorphic nodes, optimizing only node coordinates, and lacking a compensation loop after outputting the solution, this application addresses these issues through the following steps: Step 2 sets dynamic weight coefficients according to the rapid containment and long-term maintenance phases; Step 3 uses a greedy approach with marginal gains to construct a connectivity support tree; Step 4, under skeleton constraints, uses neighborhood local information exchange to jointly encode node positions, types, and orientation angles and performs distributed evolutionary optimization; and Step 5 selects a compromise solution based on the current stage weights and adds feature combinations of compensating nodes to weak areas. This allows the three conflicting objectives of coverage, connectivity, and energy consumption to be solved collaboratively within the same optimization framework. Deployment planning can be completed quickly without a central computing unit, and deployment emphasis can be adjusted according to the task phase. Simultaneously, it adapts heterogeneous node types and orientations to surface conditions, eliminates residual blind spots through verification and compensation loops, and finally outputs an integrated deployment list of positions, types, and orientations that can directly guide on-site installation.

[0018] Other effects of the above-mentioned alternative methods will be described below in conjunction with specific embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this application. Wherein: Figure 1 This is a schematic diagram of a multi-target node optimization deployment method for portable perimeter surveillance provided in an embodiment of this application; Figure 2 This is a comparative schematic diagram of the perception probability models provided in the embodiments of this application; Figure 3 This is a schematic diagram of the perimeter grid intrusion threat distribution provided in an embodiment of this application; Figure 4 This is a schematic diagram of the three-dimensional distribution of the Pareto front provided in the embodiments of this application; Figure 5 This is a schematic diagram of the convergence curve of the evolutionary algorithm provided in the embodiments of this application. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] To facilitate understanding of this application, the embodiments of this application will be briefly described below: In areas lacking fixed security facilities, such as temporary campsites, checkpoints, temporary security zones for major events, or disaster relief sites, personnel need to rapidly deploy a batch of battery-powered heterogeneous sensor nodes along the perimeter within a short period after arriving on site. These nodes include vibration nodes, acoustic nodes, infrared nodes, and small radar nodes. Each node self-organizes into a network to detect and alert personnel who illegally cross or infiltrate the perimeter, and must maintain continuous monitoring for a considerable period after the perimeter is established. For example, when a detachment temporarily camps at the edge of a mountainous jungle, it needs to deploy nodes along the camp's perimeter to form a control line: the north side of the camp is a steep cliff, making it difficult for personnel to pass; the east side is an open grassland, making it easy for personnel to be spotted when approaching; the south side is a gentle slope covered with shrubs, with multiple concealed access paths; and the west side is a seasonal gully. The deployment team first needs to acquire the terrain around the camp and divide the perimeter into planning units to determine which locations are easy for personnel to approach covertly. Then, under the conditions of limited number of nodes, battery power and communication distance, they need to determine where each node should be set up, what type of node to use and the detection orientation of the directional nodes, so that high-threat sections such as the gentle shrub slope on the south side can be covered first. At the same time, they need to ensure that the communication link from each node to the aggregation node is not blocked by the terrain. After the initial deployment is completed, nodes are added to the remaining blind spots.

[0022] Existing node deployment methods calculate coverage on a planar topographic map using a geometrically distributed, equally spaced sensing disk, without considering the signal attenuation caused by shrubs, gentle slopes, and undulating terrain. While the planned map shows covered sections, actual on-site blind spots exist. Furthermore, the equal-weighted coverage of all perimeter locations results in a limited number of nodes being distributed across low-threat areas like open grasslands on the east side, while high-threat shrub slopes are missed. Utilizing a central computing device for global multi-objective optimization is time-consuming and relies on computing power often unavailable on-site. The use of fixed weights during rapid containment and long-term monitoring phases makes it difficult to balance coverage priorities during containment and power balance during nighttime monitoring. Additionally, the method often assumes homogeneous nodes and only optimizes node coordinates, neglecting to plan for heterogeneous node types and orientation. The output also lacks verification and compensation for actual coverage and communication connectivity, making it unsuitable for the rapid, reliable, and long-term deployment requirements of portable perimeter surveillance.

[0023] See Figure 1 This is a flowchart illustrating a multi-target node optimization deployment method for portable perimeter surveillance provided in an embodiment of this application. Figure 1 The execution subject of the method shown can be a combination of software and / or hardware, specifically, it can be one or more of various types of terminals, hardware systems, cloud computing, etc.

[0024] Figure 1The proposed multi-target node optimization deployment method for portable perimeter surveillance includes steps S101 to S105, as detailed below: S101: Collect terrain elevation and surface classification data of the deployment area, divide the warning perimeter into several grid cells, establish a probability perception model for each node to be deployed that considers terrain occlusion and distance attenuation, and calculate the intrusion threat level of each grid cell based on the visibility from the grid cell to the outside of the perimeter and the surface accessibility.

[0025] In step S101, the warning perimeter is divided into several grid cells, and a probabilistic perception model considering terrain occlusion and distance attenuation is established for each node to be deployed. The specific steps are as follows: Furthermore, the method used to obtain the model's input parameters is as follows: The coordinates of each node to be deployed are selected on an electronic topographic map using a portable planning terminal or obtained through actual measurement using a satellite positioning module. The planar coordinates and terrain elevation data use the same coordinate system. The sensor mounting height is a preset value entered into the parameter table according to the node type before deployment. The mounting height for ground-mounted vibration and acoustic nodes is 0.1 to 0.3 meters, while the mounting height for pole-mounted radar and photoelectric nodes is 1.5 to 2 meters. The effective sensing radius is also a preset value, first taken from the node's factory-calibrated rated sensing radius, then reduced by a preset correction factor based on the surface type of the deployment area. Each type of node corresponds to a specific effective sensing radius. The center coordinates of the grid cell are directly determined by the grid division results, and the grid ground elevation is taken as the arithmetic mean of all elevation sampling points within the grid. The horizontal distance from the node to the center of the grid cell is equal to the square root of the sum of the squares of the differences between the horizontal and vertical coordinates of the two points. Furthermore, the method used to determine visibility based on terrain obstruction is as follows: Along the line connecting the node and the center of the grid cell, elevation profiles are extracted point by point at sampling intervals no greater than 1 meter. The actual ground elevation of each sampling point is obtained by bilinear interpolation of the terrain elevation data. The 1-meter sampling interval is a preset value, determined by the elevation data sampling interval and the minimum width of the obstructing features to be identified, ensuring that terrain obstructions wider than 1 meter are not missed. The elevation of the starting point of the line is equal to the sum of the ground elevation at the node and the sensor installation height, and the elevation of the ending point of the line is equal to the ground elevation of the center of the grid cell. The line-of-sight elevation of any sampling point on the line is obtained by linear interpolation based on the position of the sampling point between the starting and ending points of the line. The actual ground elevation of each sampling point is compared with its line-of-sight elevation one by one. If the actual ground elevation of any sampling point is higher than the line-of-sight elevation, the line is determined to be obstructed by terrain, and the perception probability of that node for that grid cell is set to 0. Only when the actual ground elevation of all sampling points is not higher than the line-of-sight elevation is it determined to be a line-of-sight, and distance attenuation calculation begins.

[0026] Furthermore, the method used to determine the distance attenuation probability is as follows: For line-of-sight connections, the detection probability decreases exponentially with horizontal distance. This decrease is uniquely determined by two boundary conditions: the detection probability is 1 when the horizontal distance is 0, and 0.5 when the horizontal distance equals the effective sensing radius. Therefore, the decay constant is 0.693 divided by the effective sensing radius. Those skilled in the art can determine the detection probability at each distance based on this. The truncation rule is as follows: when the horizontal distance is greater than twice the effective sensing radius, or when the detection probability calculated according to the exponential relationship is lower than the 0.05 probability threshold, the detection probability is set to 0. The first of the two conditions is taken. The 0.05 probability threshold is a preset value, meaning that a single node detection probability below 5% is considered unreliable detection and is no longer counted.

[0027] Furthermore, the method used to fuse the probabilities of multi-node integrated perception is as follows: The probability of a single node missing a grid cell is equal to 1 minus the node's detection probability. The combined detection probability of all nodes covering the same grid cell is equal to 1 minus the product of the missing detection probabilities of each node, which is equivalent to the joint detection probability under the condition that each node detects independently. The calculation order is as follows: first, calculate the detection probability of each node, then calculate the missing detection probability of each node and multiply them together, and finally take the complement to obtain the combined detection probability of the grid cell. The result always falls within the interval of 0 to 1, which is used for subsequent threat-weighted coverage calculation.

[0028] Please see Figure 2 , Figure 2 This paper compares the changes of three perception probability models with horizontal distance. The traditional ideal disk model has a constant perception probability of 1 within the effective perception radius R=60m, which drops sharply to 0 beyond that radius. This is a highly simplified assumption that does not consider the signal attenuation with distance or the influence of terrain occlusion. The probability perception model proposed in this embodiment (without line-of-sight occlusion) decays the perception probability exponentially. The decay constant is uniquely determined by two boundary conditions: the probability is 1 when the distance is 0 and 0.5 when the distance is equal to the effective perception radius. When the probability is below the 0.05 threshold or the distance exceeds twice the effective perception radius, it is truncated to 0. Considering the actual perception probability after terrain occlusion, it can be clearly seen that in the 20-35m range, the perception probability is significantly reduced to about 30% of the ideal value due to terrain occlusion. This comparison clearly shows that the traditional ideal disk model will seriously overestimate the actual coverage effect under complex terrain conditions, while the probability perception model of this embodiment can accurately reflect the combined effects of terrain occlusion and distance attenuation, so that the planned coverage is consistent with the actual coverage on site.

