Network-based heterogeneous particle swarm optimization-based coverage deployment method and device

Abstract

The application relates to a networked heterogeneous particle swarm optimization-based coverage deployment method and device. The method comprises the following steps: obtaining task parameters corresponding to a sensor coverage deployment task; constructing a coverage deployment optimization model with a weighted sum of maximum general monitoring area coverage rate, key monitoring area coverage rate and sensor resource utilization rate as a target function, and taking an upper limit of the number of sensors, full connectivity constraint and deployment range constraint as constraint conditions; initializing a heterogeneous group containing priori particles and following particles, and the priori particles performing three-layer search of neighborhood disturbance, directional refinement and stagnation trigger reset; the following particles switching a foraging-alarm dual mode position update according to a probability, iteratively updating an optimal solution after boundary constraint and fitness reevaluation, and outputting an optimal deployment scheme when termination. The method can simultaneously improve the comprehensive coverage quality and optimization solving efficiency of the deployment scheme in a wireless sensor network coverage deployment scene containing heterogeneous priorities.

Description

Technical Field

[0001] This application relates to the field of wireless sensor deployment technology, and in particular to a coverage deployment method and apparatus based on networked heterogeneous particle swarm optimization. Background Technology

[0002] With the rapid development of wireless sensor networks and UAV communication coverage technologies, how to efficiently deploy a limited number of sensor nodes to maximize coverage of the monitored area has become a key issue in network planning and situational awareness. Particle Swarm Optimization (PSO) is widely used to solve such non-convex, multi-modal sensor deployment optimization problems due to its simplicity and fast convergence. Its basic idea is to simulate the social behavior of bird flocks, guiding candidate solutions through the historical best experiences of individuals and the group, driving iterative evolution in the search space. Currently, PSO-based coverage deployment methods typically employ a homogeneous design where all particles execute uniform velocity update rules, have the same acceleration coefficients and inertia weights, or introduce finite heterogeneous variants at the parameter, topology, or rule levels to seek a balance between exploring unknown regions and developing known high-quality solutions.

[0003] However, the homogeneous behavior of standard PSO forces each particle to reconcile the conflict between exploration and development. On the discontinuous fitness gradients and large-area local optima generated by the binary perception model, it is prone to premature convergence or ineffective oscillations, severely limiting optimization performance on complex overlay landscapes. While various heterogeneous PSO variants introduce strategy diversity, they struggle to achieve true synergy due to blurred role boundaries and mutual interference between explorer and developer behaviors. Furthermore, the inefficient transfer of high-quality solution information between the exploration and development phases makes it difficult to support efficient iterative refinement.

[0004] To address the heterogeneous priority deployment requirements of surveillance areas encompassing general and key monitoring zones, existing methods typically handle multi-index coverage issues by adjusting the fitness function or introducing resampling strategies. However, this process inherently contradicts the wide-area sensing needs of general areas and the dense tracking needs of key areas. When sensor resources are limited, excessive tilting towards key areas weakens the coverage breadth of general areas, while uniform and distributed deployment cannot guarantee the coverage accuracy of key areas. Existing strategies struggle to achieve an effective balance between these two considerations. Furthermore, the two-dimensional discretized grid partitioning used in coverage assessment generates regions with gently changing fitness values ​​and numerous locally optimal platforms. In such terrains, existing solutions, when dealing with individuals in long-term stagnant states, fail to properly manage the boundary between retaining and resetting accumulated effective search information, easily leading to wasted search resources or delayed recovery from stagnation. This negatively impacts the algorithm's search efficiency and final deployment quality in complex geographical environments. These shortcomings collectively limit the practicality and robustness of coverage deployment schemes in complex geographical environments. Summary of the Invention

[0005] Therefore, it is necessary to provide a coverage deployment method and apparatus based on networked heterogeneous particle swarm optimization to address the aforementioned technical problems.

[0006] A coverage deployment method based on networked heterogeneous particle swarm optimization, the method comprising: Obtain the task parameters corresponding to the sensor coverage deployment task; the task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of sensors, and the sensing radius of each sensor; Based on the task parameters, a coverage deployment optimization model is constructed. The coverage deployment optimization model takes the weighted sum of the coverage rate of general monitoring areas, the coverage rate of key monitoring areas, and the utilization rate of sensor resources as the objective function, and the upper limit constraint of the number of sensors, the full connectivity constraint, and the deployment range constraint as the constraint conditions. A particle swarm containing prophet particles and follower particles is initialized in the search space, and the position of each particle represents the distribution position of each sensor in the monitored area. For each prophet particle, a neighborhood perturbation search is performed along the direction away from the randomly selected reference prophet particle, and a directional refinement search is performed along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches a preset migration threshold, the position of the current prophet particle is randomly reset. For each of the following particles, the behavior mode is switched according to a preset probability. In the foraging mode, a random prophet particle is selected as a social attraction target to update its own speed and position. In the vigilance mode, a fitness-adaptive position update is performed towards the centroid of the following particle group. The fitness is calculated by the objective function. Boundary constraints are applied to all particles and their fitness is recalculated. The individual historical optimal solution and the global optimal solution are updated. The position of the prophet particle with better fitness is synchronized to the corresponding follower particle. The particle population is iteratively updated. The iteration stops when the preset iteration termination condition is met. The sensor coverage deployment scheme corresponding to the global optimal solution is output.

[0007] A coverage deployment device based on networked heterogeneous particle swarm optimization, the device comprising: The parameter acquisition module is used to acquire the task parameters corresponding to the sensor coverage deployment task; the task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of sensors, and the sensing radius of each sensor; The model building module is used to build a coverage deployment optimization model based on the task parameters. The coverage deployment optimization model takes the weighted sum of the coverage rate of general monitoring areas, the coverage rate of key monitoring areas and the utilization rate of sensor resources as the objective function, and the upper limit constraint of the number of sensors, the full connectivity constraint and the deployment range constraint as the constraints. An initialization module is used to initialize a particle swarm containing prophet particles and follower particles in the search space, where the position of each particle represents the distribution position of each sensor in the monitored area. The particle search module is used to perform a neighborhood perturbation search along the direction away from the randomly selected reference prophet particle for each prophet particle, and to perform a directional refinement search along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches the preset migration threshold, the position of the current prophet particle is randomly reset. The position update module is used to switch the behavior mode of each of the following particles according to a preset probability. In the foraging mode, it selects a random prophet particle as a social attraction target to update its own speed and position. In the vigilance mode, it performs a fitness-adaptive position update towards the centroid of the following particle group. The fitness is calculated by the objective function. The results output module is used to perform boundary constraint processing on all particles and recalculate their fitness, update the individual historical optimal solution and the global optimal solution, synchronize the position of the prophet particle with better fitness to the corresponding follower particle, iteratively update the particle population, stop the iteration when the preset iteration termination condition is met, and output the sensor coverage deployment scheme corresponding to the global optimal solution.

[0008] The aforementioned coverage deployment method and apparatus based on networked heterogeneous particle swarm optimization achieves structural decoupling between global exploration and local development by dividing the particle swarm into two heterogeneous roles: prophet particles and follower particles. This overcomes the inherent limitations of homogeneous particle swarm algorithms in balancing exploration and development. By configuring a multi-layered search mechanism and a stagnation-triggered reset strategy for prophet particles, accurate local optimal escape can be achieved in non-convex coverage fitness landscapes, reducing the computational cost of ineffective iterations. By setting a dual-mode motion with probability switching and a multi-attractor guidance method for follower particles, a balance between the diversity and stability of the swarm search can be maintained, avoiding the risk of premature homogeneous convergence. By constructing a multi-index weighted coverage optimization model and establishing information feedback pathways between roles, the coverage requirements of differentiated priority areas can be coordinated, improving the efficiency of wireless sensor resource utilization. In this embodiment of the invention, in wireless sensor network coverage deployment scenarios with heterogeneous priorities, the overall coverage quality and optimization efficiency of the deployment scheme can be improved simultaneously, balancing the practical needs of dense sensing in key areas and wide-area coverage in general areas with limited resources. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a coverage deployment method based on networked heterogeneous particle swarm optimization in one embodiment. Figure 2 This is a schematic diagram of a coverage deployment framework based on networked heterogeneous particle swarm optimization in one embodiment; Figure 3 This is a schematic diagram showing the area division and sensor deployment results covering a deployment scenario in one embodiment. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0011] This invention proposes a coverage deployment method based on Network-based Heterogeneous Particle Swarm Optimization (HNPSO). Its core design concept stems from three key insights: Insight 1: Exploration and exploitation require information of different qualities. Exploration requires diverse and dispersed search directions to cover a vast unknown space; exploitation requires fine-grained local searches around known areas of potential. These two needs differ fundamentally in the type of information and the scale of the search, therefore a single rule of action is inherently insufficient.