[0029] In step S101, the intrusion threat level based on visibility and surface accessibility is calculated. The specific steps are as follows: Furthermore, the visibility factor is first calculated using the following method: First, determine the outer normal of the perimeter: the tangent of the perimeter centerline at the grid cell follows the centerline, and the outer normal is perpendicular to the tangent and points outward from the perimeter. Starting from the grid cell and centering on the outer normal, three lines of sight are laid out every 15 degrees to both sides of the normal, forming a total of seven lines of sight. The 15-degree angular interval and the number of seven lines of sight are preset values, chosen to balance angular resolution with the computational load of the portable terminal. Each line of sight is assessed point-by-point at 1-meter intervals within a 50-meter warning depth from the perimeter. The 50-meter warning depth is a preset value, chosen based on the minimum warning distance required for personnel to approach the boundary on foot from the outside of the perimeter; the assessment method is the same as the elevation profile method in Example 1. For each line of sight, the unobstructed segment extends from the grid cell to the first obstruction point. The length of the unobstructed segment is divided by the total length of the line of sight within the warning depth to obtain the visibility ratio of that line of sight. When the entire line of sight is unobstructed, the visibility ratio is taken as 1. The arithmetic mean of the visibility ratios of all seven lines of sight is taken as the visibility factor. The larger the value, the easier it is to observe people approaching from the outside from the inside of the vigilance, and the lower the intrusion threat at the corresponding location.

[0030] Furthermore, the method used to calculate the cost of passage due to gradient is as follows: The slope value is calculated from the elevation gradient of the terrain elevation data: taking the eight surrounding grid cells as the center to form a nine-square grid neighborhood, the elevation change rate is calculated along both the horizontal and vertical directions. The slope angle is equal to the square root of the sum of the squares of the horizontal and vertical change rates, and then taking the arctangent. The slope passage cost is segmented according to two preset thresholds based on the walking ability of personnel: when the slope is less than 15 degrees, personnel can walk normally and quickly, and the slope passage cost is 0; when the slope is greater than 30 degrees, personnel need to climb and it is difficult to pass quickly, and the slope passage cost is 1; when the slope is between 15 degrees and 30 degrees, it is determined by linear interpolation. Both the 15-degree and 30-degree thresholds are preset values ​​pre-calibrated based on human walking ability.

[0031] Specifically, the method used for the pre-establishment and lookup of the surface access cost mapping table is as follows: The surface access cost is obtained from a pre-established surface access cost mapping table stored in a portable planning terminal. This mapping table is calibrated offline before deployment. The establishment process is as follows: for each standard surface type, the average travel time per unit distance is determined through personnel walking tests. The average travel time for hardened and leveled surfaces is used as the baseline time, and the access cost for surfaces that are completely inaccessible to personnel on foot is set to 1. The increment of the travel time for each type of surface relative to the baseline time is divided by the maximum increment among all categories and normalized to the interval between 0 and 1 to obtain the access cost for each type of surface and form a mapping table of "surface type to access cost". This embodiment provides a set of calibration examples: hard surface is set to 0.1, bare soil to 0.2, grassland to 0.4, shrubland to 0.6, woodland to 0.8, and water body to 1. In actual deployment, the values ​​are retrieved by looking up the table grid by grid from the surface classification data; for surface types not directly included in the mapping table, the cost of the labeled type with the closest passage characteristics is taken according to the difficulty of walking by people, so as to ensure that each grid can obtain a unique and determined surface passage cost.

[0032] Furthermore, the method used to fuse reachability factors and intrusion threat levels is as follows: The comprehensive passage cost is obtained by weighting the slope passage cost with a weight of 0.6 and the surface type passage cost with a weight of 0.4. The weights 0.6 and 0.4 are preset values, chosen based on the assumption that slope has a more direct impact on short-term, rapid accessibility for personnel. These weights can be adjusted before deployment based on the terrain of the deployment area. The accessibility factor is obtained by subtracting the comprehensive passage cost from 1; the easier the surface is for personnel to pass, the larger the accessibility factor. The concealment access factor is obtained by subtracting the visibility factor from 1; the more severe the terrain obstruction and the easier it is to conceal access, the larger the concealment access factor. The concealment proximity factor and the accessibility factor are each weighted by 0.5 and summed. 0.5 and 0.5 are preset equal weights, indicating that "whether it can be approached concealably" and "whether it can pass smoothly" are equally important to the intrusion threat. The weighted result itself falls in the range of 0 to 1. Then, the maximum and minimum values ​​of the weighted results of all grids in the deployment area are used for extreme value normalization. Finally, the intrusion threat level of each grid distributed in the range of 0 to 1 is obtained. The larger the value, the higher the threat of the grid being invaded. This value is used for threat weighting effective coverage and priority deployment of high-threat grids.

[0033] In this embodiment, step S101 collects terrain elevation and surface classification data of the deployment area and rasterizes the warning perimeter. On the one hand, it uses elevation profile visibility judgment to characterize terrain occlusion, uses the exponential law uniquely determined by double boundary conditions to characterize distance attenuation, and multiplies the missed detection probabilities of each node to achieve multi-node joint perception. This ensures that the comprehensive perception probability of each grid is based on the actual visibility conditions and actual detection distance of the node, overcoming the defect of mismatch between planned coverage and actual coverage in complex terrain under the ideal perception disk model, and eliminating the perception blind spots caused by terrain occlusion from the data source. On the other hand, it quantifies the concealed access conditions outside the perimeter by the visibility ratio of the 7 lines of sight, and quantifies and merges the personnel accessibility conditions by the slope access cost and the surface access cost. To normalize the intrusion threat level, the difficulty of intrusion into each perimeter grid is objectively distinguished, changing the existing deployment method that treats all perimeter locations equally and distributes limited nodes to low-threat areas. Both types of quantification results share the same set of grid cells and terrain data, providing a unified, reproducible, and rapidly computed data foundation on portable planning terminals for constructing the threat-weighted effective coverage rate in subsequent multi-objective optimization functions, prioritizing the coverage of high-threat grids by skeleton nodes, and verifying and compensating for the threat-weighted perception probability after deployment. This allows node resources to be concentrated on high-threat road sections that are severely obscured by terrain and easily accessible to personnel, thereby improving the overall matching degree between the final deployment scheme and the actual terrain conditions and actual threat distribution.

[0034] S102: Construct a multi-objective optimization function with three sub-objectives: threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability. The threat-weighted effective coverage is the weighted sum of the comprehensive perception probability of each grid with its intrusion threat level as the weight, and dynamic weight coefficients are set according to the stage of the alert task.

[0035] In step S102, a target vector is constructed and a multi-objective optimization function is synthesized using threat-weighted effective coverage, node residual energy balance, and network connectivity reliability as three sub-objectives. The specific steps are as follows: Furthermore, the method used to calculate the threat-weighted effective coverage is as follows: Using the comprehensive perception probability of each grid cell obtained in step S101 as the numerator and the intrusion threat level of each grid cell as the weight, the product of the comprehensive perception probability and the intrusion threat level is calculated for each grid cell, and the sum is applied to all grid cells. This product is then divided by the sum of the intrusion threat levels of all grid cells. The resulting value is the threat-weighted effective coverage rate. Since both the comprehensive perception probability and the intrusion threat level fall within the range of 0 to 1, this coverage rate is also normalized to the range of 0 to 1. Its physical meaning is that when a grid cell with a higher threat level is reliably covered, it contributes more to the target value, thereby guiding optimization towards high-threat areas. A divide-by-zero fallback rule is set: when the sum of the intrusion threat levels of all grid cells is 0, that is, in the extreme case where the entire perimeter is an inaccessible area with no possibility of intrusion, the threat-weighted effective coverage rate is directly set to 0. This rule ensures that the target value can be uniquely determined under any input and that no divide-by-zero anomaly occurs.

[0036] Furthermore, the method used to calculate the residual energy balance at the nodes is as follows: First, predict the energy consumption of each node within the set monitoring period. The predicted energy consumption of a node is the sum of its sensing energy consumption, communication energy consumption, and standby energy consumption. The monitoring period is the continuous monitoring duration preset before deployment based on the alert task. Sensing energy consumption equals the product of the node's sensing power and the monitoring duration, where the sensing power is the preset rated value corresponding to the node model. Standby energy consumption equals the product of the node's standby power and the monitoring duration, where the standby power is also the preset value for the node model. Communication energy consumption is determined along the connected support tree according to the communication distance between the node and its parent node, specifically the product of the transmission power and the cumulative communication duration. The transmission power is based on the minimum transmission power at a 1-meter reference distance and increases according to the square of the communication distance; that is, for every doubling of the communication distance relative to the reference distance, the transmission power increases to four times the original value. The cumulative communication duration equals the monitoring duration divided by the preset reporting period, the product of the number of reports, and the duration of a single transmission. Therefore, communication energy consumption increases monotonically with the increase of the communication distance between parent and child nodes.