[0012] Insight 2: Leadership in natural groups is sparsely distributed. The groundbreaking research by Couzin et al. shows that in animal group decision-making, only 5%-15% of individuals (informed individuals) can exert a decisive influence on the large-scale movement direction of the entire group, and this sparse leadership mechanism remains effective even when followers cannot identify the leader. This provides a biological basis for the minority-knowledge-particle → majority-follower-particle architecture of this invention.

[0013] Insight 3: The fitness landscape generated by the cover-optimized binary perception model has a discontinuous gradient structure. Optimization algorithms lacking an effective escape mechanism are prone to stagnation in configurations that are locally stable but globally insufficient. A soft-restart mechanism is needed that can both escape local optima and preserve the accumulated search information.

[0014] Based on the above insights, this invention specifically decomposes the particle swarm function into two types of heterogeneous roles: Prophet Particles: These particles lead the exploration process. Relatively few in number (approximately half the total population), they continuously search for potentially new regions in the search space through a three-tiered search mechanism: neighbor-guided local perturbations, global-optimal refinement, and stagnation-triggered migration. Each Prophet Particle maintains a stagnation counter, recording the number of consecutive iterations without improvement, which serves as a signal to trigger migration (soft restart).

[0015] Follower Particles: These particles perform adaptive execution. Their numbers are equal to those of Seer Particles, and they probabilistically switch between two behavior modes: In Seer-guided foraging mode (probability pf=0.8), Follower Particles select a random Seer Particle as their social attraction target to perform PSO-style updates; in safe-gathering vigilance mode (probability 1-pf=0.2), Follower Particles move towards the group's center of mass, their movement speed adaptively adjusted by their own fitness.

[0016] The two types of roles interact through a structured information transmission mechanism: at the end of each iteration, the prophet particle injects a solution that is superior to the corresponding follower particle into the follower population; the follower population updates its individual historical best and global best after each movement; in the next iteration, the prophet particle uses the updated global best to guide refinement. This forms a virtuous cycle of exploration and discovery (prophet) → development and refinement (follower) → feedback and update (information injection).

[0017] In one embodiment, such as Figure 1 As shown, a coverage deployment method based on networked heterogeneous particle swarm optimization is provided, including the following steps: Step 102: Obtain the task parameters corresponding to the wireless sensor coverage deployment task. The task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of wireless sensors, and the sensing radius of each wireless sensor.

[0018] The general monitoring area refers to the area outside the key monitoring area within the monitoring range. Its coverage requirements are: the number of monitoring nodes configured per unit area does not exceed the first threshold (for example, no more than 1 node per 100 square meters), and the maximum distance between adjacent monitoring nodes is not less than the first distance (for example, not less than 50 meters) to achieve wide-area sparse coverage perception. The key monitoring area is usually a smaller rectangular area located in the center of the monitoring area, which together with the general monitoring area forms a complete monitoring range. The two do not overlap in space. The sensing radius of the wireless sensor refers to the maximum spatial distance at which a single wireless sensor can effectively detect a target. It is the core hardware parameter that determines the coverage range of a single wireless sensor.

[0019] Step 104: Construct a coverage deployment optimization model based on the task parameters. The coverage deployment optimization model takes the weighted sum of maximizing the coverage rate of general monitoring areas, the coverage rate of key monitoring areas, and the utilization rate of wireless sensor resources as the objective function, and takes the upper limit constraint of the number of wireless sensors, the full connectivity constraint, and the deployment range constraint as the constraints.