[0037] After the three types of energy consumption prediction are completed, the predicted energy consumption is subtracted from the initial energy of the node battery (a preset value of the battery's rated energy) to obtain the remaining energy after the node is on duty. Then, the remaining energy is divided by the initial energy to obtain the remaining energy ratio of the node. The standard deviation of the remaining energy ratio of all nodes is calculated. That is, the arithmetic mean of the remaining energy ratio of all nodes is first calculated, and then the arithmetic mean of the squares of the differences between the remaining energy ratio of each node and the average value is calculated and the square root is taken. Then, it is normalized according to the preset maximum standard deviation, which is 0.5. The value is based on the theoretical upper limit of the standard deviation when the remaining energy ratio is distributed in the range of 0 to 1 (obtained when half of the nodes have a remaining energy ratio of 0 and half have a remaining energy ratio of 1). The normalized value is obtained by dividing the actual standard deviation by 0.5. Finally, the normalized value is subtracted from 1. The result is the node remaining energy balance, which falls in the range of 0 to 1 and the value is closer to 1 as the remaining energy of each node becomes more consistent.

[0038] Furthermore, the method used to calculate network connectivity reliability is as follows: First, determine the single-hop link reliability between adjacent nodes, which consists of two parts: distance attenuation and terrain obstruction reduction. The distance attenuation is segmented according to two preset distances calibrated by the node radio at the factory: 1 when the reliable communication distance is less than 50 meters and 0 when the maximum communication distance is greater than 100 meters. When the distance is between 50 and 100 meters, it is determined by linear attenuation, that is, the single-hop distance reliability is equal to the difference between 100 meters and the actual distance divided by 50 meters. The terrain obstruction reduction extracts the obstruction segment along the line connecting the two nodes using the same elevation profile method as in step S101. The proportion of the obstruction segment length to the total link length is the obstruction rate. Finally, the single-hop link reliability is equal to the distance reliability multiplied by 1 minus the obstruction rate, so that the more severe the terrain obstruction, the lower the link reliability. Based on this, along the connected support tree, the reliability of all single-hop links along the path from each node to the sink node is multiplied together to obtain the end-to-end connectivity reliability of that node, which physically means the probability that the series link will not be interrupted throughout the entire process; then the arithmetic mean of the end-to-end connectivity reliability of all nodes is calculated to obtain the network connectivity reliability, and the result falls in the interval between 0 and 1.

[0039] Furthermore, the method used to synthesize the comprehensive optimization objective function and determine the optimization direction is as follows: The three sub-objectives—threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability—have all been normalized to the range of 0 to 1 and are uniformly considered to be better the larger the value. The three sub-objectives are multiplied by their respective weight coefficients and then summed to obtain the comprehensive optimization objective function. It should be noted that the weighting coefficients are dynamically set according to the stages of the alert task. Specifically, the alert task is divided into a rapid containment stage and a long-term monitoring stage, with weights assigned to each of the three sub-targets. The rapid containment stage corresponds to the period immediately after node deployment, where perimeter containment needs to be established as quickly as possible. Following the principle of containment first, then long-term maintenance, the weights for threat-weighted effective coverage, network connectivity reliability, and node remaining energy balance are set to 0.7, 0.2, and 0.1 respectively. The long-term monitoring stage corresponds to the period after containment is established, where sustained monitoring is required for an extended period. Following the principle of minimizing energy consumption and maximizing maintenance duration, the weights for the three sub-targets are set to 0.3, 0.2, and 0.5 respectively. The weight for network connectivity reliability in both stages is 0.2, based on the fact that communication connectivity is fundamental for data transmission and intrusion alarm reporting, and cannot be compromised in any stage of the task. The sum of the two weights is always equal to 1.

[0040] Phase switching employs a dual-criteria approach based on time and coverage rate, and incorporates a hysteresis interval to prevent weight fluctuations: Within two hours of node deployment, or if the threat-weighted effective coverage rate has not yet reached 0.85, the system is in the rapid containment phase. When deployment exceeds two hours and the threat-weighted effective coverage rate reaches 0.85, it switches to the long-term standby phase. If the threat-weighted effective coverage rate drops below 0.7 during the long-term standby phase, the rapid containment phase weight is temporarily restored until the coverage rate recovers to 0.85, at which point it switches back to the long-term standby weight. The hysteresis interval between 0.7 and 0.85 prevents frequent weight switching between phases when coverage fluctuates around a single threshold. The 2-hour, 0.85, and 0.7 values ​​are preset values, determined based on the typical operation time for portable teams to complete perimeter containment, the minimum coverage level required for containment formation, and the coverage threshold for containment failure, respectively. These values ​​can be adjusted before deployment according to mission requirements.

[0041] The specific method for invoking the weights is as follows: the portable planning terminal records the deployment duration using a built-in timer and recalculates the current threat-weighted effective coverage rate according to a set period. Based on this, the current task stage is determined, and the corresponding set of weights is read and substituted into the comprehensive optimization objective function. During stage switching, the three weights gradually change from the old value to the new value within a set transition period, rather than abruptly, to avoid abrupt changes in the comprehensive objective value ranking of adjacent candidate schemes. It should be noted that the non-dominated ranking in step S104 directly compares the dominance relationship using the target vector composed of the three sub-objectives, without substituting the weights. The dynamic weights are only substituted into two stages: the weighting of the three marginal gains in step S103 (skeleton greedy selection) and step S105 (selection of a compromise scheme from the Pareto optimal deployment scheme set). This ensures that the Pareto solution set obtained by evolutionary optimization is not affected by task stage preferences. Stage preferences only affect skeleton optimization and final scheme selection, achieving one-time optimization and stage-based solution selection.

[0042] The comprehensive optimization objective function takes the maximization of the objective function as the direction of deployment optimization. That is, it seeks the scheme that maximizes the objective function value among all candidate deployment schemes that meet the constraints. The objective function also serves as the evaluation basis for the marginal gain when greedily selecting skeleton nodes in step S103, the evaluation basis for the individual fitness when evolutionarily optimizing in step S104, and the scoring basis for selecting the compromise scheme in step S105.

[0043] In this embodiment, step S102 unifies the three sub-objectives—threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability—to the 0-1 range, with higher values ​​being better. This integrates the conflicting requirements of coverage quality, persistent monitoring, and communication connectivity in perimeter deployment into a unified metric: Threat-weighted effective coverage inherits the comprehensive perception probability and intrusion threat level from step S101, and uses division by zero to ensure a unique value for any input, continuously tilting coverage optimization towards high-threat grids; Node remaining energy balance predicts the energy consumption of perception, communication, and standby for each node, and uses the inverse of the standard deviation of the remaining energy ratio to characterize the dispersion of energy consumption across the entire network, preventing individual nodes from failing prematurely due to excessive energy consumption, thereby extending the continuous monitoring time of the entire network; Network connectivity reliability incorporates terrain occlusion into single-hop link reduction and multiplies it hop-by-hop along the connectivity support tree, ensuring that coverage optimization and connectivity maintenance are aligned. The simultaneous solution within the same target system overcomes the problem of communication topology breakage caused by the separation of the two, which leads to coverage achievement. On this basis, the weights are dynamically configured according to the two stages of rapid containment and long-term maintenance. The weights are smoothly transitioned by using time and coverage as dual criteria, hysteresis intervals and gradual switching. This allows the same optimization framework to adapt to the target emphasis of different task stages. Moreover, the weights only intervene in the greedy selection of the skeleton and the selection of compromise solutions, without intervening in non-dominated sorting. This ensures that a complete Pareto solution set can be obtained in one evolution and a solution can be quickly selected according to the current stage without repeated optimization. The comprehensive optimization objective function constructed in this way provides a unified evaluation basis for steps S103 to S105. This ensures that the marginal gain evaluation, individual fitness comparison and final solution scoring follow the same evaluation scale. Overall, it ensures that the deployment scheme achieves an optimal balance that can be adjusted according to the task stage among threat matching coverage, energy consumption balance and durability and network connectivity reliability.

[0044] S103: Extract the extreme points of the perimeter geometric curvature and the high points of the terrain as candidate backbone locations. For each candidate location selected, evaluate its marginal gain using a multi-objective optimization function. Under the constraint of communication connectivity, prioritize the location that maximizes the coverage gain of the high-threat grid. Use a greedy strategy to successively select the set of skeleton nodes and construct a connected support tree.

[0045] In step S103, the extreme points of the perimeter geometric curvature and the high points of the terrain are extracted as candidate backbone positions. The marginal gain is evaluated round by round using a multi-objective optimization function. The backbone nodes are greedily selected and a connected support tree is constructed. The specific steps are as follows: Furthermore, the method used to extract the extreme points of the perimeter geometric curvature is as follows: The perimeter centerline is generated by personnel plotting along the warning boundary on an electronic topographic map, or automatically extracted from the centerline of the perimeter buffer zone. The perimeter centerline is discretized into an ordered sequence of boundary points at a preset interval of 5 meters. The 5-meter interval matches the side length of the raster unit, ensuring that the spatial resolution of curvature discrimination is not lower than the raster level. For each ordered boundary point, its five preceding and five following adjacent points along the perimeter direction form a local support window, with the five preceding and following points being preset values. Together with the center point, it covers an arc length of approximately 50 meters to reflect the overall direction of the road segment and suppress single-point sampling noise. The line connecting the first and last points of the preceding window is used as the incoming line segment, and the line connecting the first and last points of the following window is used as the outgoing line segment. The directional deflection angle between the two line segments is calculated. When the directional deflection angle is greater than a preset threshold of 30 degrees, the boundary point is marked as a geometric curvature extreme point. The 30-degree threshold is determined based on the turning point where the deflection angle exceeds this value. Nodes deployed along a single direction cannot simultaneously cover the road segments on both sides of the turning point, so backbone nodes must be deployed.

[0046] Furthermore, the method used to extract the highest points of the terrain is as follows: Within the perimeter's warning buffer zone on both sides, a circular sliding window with a radius of 20 meters is used to traverse the grid cell by cell. The width of the buffer zone is consistent with the 50-meter warning depth in step S101. The 20-meter window radius is a preset value, which is on the same order of magnitude as the effective sensing radius of the node, so that the selected high point has a visual advantage over its surrounding area. When the elevation of a certain grid cell simultaneously meets the two conditions of being higher than all other grid cells in the window and having a height difference of more than 2 meters between it and the lowest point of the window edge, it is marked as a terrain high point. The 2-meter height difference is a preset value, used to exclude false high points caused by normal sampling undulations on flat ground.

[0047] Furthermore, the method used to merge candidate backbone positions and remove duplicates is as follows: The extreme points of geometric curvature are projected onto their respective grids and merged with the terrain high point grids into the same candidate set. For candidate locations in the set with a pairwise distance of less than 10 meters, only the one with the higher elevation is retained. The 10-meter deduplication distance is a preset value and does not exceed half of the minimum effective sensing radius of the node, so as to avoid overlapping coverage and node waste caused by two backbone locations being too close. After deduplication, a set of candidate backbone locations is formed.

[0048] Furthermore, the skeleton set is initialized and three types of marginal gains are defined, based on the following method: The initial skeleton node set contains only the aggregation node, whose location is a preset value, taken as the location of the temporary command post or communication hub, and serves as the root node of the connectivity support tree. Starting from the first round, a trial placement is performed for each candidate backbone location, calculating the marginal gains of the three sub-objectives of the multi-objective optimization function: Coverage gain is the difference between the threat-weighted effective coverage rate after the trial placement and the current value, equivalent to the weighted increment of the intrusion threat level of a newly added grid reaching the 0.7 coverage threshold, where 0.7 is the preset reliable coverage threshold; Connectivity gain is the increase in the arithmetic mean of the end-to-end connectivity reliability of the entire network relative to the current value after the candidate accesses the support tree via a reliable link; Energy balance gain is the difference between the standard deviation of the remaining energy ratio of all nodes before access and the standard deviation after access, i.e., the decrease in standard deviation. In each round, the three types of gains of all admission candidates are normalized by dividing them by the maximum value of that type of gain in this round, ensuring that all three types of gains fall within the range of 0 to 1. Then, the weighted sum is calculated according to the weight coefficient of the current task stage in step S102 to obtain the comprehensive marginal gain of the candidate.

[0049] Furthermore, the method for setting access constraints and selecting access points round by round is as follows: Candidate backbone locations can only participate in this round of selection if the distance between them and a selected backbone node is less than 100 meters (maximum communication distance) and the single-hop link reliability between them is higher than 0.5. 100 meters is the preset maximum communication distance value calibrated by the node radio at the factory, and 0.5 is the preset lower limit of link reliability. Links with a reliability lower than this value have too low end-to-end reliability after multi-hop serialization. In the initial round, the aggregation node serves as the first access anchor point. In each round, the node with the largest comprehensive marginal gain among all admission candidates is added to the backbone node set, and the parent node is determined among all selected backbone nodes that meet the admission constraints. That is, the single-hop link reliability from the candidate to the selected backbone is multiplied by the end-to-end connectivity reliability from the selected backbone to the aggregation node, and the node with the largest product is taken as the parent node. The new backbone node is added to the connectivity support tree as its child node. The node type of the backbone node is determined according to the surface classification of the grid it is located in. The orientation of the directional backbone node is aligned with the geometric centroid of the high-threat grid that it is newly covered in this round. The position, type, and orientation of the backbone node remain fixed in step S104.

[0050] Specifically, the method used to construct a high-threat grid priority coverage queue and perform step-by-step comparisons is as follows: All grid cells are sorted from highest to lowest intrusion threat level. Grid cells with an intrusion threat level higher than a preset threshold of 0.6 are marked as high-threat grid cells and form a priority coverage queue. In each round of greedy selection, the number of newly covered high-threat grid cells after each candidate hypothesis is placed is counted and their intrusion threat levels are summed as the high-threat coverage gain. The candidate position with the largest high-threat coverage gain is selected first. When two or more candidates have the same high-threat coverage gain, their connectivity gain and energy balance gain are compared in turn and the best is selected. The comparison at each level ensures that the selection result is unique in any case and that no parallel deadlock occurs.

[0051] Specifically, the method used to perform round-by-round gain updates and primary / backup link redundancy maintenance is as follows: After each skeleton node is selected, the gratings whose comprehensive sensing probability has reached the coverage determination threshold of 0.7 are removed from the subsequent marginal gain calculation to avoid reliably covered gratings being counted in the gain repeatedly. Based on the updated grating coverage status and the current connected support tree structure, the three types of marginal gains for all remaining candidate skeleton locations are recalculated. When a candidate location is connected to the support tree, if it meets the communication distance and link reliability constraints with two or more selected skeleton nodes, the one with the highest end-to-end connectivity reliability is used as the parent node to establish the primary link. The other links that meet the constraints are only recorded as backup links and do not participate in data forwarding to avoid forming loops. When the single-hop link reliability of the primary link is lower than 0.5, it switches to the backup link with the highest single-hop reliability. It switches back to the primary link after the reliability of the primary link is restored.

[0052] Furthermore, the method used to determine the termination of the iteration and output the result is as follows: Repeat the above greedy selection process round by round until one of the following conditions is met: First, the overall perception probability of all high-threat grids with an intrusion threat level higher than 0.6 in the priority coverage queue reaches the coverage judgment threshold of 0.7; Second, all candidate backbone positions are selected; Third, the number of skeleton nodes reaches the set upper limit, which is preset to 40% of the planned number of nodes, to reserve sufficient nodes for the node filling optimization in step S104; After stopping, output the skeleton node set and the connected support tree, which will be used for the distributed evolution optimization of subsequent node filling based on it.

[0053] In this embodiment, step S103 first uses two types of geometric and terrain priors—perimeter curvature extrema points and terrain high points—to compress the backbone candidate locations from a massive number of grids across the entire region into a small number of key points. Then, it greedily selects points round by round using the marginal gain of a multi-objective function, and simultaneously connects to the connectivity support tree under the constraints of communication distance and link reliability. This ensures that the selected skeleton nodes prioritize suppressing high-threat grids while naturally guaranteeing connectivity, avoiding the problem of communication interruption despite coverage being achieved. In each round, grids that have met the standards are removed, and the remaining candidate gains are recalculated. Primary and backup redundancy is established for multiple link candidates to ensure that the gains are not recalculated and that the skeleton topology has the ability to withstand destruction. The fixed skeleton further reduces the global deployment problem for all nodes to a local optimization problem for filling nodes, laying the foundation for the distributed and rapid solution on portable nodes in step S104. This achieves the deployment effect of first blocking key locations, maintaining the communication skeleton without interruption, and minimizing the scale of subsequent optimization.

[0054] S104: Based on the skeleton nodes, each node to be deployed exchanges local information in its neighborhood, jointly encodes the node position, node type and orientation angle, uses the three sub-objectives of the multi-objective optimization function as the fitness evaluation basis, adopts a multi-objective evolutionary algorithm to iteratively optimize, maintains the diversity of the solution set through non-dominated sorting and crowding distance, and obtains the Pareto optimal deployment scheme set.

[0055] In step S104, with the skeleton nodes as a fixed foundation, each filling node exchanges local information in its neighborhood, jointly encodes its position, type, and orientation, and uses three sub-objectives as fitness. A multi-objective evolutionary algorithm is then used to iteratively obtain a Pareto-optimal deployment scheme set. The specific steps are as follows: Furthermore, the method used to perform node partitioning and construct fixed-length chromosomes is as follows: Before deployment, all nodes are considered usable nodes, and their total number is a preset value. A 10% to 20% reserve of spare nodes is initially reserved and rounded down. This reserve ratio is a preset value determined based on experience in compensating for redundancy. Rounding down ensures that the number of nodes used for planning after reservation does not exceed the actual number of nodes carried. The remaining nodes are planning nodes. After the skeleton node set is determined in step S103, the number of filling nodes equals the number of planning nodes minus the number of skeleton nodes. Each individual in the population corresponds to one chromosome. The total number of chromosome gene loci equals the number of planning nodes, and the length is fixed throughout the evolutionary process. The position, type, and orientation of the gene loci corresponding to the skeleton nodes are written according to the results of step S103 and remain fixed during initialization and all iterations, without participating in crossover or mutation. Only the gene loci corresponding to the filling nodes participate in evolution. Spare nodes and compensation nodes added in step S105 are post-processing resources after the scheme is determined. They do not enter the population or chromosome and do not participate in crossover or mutation, thus ensuring that the crossover operation gene loci on the fixed-length chromosome are always aligned.