[0020] General monitoring area coverage refers to the proportion of the general monitoring area covered by the sensing range of wireless sensors to the total area of ​​the general monitoring area, used to measure the coverage breadth of wide-area sensing. Key monitoring area coverage refers to the proportion of the key monitoring area covered by the sensing range of wireless sensors to the total area of ​​the key monitoring area, used to measure the coverage accuracy of core areas. Wireless sensor resource utilization refers to the proportion of the union area of ​​all wireless sensor coverage areas to the total area of ​​the entire monitoring area, used to measure the efficiency of the spatial distribution of wireless sensors. Full connectivity constraint means that all wireless sensors must form a complete network capable of communicating with each other through wireless communication links, and data can be transmitted between any two wireless sensors via multi-hop paths.

[0021] It is understandable that this step, by weighting and integrating multi-dimensional coverage indicators and setting boundary constraints, can transform heterogeneous priority coverage requirements into a mathematical optimization problem that can be quantified and solved. This is beneficial for taking into account the coverage objectives of different priority areas under limited wireless sensor resources.

[0022] Step 106: Initialize a particle swarm containing prophet particles and follower particles in the search space, where the position of each particle represents the distribution location of each wireless sensor in the monitored area.

[0023] The search space refers to the multidimensional solution space comprised of all feasible wireless sensor deployment schemes. A particle is a candidate solution vector in the solution space, and its position refers to its coordinates in the multidimensional search space. Each coordinate dimension corresponds to the position component of a wireless sensor, thus characterizing the distribution of a group of wireless sensors within the monitored area. Fitness refers to the evaluation value of a candidate solution under the objective function, used to quantify the overall performance of the deployment scheme. The swarm refers to the collection of all particles. Among them, the prophet particles are those responsible for global exploration, searching for deployment areas with high fitness potential in the solution space; follower particles are those responsible for local refinement, conducting detailed searches near areas known to have search potential.

[0024] It is understandable that this step, by dividing the particle swarm into functional roles, can decouple global exploration from local development at the structural level, which is conducive to breaking the behavioral limitations of homogeneous particle swarms and laying the foundation for division of labor and cooperation for subsequent iterative optimization.

[0025] Step 108: For each prophet particle, perform a neighborhood perturbation search along the direction away from the randomly selected reference prophet particle, and perform a directional refinement search along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, accumulate the number of consecutive search failures. When the number of consecutive search failures reaches the preset migration threshold, perform a random reset on the position of the current prophet particle.

[0026] Neighborhood perturbation search refers to the Seer particle adjusting its position in directions away from its companions, using them as a reference, to maintain the spatial search diversity of the group. Targeted refinement search refers to the Seer particle exploring its neighborhood in directions away from the global optimum, using its current global optimum as a reference, to uncover potential better solutions around the elite solution. Consecutive search failures refer to the cumulative number of times a single Seer particle's position has not improved in multiple consecutive iterations. The migration threshold is the upper limit of consecutive failures that trigger a particle position reset, used to balance exploration patience and escape efficiency.

[0027] It is understandable that this step, through multi-layered progressive search and individual-level stagnation reset mechanism, can enhance the ability of the prophet particle to explore non-convex fitness landscapes, which is conducive to timely escape from local optimum platforms and reduce the consumption of computational resources caused by invalid iterations.

[0028] Step 110: For each following particle, switch its behavior mode according to a preset probability. In foraging mode, select a random prophet particle as a social attraction target to update its own velocity and position. In vigilance mode, perform a fitness-adaptive position update towards the centroid of the following particle group. The fitness is calculated using the objective function.

[0029] Foraging mode refers to the behavior of follower particles moving with the prophet particle as a guide, used to follow the prophet particle's exploration results and perform local refinement. Alert mode refers to the behavior of follower particles gathering towards the group's center of mass, used to maintain group structural stability and prevent excessive particle dispersion. Social attraction target refers to the reference particle position used by follower particles for social learning during movement updates. Fitness adaptation refers to the dynamic adjustment of a particle's movement step size based on its fitness level; higher fitness indicates a stronger tendency to gather.