[0056] Furthermore, for each filler gene site, three gene segments are assigned: location, type, and orientation. The method used is as follows: Location genes are represented by raster cell numbers, and their value range is the deployable raster after excluding undeployable raster cells. Undeployable raster cells include water bodies and raster cells with a slope greater than 30 degrees. 30 degrees is consistent with the upper limit of slope cost in step S101, corresponding to terrain where personnel cannot stably set up nodes. Type genes are represented by node type numbers, and their value range is the set of node types allowed by the surface classification of the raster corresponding to the location gene. Among them, vibration-type nodes are excluded from soft soil raster cells because vibration signals attenuate too quickly in soft soil. Insufficient distance measurement and dense vegetation grids exclude infrared nodes because infrared beams are easily blocked by branches and leaves, resulting in false alarms and missed alarms. The orientation gene is discrete into 24 orientation levels from 0 to 345 degrees at 15-degree intervals, with an additional omnidirectional marker level. The 15-degree interval and 24 levels are preset values ​​that take into account the orientation resolution capability of the orientation sensor and the encoding length. When the type gene is an omnidirectional sensing node type, the orientation gene takes the omnidirectional marker level value. When the type gene is an orientation sensor type, the orientation gene takes one of the 24 orientation levels.

[0057] Furthermore, the initial population is generated using the following method: Several individuals are generated according to a preset population size of 40, which is a preset value that balances the edge computing power of portable terminals and population diversity. The skeleton gene of each individual is fixedly written according to the result of step S103. The position gene of the filling node is uniformly and randomly selected in its deployable grid. The type gene is uniformly and randomly selected in the set of allowed types in the grid. The orientation gene of the type gene is directional and is uniformly and randomly selected from 24 orientation levels. The orientation gene of the type gene is omnidirectional and is directly taken from the omnidirectional identifier level value. Thus, each individual in the initial population is a valid solution.

[0058] Furthermore, the method used to organize local information exchange within the neighborhood and calculate the fitness vector is as follows: Each skeleton node is used as a local computing unit, and evolutionary partitions are divided according to the one-hop communication range. Each partition is responsible for the local computation of filling gene loci within its range. During the optimization process, each filling node only exchanges local information with its neighboring nodes within a 100-meter communication range. The exchanged information includes the current gene of the node, the comprehensive perception probability of the grid covered by the node, and the three local sub-target components. The boundary grids with overlapping coverage ranges are uniformly summarized by the skeleton node closest to them to avoid duplicate measurement. Each local computing unit combines the local information obtained from the neighborhood exchange to generate complete fitness evaluation data for each individual. The sub-target calculation method is the same as in step S102, and the three sub-target values ​​together constitute the fitness vector of the individual. All computation and neighborhood combination are completed on the node side without uploading global data to the central node, thus realizing distributed optimization in the field without central computing power.

[0059] Furthermore, the methods used to perform selection, crossover, and mutation operations are as follows: The selection operation uses a binary tournament format, randomly selecting two individuals each time. The winner is determined by the individual with the higher non-dominant level; if the non-dominant levels are the same, the individual with the greater crowding distance wins. The crossover operation is performed in a fixed order of position, type, and orientation, with a crossover probability of 0.8: First, single-point crossover is performed on position genes according to their grid numbers, with crossover points randomly selected within the range of filled gene positions. Then, type genes are determined by calculating the intersection of the set of type gene values ​​from the two parents at the same gene position with the allowed type set of the surface classification of the offspring's own position grid after crossover. If the intersection contains two valid values, they are randomly selected discretely and uniformly from that set; if the intersection contains only one valid value, the offspring directly inherits that valid value; if the intersection is empty, the offspring's position grid at that gene position is empty. Within the allowed type set, a type is randomly and uniformly reselected. Finally, the orientation gene is determined. If the type gene is an omnidirectional sensing node, the orientation gene directly takes the omnidirectional identifier value. If the type gene is a directional sensor and both parents have orientation values ​​for that position, the middle position of the two parents' orientation positions along the shortest angular distance direction is taken. If the type gene is a directional sensor but the parent has an omnidirectional identifier value for that position, one position is randomly selected from 24 orientation positions. If the parent's orientation gene has an omnidirectional identifier value, the offspring directly inherits the identifier value without orientation crossing.

[0060] During mutation, the location gene mutates with a 0.1 probability to its nearest reachable grid within its eight neighborhoods, the type gene mutates with a 0.1 probability to another type in the allowed type set of the location grid, and the orientation gene only shifts randomly by one position with a 0.1 probability when the type gene is a directional sensor type. The omnidirectional identifier position value does not participate in orientation mutation. When the type gene mutation causes a node to switch between omnidirectional and directional types, the orientation gene is set accordingly: mutated to omnidirectional, it is forced to take the omnidirectional identifier position value; mutated to directional, it randomly selects one position from 24 orientation positions. The 0.8 crossover probability and 0.1 mutation probability are preset values ​​for the evolutionary algorithm to balance global search and diversity. If, after crossover or mutation, an illegal situation occurs where the location falls into an undeployable grid or the type does not match the surface classification, gene repair is performed. The location gene is repaired to the nearest deployable grid, and the type gene is repaired to a legal value in the allowed type set, ensuring that each offspring individual is always legal.

[0061] Furthermore, non-dominated sorting and crowding distance calculation are performed using the following method: For any two individuals, if one individual's three sub-objective values ​​are all no worse than the other individual's and at least one sub-objective value is better than the other individual's, then the former dominates the latter. First, individuals in the current population that are not dominated by any other individual are classified as the first non-dominated layer. Then, these individuals are temporarily removed. The dominance determination is repeated for the remaining individuals to obtain the second non-dominated layer, and so on until all individuals are classified. Within the same non-dominated layer, the crowding distance is calculated as follows: individuals within the layer are sorted from smallest to largest according to their three sub-objective values. The crowding distance between the first and last individuals in each sub-objective dimension is the maximum value of that layer. The crowding component of the remaining individuals in that dimension is equal to the difference between the sub-objective values ​​of their two adjacent individuals divided by the difference between the maximum and minimum values ​​of that sub-objective value within the layer. The crowding distance of the individual is obtained by summing the crowding components of the three dimensions. A larger crowding distance indicates that the individual is sparser in the solution space and should be retained to maintain solution set diversity.

[0062] Furthermore, the methods for implementing elite retention, determining iteration termination, and outputting the solution set are as follows: After merging the parent and offspring populations in each generation, a unified non-dominated sort is performed. Individuals are selected in ascending order of non-dominated level and descending order of crowding distance within the same level, with an equal number of individuals to the population size retained for the next generation to ensure that superior individuals are not lost due to generational changes. The iteration stops when 100 generations have been reached, or when the individuals in the first non-dominated layer have remained unchanged for 10 consecutive generations. 100 generations and 10 consecutive generations are preset values ​​determined based on the algorithm's convergence experience. After stopping, all individuals in the first non-dominated layer of the final population are output as a Pareto optimal deployment scheme set. Any two individuals in this layer are not mutually dominant and correspond to several candidate deployment schemes with different focuses on coverage, energy consumption, and connectivity. Each individual corresponds to a complete deployment scheme containing the positions, types, and orientations of all filling nodes, which is used by step S105 to select a compromise scheme according to the task stage weight.

[0063] Please see Figure 5 , Figure 5 The convergence process of the three sub-objectives and the comprehensive optimization objective of the multi-objective evolutionary algorithm with the number of generations is shown. It can be seen that the threat-weighted effective coverage converges the fastest, reaching above 0.85 at around 30 generations. This is because the skeleton nodes have prioritized covering high-threat grids, and the evolution of fill nodes mainly supplements the coverage of medium and low-threat areas. The energy balance converges more slowly, stabilizing at around 50 generations, reflecting that energy balance requires global node coordination. The connectivity reliability steadily improves based on the skeleton connectivity support tree. The comprehensive optimization objective enters the convergence region after about 40 generations, indicating that the algorithm can fully converge within the preset upper limit of 100 generations.

[0064] This convergence characteristic verifies the effectiveness of the hierarchical optimization strategy of fixing the skeleton node and evolving the fill node. The skeleton node reduces the global deployment problem to local optimization, greatly compressing the search space, so that the edge computing power of the portable terminal can complete the solution in a reasonable time.

[0065] In this embodiment, step S104, based on the fixed skeleton in step S103, uses a fixed-length chromosome with a total number of gene loci equal to the planned number of nodes to encode only the filling nodes. This further compresses the original search for all nodes into a small-scale local optimization. By jointly encoding the location, type, and orientation, a single evolution can simultaneously determine "where the node is placed, what type it is, and what orientation the node faces." Combined with the value range of surface classification constraints, the intersection of three branches, and the repair of illegal genes, it ensures that legal individuals are always produced throughout the crossover and mutation process. This fundamentally solves the problem of node type mismatch with terrain and orientation deviation caused by only optimizing coordinates under the isomorphic assumption. During the optimization process, each node only exchanges local information with its 100-meter neighborhood and combines fitness on the node side. There is no need for the central computing unit to upload global data, allowing portable nodes with limited computing power to complete the solution in a distributed manner on-site. Then, by using non-dominated sorting, crowding distance, and elite retention, a Pareto optimal solution set with different focuses on coverage, energy consumption, and connectivity is obtained and evenly distributed, avoiding premature locking into a single solution by fixed weights.

[0066] S105: Using the dynamic weighting coefficients of the current task phase, the three sub-objectives of each scheme in the Pareto optimal deployment scheme set are weighted and summed. The scheme with the largest comprehensive optimization objective value is selected as the compromise deployment scheme. The threat weighted perception probability and network connectivity of each grid are verified. Compensation nodes are added to areas below the threshold. The deployment list containing the node location, type and orientation is output.