[0030] It is understandable that this step, through the dual-mode motion with probability switching and the multi-attractor design, can achieve parallel refinement of multiple exploration directions, which helps to avoid premature homogenization and convergence of the population, and maintains the robustness and diversity of the population while continuing the search.

[0031] Step 112: Perform boundary constraint processing on all particles and recalculate their fitness. Update the individual historical optimal solution and the global optimal solution. Synchronize the position of the prophet particle with better fitness to the corresponding follower particle. Iterate and update the particle population. Stop the iteration when the preset iteration termination condition is met. Output the wireless sensor coverage deployment scheme corresponding to the global optimal solution.

[0032] Boundary constraint handling refers to correcting the positions of particles that exceed the deployment range, ensuring they always remain within the legal solution space. The individual historical optimal solution refers to the position with the highest fitness found by a single follower particle during the iteration process, recording the particle's own historical best experience. The global optimal solution refers to the position with the highest fitness found by the entire follower particle swarm during the iteration process, representing the swarm's current optimal deployment plan. The iteration termination condition refers to the rule for stopping iterative evolution, typically reaching a preset maximum number of iterations or the optimal solution showing no improvement for multiple consecutive rounds.

[0033] It is understandable that this step, through boundary verification, iterative updates of the optimal solution, and information transfer between roles, can form a closed-loop information feedback for exploration and development. This is conducive to efficiently transforming the exploration results of the prophet particles into the overall optimization benefits of the group, and ultimately outputting a wireless sensor coverage deployment scheme with the best comprehensive performance.

[0034] In the aforementioned coverage deployment method based on networked heterogeneous particle swarm optimization, by dividing the particle swarm into two heterogeneous roles—prophet particles and follower particles—structural decoupling of global exploration and local development can be achieved, overcoming the inherent limitations of homogeneous particle swarm algorithms in balancing exploration and development. By configuring a multi-layered search mechanism and a stagnation-triggered reset strategy for prophet particles, accurate local optimal escape can be achieved in non-convex coverage fitness landscapes, reducing the computational cost of ineffective iterations. By setting a dual-mode motion with probability switching and a multi-attractor guidance method for follower particles, a balance between the diversity and stability of the swarm search can be maintained, avoiding the risk of premature homogeneous convergence. By constructing a multi-index weighted coverage optimization model and establishing information feedback paths between roles, the coverage requirements of differentiated priority areas can be coordinated, improving the efficiency of wireless sensor resource utilization. This invention can simultaneously improve the overall coverage quality and optimization efficiency of deployment schemes in wireless sensor network coverage deployment scenarios with heterogeneous priorities, balancing the actual needs of dense sensing in key areas and wide-area coverage in general areas with limited resources.

[0035] In step 104 above, a rigorous mathematical model is performed on the target problem—sensor deployment for UAV communication coverage—to obtain a coverage deployment optimization model. This model will be directly used as the fitness evaluation function for HNPSO. n A wireless sensor is deployed in a two-dimensional monitoring area A(R) 2 Within the space, area A is divided into two parts: a general monitoring area G and a key monitoring area K. K is typically a smaller rectangular area centered on A. The two areas differ significantly in monitoring priority: the key area has a much higher monitoring value than the general area.

[0036] Each sensor i Deployed at coordinates ( x i , y i At a location (r), there is a circular sensing area that can detect any target within the sensing range. The sensors are connected wirelessly with a communication radius of R, typically R > r to reflect typical hardware specifications (communication capability greater than sensing capability).

[0037] sensor i For any point in space p = ( x p , y p The detection capability is described using a binary sensing model:

[0038] in For sensors iTime p Euclidean distance.

[0039] This model abstracts physical layer details such as signal attenuation, multipath effects, and environmental occlusion, but accurately captures the first-order geometry of coverage optimization—the binary nature of whether or not coverage occurs. Computation based on this binary model has a linear complexity of O(n * |G|) (where |G| is the number of evaluation grid points), making it suitable as a fast fitness evaluation function for iterative evolution. For scenarios requiring higher precision, a probabilistic or terrain-aware model can be used instead, while the optimizer framework remains unchanged.