[0067] In step S105, a compromise solution is selected from the Pareto optimal deployment scheme set according to the weight of the current task stage. After double verification of coverage and connectivity, compensation nodes are added, and a deployment list that can directly guide on-site setup is output. The specific steps are as follows: Furthermore, the method for reserving backup nodes and defining their composition is as follows: All nodes carried before deployment are available nodes. A reserve of 10% to 20% of the available nodes is reserved as spare nodes, rounded down. This reserve ratio is a preset value determined based on on-site redundancy compensation requirements. Rounding down ensures that the planned number of reserved nodes does not exceed the actual number carried. The remainder are planned nodes. The planned nodes are determined as skeleton nodes and fill nodes through skeleton selection in step S103 and evolutionary optimization in step S104. Spare nodes do not participate in skeleton selection or evolutionary coding; their initial number is equal to the total number of available nodes minus the number of planned nodes. Spare nodes consist of two types: directional sensing nodes and omnidirectional relay nodes, configured according to a preset ratio. Directional sensing nodes account for two-thirds of the reserved number, and omnidirectional relay nodes account for one-third, rounded down. At least one of each type is retained. Directional sensing nodes are used for coverage compensation, and omnidirectional relay nodes are used for connectivity compensation. The number of each type of spare node is counted independently, decreasing by one for each node withdrawn.

[0068] Furthermore, the method for scoring and selecting a compromise deployment plan based on dynamic weights is as follows: For each scheme in the Pareto optimal deployment scheme set, its three sub-objective values—threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability—are multiplied by the current task stage weight coefficient determined in step S102, and then summed to obtain the comprehensive optimization objective value of the scheme. The weights for the rapid containment stage are 0.7, 0.2, and 0.1, and the weights for the long-term defense stage are 0.3, 0.2, and 0.5. The scheme with the largest comprehensive optimization objective value is selected as the compromise deployment scheme. When the comprehensive optimization objective values ​​of two schemes are the same or the difference is within the preset equal value tolerance, the one with the higher threat-weighted effective coverage is selected. The basis for this is that coverage containment is the primary objective in any stage, thus ensuring that the selected solution is unique.

[0069] Furthermore, the method for performing coverage and connectivity double verification is as follows: For coverage verification, the product of the overall perception probability and the intrusion threat level is calculated for each grid in the compromise deployment scheme, serving as the threat-weighted perception probability. Since this product does not exceed the grid's intrusion threat level itself, only grids with an intrusion threat level higher than the lower limit of 0.5 are considered weak. When their threat-weighted perception probability is lower than 0.5, they are marked as weak grids. Grids with an intrusion threat level not higher than 0.5 are not included in the weak grid determination even if the product is low. The 0.5 threshold and the 0.5 lower limit work together to ensure that the verification is mathematically convergent. Weak grids with eight adjacent neighborhoods are connected and aggregated into a compensation region, while isolated weak grids form a separate compensation region. For connectivity verification, the end-to-end connectivity reliability is calculated for each node along the deployment topology of the compromise scheme. Nodes with an end-to-end connectivity reliability lower than 0.6, or without adjacent nodes within 100 meters, are marked as weak nodes. 0.6 is the preset lower limit for end-to-end reliability, and 100 meters is consistent with the maximum communication distance in step S104.

[0070] Furthermore, the method used to determine the location, type, and orientation of the compensation and to add relays is as follows: For each compensation area, taking the grid with the highest intrusion threat level in the area as the benchmark, among the grid and its adjacent deployable grids (excluding water bodies and grids with a slope greater than 30 degrees), the grid that maximizes the weighted coverage gain of the new threat is selected as the compensation location. One is taken from the directional sensing backup nodes as the compensation node, and its type is determined according to the surface classification of the grid. The adaptation rule is consistent with the value range of the type gene in step S104. The orientation of the directional compensation node is aligned with the geometric centroid of the coordinates of all grids in the compensation area. For weakly connected nodes, the compensation location is selected from the deployable grids near the midpoint of the link between it and the nearest skeleton node. If the midpoint is not deployable, the deployable grid closest to the midpoint is selected. One is taken from the omnidirectional relay backup nodes for addition. Omnidirectional relay nodes have no orientation parameters, and their access parent node is the nearest skeleton node or fill node.

[0071] Furthermore, the method for iteratively supplementing and setting an exhaustion fallback is as follows: After each node is added, the comprehensive perception probability of the relevant grids is re-fused according to step S101, the end-to-end connectivity reliability is updated according to step S102, the coverage and connectivity double verification is re-executed, and the weak areas and weak connectivity nodes are updated. Iterative addition continues until the threat-weighted perception probability of all grids to be verified is not less than 0.5, the end-to-end connectivity reliability of all nodes is not less than 0.6, or the remaining number of any type of backup nodes is zero, i.e., exhausted. When the corresponding type of backup nodes are insufficient or the backup nodes are exhausted, the area that still does not meet the standard will not be added nodes but will be marked as a manual patrol area and its grid range will be recorded, ensuring that the method can still be executed when resources are limited and a clear disposal exit is given.

[0072] Furthermore, the deployment manifest is generated and output using the following method: After verification and compensation are completed, all nodes are arranged in the order of skeleton nodes, then fill nodes, and finally uncompensated nodes. Within the same level, they are arranged according to the order in which they were selected, forming a deployment list containing five fields: deployment coordinates, node type, orientation angle, access parent node, and deployment order for each node. The orientation field for omnidirectional relay nodes is marked as none. The deployment list is output to a portable planning terminal. Deployment personnel navigate to the nodes one by one according to the deployment order and complete the network according to the access parent node, realizing a direct connection from optimization results to on-site setup.

[0073] In this embodiment, step S105 uses the dynamic weights of the current task stage to uniformly score the Pareto optimal solution set output in step S104 and selects the one with the largest comprehensive optimization target value. This allows the same evolution result to directly select an appropriate solution based on the stage emphasis of rapid containment or long-term monitoring without re-optimization. When the comprehensive values ​​are the same, coverage rate is prioritized to ensure the uniqueness of the selected solution. Subsequently, the selected solution is placed under real perception and communication conditions to perform coverage and connectivity dual verification grid by grid and node by node. This accurately identifies the weak coverage areas and weak connectivity nodes left by the evolution, and then uses directional compensation nodes to align with the compensation areas and omnidirectional compensation nodes to align with the weak connectivity nodes. Relay nodes repair broken links, enabling the precise deployment of limited backup nodes on demand rather than blind encryption. Each additional node is re-verified and iterated until all verifiable grids and all nodes meet the standards. When backup nodes are exhausted, the remaining area is explicitly handed over to manual patrol, balancing the compensation loop with the feasibility under resource constraints. Finally, a deployment list containing deployment coordinates, node types, orientation angles, access parent nodes, and deployment order is output to a portable terminal in the order of skeleton, filling, and compensation. This allows the optimization results to be directly converted into on-site construction guidelines that can be set up according to the list, forming a complete closed loop from planning calculation to on-site deployment.

[0074] Please see Figure 4 , Figure 4 The diagram illustrates the three-dimensional distribution of the Pareto front obtained by the multi-objective evolutionary algorithm. Blue dots represent individuals in the Pareto optimal solution set, gray dots represent dominated solutions, and red dots represent fast containment compromise schemes selected from the Pareto solution set with a coverage weight of 0.7, a connectivity weight of 0.2, and an energy consumption weight of 0.1, emphasizing coverage quality. Green dots represent long-term maintenance compromise schemes selected from the Pareto solution set with a coverage weight of 0.3, a connectivity weight of 0.2, and an energy consumption weight of 0.5, emphasizing energy balance and durability. Both compromise schemes are located on the Pareto front, demonstrating that this embodiment can find the optimal balance among three conflicting objectives.

[0075] The distribution characteristics of the Pareto front indicate a clear inverse relationship between coverage and energy balance. Solutions with high coverage often require more nodes to operate continuously, leading to uneven energy consumption. Connectivity reliability reaches its optimal value in the medium coverage range because excessively high coverage requires nodes to be concentrated in high-threat areas, which may cause some nodes to be far from the sink node and links to become too long, thus reducing connectivity. The dynamic weighting mechanism allows the same Pareto solution set to quickly switch to the optimal solution as the task progresses, eliminating the need for repeated optimization.

[0076] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-target optimized deployment method for nodes in portable perimeter surveillance, characterized in that, include: Collect terrain elevation and surface classification data and rasterize the perimeter. Establish a probabilistic perception model that considers terrain occlusion and distance attenuation. Calculate the intrusion threat level of each grid based on visibility and surface accessibility. A multi-objective optimization function is constructed with threat-weighted effective coverage, node remaining energy balance, and network connectivity reliability as sub-objectives. The threat-weighted effective coverage is the sum of the comprehensive perception probabilities of each grid weighted by the threat degree, and dynamic weights are set according to the stage. We extract the extreme points of perimeter curvature and the highest points as candidate backbone positions. Under connectivity constraints, we evaluate the marginal gain using an optimization function, prioritize the high-threat grid coverage with the largest gain, greedily select the backbone nodes, and construct a connected support tree. Based on the skeleton, each node exchanges neighborhood information and jointly encodes its position, type and orientation angle. Using the three objectives as fitness, a multi-objective evolutionary algorithm is used for iteration. After non-dominated sorting and crowding, a Pareto optimal deployment scheme set is obtained. The sub-objectives of each scheme in the scheme set are weighted and summed according to the current stage dynamic weights. The compromise deployment scheme with the largest comprehensive objective value is selected. The grid threat weighted perception probability and network connectivity are verified. Compensation nodes are added for areas below the threshold. The deployment list with location, type and orientation is output.

2. The method according to claim 1, characterized in that, Establish a probabilistic perception model that considers terrain occlusion and distance attenuation, including: Obtain the deployment coordinates, sensor installation height, and effective sensing radius of each node to be deployed, as well as the center coordinates and ground elevation of each grid cell, and calculate the horizontal distance from the node to the center of the grid cell; Elevation profiles are extracted along the line connecting the node and the center of the grid cell at sampling intervals of no more than 1 meter. The actual ground elevation of each sampling point on the line is compared with the line-of-sight elevation of the sampling point on the line connecting the node sensor installation height to the grid ground elevation. If the actual ground elevation of any sampling point is higher than the line-of-sight elevation, it is determined that there is terrain occlusion, and the perception probability of the node for the grid cell is set to 0. If there is no terrain obstruction, the perception probability decreases exponentially with horizontal distance. When the horizontal distance is 0, the perception probability is 1. When the horizontal distance is equal to the effective perception radius, the perception probability decreases to 0.

5. When the horizontal distance is greater than twice the effective perception radius or the perception probability decreases to a probability threshold below 0.05, the perception probability is set to 0. For all nodes covering the same grid cell, the overall perception probability of the grid cell is calculated by subtracting the product of the missed detection probabilities of each node for that grid cell from 1. The missed detection probability of a single node is equal to 1 minus the perception probability of that node for that grid cell.

3. The method according to claim 2, characterized in that, Intrusion threat level of each grid cell is calculated based on visibility and surface accessibility, including: Starting from a grid cell, three lines of sight are emitted along the outer perimeter normal direction and at 15-degree intervals on both sides of the normal, forming a total of seven lines of sight. Within a 50-meter warning depth range from the perimeter, elevation visibility is assessed point-by-point along each line of sight. The proportion of the unobstructed segment from the grid cell to the first obstruction point for each line of sight is calculated to the total length of that line of sight within the warning depth range. If the entire line of sight is unobstructed, this proportion is set to 1. The arithmetic mean of this proportion for all seven lines of sight is taken as the visibility factor for that grid cell. The passage cost is calculated based on the slope value and surface type at each point within the warning depth range outside the grid cell. The slope value is calculated from the elevation gradient of the terrain elevation data, and the slope is less than 15 degrees. The slope access cost is set to 0 when the slope is greater than 30 degrees, and to 1 when the slope is greater than 30 degrees. For slopes between 15 and 30 degrees, linear interpolation is used to determine the access cost. The surface access cost is determined based on surface classification data. The surface access cost is obtained from a preset surface access cost mapping table and normalized to the interval between 0 and 1. The slope access cost is weighted with a weight of 0.6 and the surface type access cost is weighted with a weight of 0.4 to obtain the comprehensive access cost. The accessibility factor of the grid cell is obtained by subtracting the comprehensive access cost from 1. The concealment access factor is obtained by subtracting the visibility factor from 1. The concealment access factor and the accessibility factor are weighted with a weight of 0.5 and normalized to the interval between 0 and 1 to obtain the intrusion threat level of the grid cell.

4. The method according to claim 3, characterized in that, Construct a multi-objective optimization function with threat-weighted effective coverage, node residual energy balance, and network connectivity reliability as sub-objectives, including: The threat-weighted effective coverage rate is the sum of the comprehensive perception probabilities of each grid cell, weighted by their intrusion threat levels, and then divided by the sum of the intrusion threat levels of all grid cells, resulting in a value normalized to the range of 0 to 1. If the sum of the intrusion threat levels of all grid cells is 0, the threat-weighted effective coverage rate is 0. The node remaining energy balance is determined as follows: the predicted energy consumption of a node within a set duty period is the sum of its perception energy consumption, communication energy consumption, and standby energy consumption. Perception energy consumption is the product of the node's perception power and duty duration. Communication energy consumption is determined along the connected support tree according to the communication distance between the node and its parent node and increases with increasing communication distance. Standby energy consumption is the product of the node's standby power and duty duration. The remaining energy after the node's duty period is obtained by subtracting the predicted energy consumption from the initial energy, and its proportion of the initial energy is calculated. The remaining energy is then calculated by subtracting the total energy consumption from the initial energy. The standard deviation of the remaining energy ratio of a node is used as the node's remaining energy balance. First, the standard deviation of the remaining energy ratio of a node is calculated, and then normalized according to the preset maximum standard deviation. The inverse of the normalized value is used as the node's remaining energy balance. The network connectivity reliability is determined in the following way: the single-hop link reliability is determined based on the distance between adjacent nodes and the terrain occlusion. When the distance between adjacent nodes is less than 50 meters (the reliable communication distance), the single-hop link reliability is 1; when it is greater than 100 meters (the maximum communication distance), it is 0; when it is in between, it decreases linearly with the increase of distance. When there is terrain occlusion in the connection between adjacent nodes, it is reduced according to the proportion of the occlusion segment. The product of the reliability of all single-hop links on the path from each node to the sink node is calculated along the connectivity support tree as the end-to-end connectivity reliability of that node. The arithmetic mean of the end-to-end connectivity reliability of all nodes is taken as the network connectivity reliability. The sub-objectives, which are all normalized to the range of 0 to 1 and whose larger values ​​are preferred, are multiplied by their corresponding weight coefficients and then summed to obtain the comprehensive optimization objective function. The deployment optimization direction is to obtain the maximum value of the comprehensive optimization objective function. The weight coefficients are dynamically set according to the stage of the alert task.

5. The method according to claim 1, characterized in that, The extreme points of perimeter curvature and the highest points are extracted as candidate backbone positions. Marginal gains are evaluated using an optimization function under connectivity constraints, including: The perimeter centerline is discretized into an ordered sequence of boundary points at 5-meter intervals. For each boundary point, five adjacent points before and after it are taken, and the directional deflection angle of the line connecting the adjacent points before and after it is calculated. Boundary points with a deflection angle greater than 30 degrees are marked as geometric curvature extreme points. Within the warning buffer zone on both sides of the perimeter, a sliding window with a radius of 20 meters is used to traverse the grid grid by grid. Grids with an elevation higher than all other grids in the window and with an elevation difference greater than 2 meters from the lowest point of the window edge are marked as terrain high points. The extreme points of geometric curvature are merged with the high points of the terrain. For two candidate locations with a distance of less than 10 meters, the one with the higher elevation is retained. After deduplication, a set of candidate backbone locations is formed. The initial skeleton node set contains only the sink node. A greedy selection is performed round by round. For each candidate backbone location, a node is assumed to be placed and the marginal gain of the three sub-objectives of the multi-objective optimization function after its access is calculated. The coverage gain is the difference between the threat-weighted effective coverage before and after the node is placed, which is the intrusion threat-weighted increment of the newly added grid that reaches the coverage determination threshold. The connectivity gain is the end-to-end connectivity of the entire network after the candidate location is connected to the support tree with a reliable link. The improvement of the reliability arithmetic mean is the energy consumption balancing gain, which is the reduction of the standard deviation of the proportion of remaining energy of all nodes after access. The three marginal gains are normalized and then weighted and summed according to the weight coefficient of the current warning task stage to obtain the comprehensive marginal gain. Only when the distance between the candidate position and the selected skeleton node is less than 100 meters of the maximum communication distance and the single-hop link reliability is higher than 0.5, the candidate position can participate in the selection. In each round, the candidate position with the largest comprehensive marginal gain is selected and added to the skeleton node set, and the selected skeleton node with the largest end-to-end connectivity reliability is used as the parent node to access the connectivity support tree. The greedy selection process continues until the comprehensive perception probability of all high-threat grids with an intrusion threat level higher than 0.6 reaches the coverage judgment threshold of 0.7, or all candidate backbone positions are selected, or the number of skeleton nodes reaches the set upper limit, thus obtaining the skeleton node set and the connected support tree.

6. The method according to claim 5, characterized in that, Prioritize selecting the grid with the greatest coverage gain for high-threat grids, greedily select skeleton nodes, and construct a connected support tree, including: All grid cells are sorted from highest to lowest intrusion threat level. Grid cells with an intrusion threat level higher than 0.6 are marked as high-threat grid cells and form a priority coverage queue. In each round of greedy selection, the number of newly covered high-threat grid cells after assuming node placement at each candidate backbone location is counted, and the sum of their intrusion threat levels is used as the high-threat coverage gain. The candidate location with the largest high-threat coverage gain is selected first. When two or more candidate locations have the same high-threat coverage gain, their connectivity gain and energy balance gain are compared sequentially, and the best one is selected. After each backbone node is selected, grid cells whose comprehensive sensing probability has reached the coverage determination threshold of 0.7 are removed from the priority coverage queue. In subsequent marginal gain calculations, the candidate nodes are removed, and based on the updated grid coverage status and the current connected support tree structure, the marginal gain of all remaining candidate backbone locations is recalculated. When a candidate location is connected to the support tree, if it meets the communication distance and link reliability constraints with two or more selected backbone nodes, the node with the highest end-to-end connectivity reliability is used as the parent node to establish the primary link, and the remaining links that meet the constraints are recorded as backup links. When the single-hop link reliability of the primary link is lower than 0.5, the system switches to the backup link with the highest single-hop reliability. This process is repeated until all high-threat grids in the priority coverage queue reach the coverage determination threshold.

7. The method according to claim 1, characterized in that, Based on the skeleton nodes, each node to be deployed exchanges local information within its neighborhood, jointly encoding its location, type, and orientation angle. The fitness evaluation is based on three sub-objectives of the multi-objective optimization function, including: The total number of available nodes is the preset value of the nodes carried before deployment. 10% to 20% of the available nodes are reserved as spare nodes and rounded down. The remainder are planned nodes. After the skeleton node set is determined, the number of filling nodes is equal to the number of planned nodes minus the number of skeleton nodes. Each individual in the population corresponds to a chromosome of fixed length. The total number of chromosome gene loci is equal to the number of planned nodes. Spare nodes do not enter the population or chromosome. The position, type, and orientation of the gene loci corresponding to the skeleton nodes remain fixed during initialization and the entire iterative optimization process and do not participate in crossover or mutation. Only the gene loci corresponding to the filling nodes participate in evolution. The supplementary compensation nodes are post-processing nodes after the compromise scheme is determined. They do not enter the population or chromosome and do not participate in crossover or mutation. Each filled node corresponds to a gene locus, and each gene locus includes three segments: location gene, type gene, and orientation gene. The location gene is represented by the raster cell number, and its value range is the deployable raster after excluding water bodies and slopes greater than 30 degrees. The type gene is represented by the node type number, and its value range is the set of node types allowed by the raster surface classification corresponding to that location gene. Among them, vibration-type nodes are excluded for soft soil raster, and infrared-type nodes are excluded for dense vegetation raster. The orientation gene is discretized into 24 orientation levels from 0 to 345 degrees at 15-degree intervals. An additional omnidirectional identifier level is set. When the type gene is an omnidirectional sensing node type, the orientation gene takes the omnidirectional identifier level value. When the type gene is a directional sensor type, the orientation gene takes one of 24 orientation levels. An initial population is generated according to a set number. The skeleton gene of each individual in the population is fixed. The position gene of the filling node is randomly selected from the deployable grid. The type gene is randomly selected from the set of allowed types in the grid. The orientation gene of the type gene is directional and is randomly selected from 24 levels. The orientation gene of the type gene is omnidirectional and directly takes the omnidirectional identifier level value. In crossover operations, if the orientation gene of the corresponding gene position in the parent generation has an omnidirectional identifier value, the offspring directly inherit the omnidirectional identifier value without orientation crossover. In mutation operations, orientation genes with omnidirectional identifier values ​​do not participate in orientation mutation, and type gene mutation causes the orientation gene to be linked when a node switches between omnidirectional and directional types. Orientation genes mutated to omnidirectional types are forced to take the omnidirectional identifier value, while orientation genes mutated to directional types randomly select one of the 24 identifiers. During the optimization process, each filling node only communicates with its counterparts within a 100-meter radius. Neighboring nodes exchange local information, including the node's current gene, the comprehensive perception probability of the covered grid, and the three local sub-target components. The boundary grids with overlapping coverage areas are summarized by the nearest skeleton node. Each filling node combines the local information obtained from the neighborhood exchange to form the individual's fitness evaluation data, without needing to upload global information to the central node. For each individual, the threat-weighted effective coverage rate, node remaining energy balance, and network connectivity reliability are calculated separately. The three sub-target values ​​together constitute the individual's fitness vector.

8. The method according to claim 7, characterized in that, A multi-objective evolutionary algorithm is used iteratively to obtain a set of Pareto optimal deployment schemes through non-dominated sorting and crowding, including: The selection, crossover and mutation operations are performed sequentially on individuals in the population. The selection operation adopts a tournament-style approach, with priority given to non-dominated hierarchies and those with greater crowding distances within the same hierarchical level. The crossover operation first performs single-point crossover on the position genes according to the raster number, and then determines the type genes. That is, it calculates the intersection of the set of type gene values ​​of the two parents at the same gene locus with the set of allowed surface classification types of the raster of the gene locus after the crossover of the offspring. If the intersection contains two valid values, a discrete uniform random selection is made within it. If the intersection contains only one valid value, the offspring directly inherits the valid value. If the intersection is empty, the offspring discretely and uniformly randomly selects a type again within the set of allowed types of the raster of the gene locus. After the type gene of the offspring gene locus is determined, the orientation gene is then determined. If the type gene is of the omnidirectional sensing node type, the orientation gene is directly taken as the omnidirectional identifier gear value. If the type gene is of the orientation sensor type and both parents have orientation gear values ​​for this gene locus, the orientation gene is taken as the middle gear of the orientation gears of the two parents along the shortest angular distance direction. If the type gene is of the orientation sensor type but there is an omnidirectional identifier gear value in the orientation gene of this gene locus of the parents, the orientation gene is randomly selected from 24 orientation gears, and the crossover probability is 0.

8. In mutation operations, the position gene mutates to an adjacent reachable grid with a probability of 0.1, the type gene mutates to another type in the set of allowed types for the position grid with a probability of 0.1, and the orientation gene only shifts randomly by one position with a probability of 0.1 when the type gene is a directional sensor type. The omnidirectional identifier position value does not participate in orientation mutation, and the orientation gene is linked when the type gene mutation causes the node to switch between omnidirectional and directional types. The orientation gene mutated to omnidirectional is forced to take the omnidirectional identifier position value, and the orientation gene mutated to directional type randomly selects one position from 24 orientation positions. If, after crossover or mutation, the position falls into an undeployable grid or the type does not match the surface classification, it is repaired to the nearest legal value. Non-dominated sorting is performed as follows: for any two individuals, if the values ​​of all three sub-targets of one individual are not inferior to those of the other individual and at least one sub-target is superior to that of the other individual, then the former dominates the latter, and the individuals in the current population that are not dominated by any other individual are classified as the first non-dominated layer. After removing the layer, repeat the above division for the remaining individuals to obtain each non-dominated layer until all individuals are layered. Within the same non-dominated layer, calculate the crowding distance. Sort the individuals within the layer according to the three sub-target values. For each sub-target dimension, the crowding distance of the first and last individuals in the sort is the maximum value of that layer. For the remaining individuals in that dimension, the crowding component is the difference between the sub-target values ​​of their two adjacent individuals divided by the difference between the maximum and minimum values ​​within that sub-target layer. Sum the three dimensional components to obtain the individual crowding distance. In each generation, merge the parent and offspring populations and perform a unified non-dominated sort. Select individuals equal to the population size in order of increasing non-dominated level and decreasing crowding distance within the same level to retain them for the next generation. Stop iteration when the first non-dominated layer no longer changes after 100 generations or 10 consecutive generations. Output all individuals in the first non-dominated layer of the final population as a Pareto optimal deployment scheme set, with each individual corresponding to a deployment scheme containing the positions, types, and orientations of all filling nodes.

9. The method according to claim 1, characterized in that, The sub-objectives of each scheme in the scheme set are weighted and summed using the current dynamic weights. The compromise deployment scheme with the largest comprehensive objective value is selected. The threat-weighted perception probability and network connectivity of each grid are verified. Compensation nodes are added to areas below the threshold. The deployment list, including node location, type, and orientation, is output, including: All nodes carried before deployment are available nodes. 10% to 20% of the available nodes are reserved as spare nodes and rounded down. The rest are planned nodes. The planned nodes are determined as skeleton nodes and fill nodes through skeleton selection and evolutionary optimization. Spare nodes do not participate in skeleton selection and evolutionary coding. Spare nodes are composed of directional sensing nodes and omnidirectional relay nodes and are only used for compensation and supplementation after verification. Their initial number is equal to the total number of available nodes minus the number of planned nodes. For each scheme in the Pareto optimal deployment scheme set, its three sub-objective values ​​are multiplied by the weight coefficient corresponding to the current task stage and then summed to obtain the comprehensive optimization objective value of the scheme. The scheme with the largest comprehensive optimization objective value is selected as the compromise deployment scheme. When the comprehensive optimization objective values ​​are the same, the one with the higher threat-weighted effective coverage is selected. For the compromise deployment scheme, the product of the comprehensive perception probability and the intrusion threat level is calculated grid-by-grid as the threat-weighted perception probability. Grids with an intrusion threat level higher than 0.5 and a threat-weighted perception probability lower than 0.5 are marked as weak grids. Grids with an intrusion threat level not higher than 0.5 are not included in the weak grid determination. Adjacent weak grids are connected to form a compensation area. At the same time, the end-to-end connectivity reliability is calculated node-by-node along the deployment topology. Nodes with an end-to-end connectivity reliability lower than 0.6 or without adjacent nodes within 100 meters are marked as weak nodes. For each compensation area, compensation locations are selected from the grid with the highest intrusion threat level and its adjacent deployable grids. The type of compensation node is determined according to the surface classification adaptation of that grid. The orientation of the compensation node is aligned with the center of the compensation area. For nodes with weak connectivity, omnidirectional relay nodes can be added to the grid near the midpoint of the link between the node and the nearest skeleton node. After each compensation node is added, the coverage and connectivity verification is re-executed and the weak area is updated. The process is iteratively added until the threat-weighted perception probability of all grids to be verified is not less than 0.5, the end-to-end connectivity reliability of all nodes is not less than 0.6, or the backup nodes are exhausted. Areas that still do not meet the standards when the backup nodes are exhausted are marked as manual patrol areas. Finally, the deployment list containing the deployment coordinates, node type, orientation angle, access parent node, and deployment order of each node is output to the portable terminal in the order of skeleton nodes first, then filling nodes, and finally compensation nodes.

Images