[0040] sensor i With sensors j The conditions for direct communication between them are:

[0041] The connectivity of a sensor is defined as: Conn(i) =

[0042] The full connectivity constraint requires that the connectivity of all sensors be at least 1, i.e., Conn(i) >= 1 for all i. This ensures that any two sensors can communicate via multi-hop paths, satisfying the basic network requirements for collaborative sensing and data fusion.

[0043] The following three complementary coverage quality metrics guide the multi-dimensional evaluation of optimization problems: (1) General area coverage ( The general monitoring area G is discretized into |G| uniform evaluation grid points. Under the binary sensing model, the joint condition for evaluation point p to be covered is that at least one sensor i satisfies d(i, p) <= r.

[0044]

[0045] (2) Coverage rate of key areas ( Similar definition, but calculated on the key monitoring area K. Since the area of ​​K is much smaller than G, achieving high... Sensors need to be centrally deployed in region K, potentially at the expense of... This creates a fundamental tension in coverage deployment at the cost of:

[0046] (3) Sensor resource utilization rate ( () Penalizes the waste of coverage caused by excessive overlap and encourages efficient spatial distribution:

[0047] in Let A be the set of coverage points of sensor i, and |A| be the total area of ​​the monitored region. Maximize Encourage the decentralized deployment of sensors to offset the prioritization of key areas. This leads to a tendency for sensors to cluster together.

[0048] The three indicators mentioned above are weighted and summed to form the final optimization objective function:

[0049] The constraints include an upper limit on the number of sensors, full connectivity constraints, and deployment range constraints.

[0050] The weights satisfy a + b + c = 1. In typical application scenarios of UAV communication coverage, based on the priority principle of tracking accuracy in key areas > surveillance breadth in general areas > resource utilization efficiency, the weights are set as follows: a = 0.2, b = 0.5, c = 0.3. That is, the contribution of improved coverage in key areas to the overall goal is b / a = 2.5 times that of general areas.

[0051] The optimization variable is the two-dimensional coordinates of n sensors. x i , y i The problem has 2n decision variables (30 dimensions when n=15). The search space is continuous but bounded, the fitness landscape is non-convex (due to the discontinuity generated by the binary perception model), and there are multiple local optimum platforms (generated by the geometric symmetry of the sensor arrangement and overlapping coverage areas).

[0052] In one embodiment, performing a neighborhood perturbation search along a direction away from a randomly selected reference prophet particle and a directional refinement search along a direction away from the current global optimal position for each prophet particle includes: randomly selecting a reference prophet particle different from the current particle from the prophet particle swarm; calculating the position deviation vector between the current prophet particle and the reference prophet particle; generating a first candidate position along a direction away from the reference prophet particle based on the position deviation vector and a first random perturbation coefficient; comparing the fitness of the first candidate position with the current prophet particle position, and updating the current prophet particle position if the first candidate position has a better fitness; calculating the updated position deviation vector between the current prophet particle and the current global optimal position, and generating a second candidate position along a direction away from the current global optimal position based on the position deviation vector and a second random perturbation coefficient; comparing the fitness of the second candidate position with the current prophet particle position, and updating the current prophet particle position if the second candidate position has a better fitness.

[0053] In one embodiment, if the fitness of the searched position does not improve, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches a preset migration threshold, the position of the current prophet particle is randomly reset. This includes: after completing a single iteration of neighborhood perturbation search and directional refinement search, comparing the fitness of the current prophet particle before and after the update; if the fitness of the updated position is not higher than the fitness of the original position, then the number of consecutive search failures of the current prophet particle is incremented by one; comparing the number of consecutive search failures with the preset migration threshold, and when the number of consecutive search failures reaches the preset migration threshold, randomly generating a new position within the deployment range to replace the position of the current prophet particle, and clearing the number of consecutive search failures.

[0054] Specifically, let the size of the Prophet Particles subpopulation be... m Each prophet particle i Maintenance location x i P (R) 2n (dimensional vector) and stag counter i In each iteration step, each prophet particle performs the following three layers of operations in sequence: Neighbor-guided local perturbation: By searching in directions away from companions, prophet particles are encouraged to disperse into areas unexplored by other prophets, maintaining spatial diversity in the search. The operation includes: (1) Prophet Particle i Randomly select another prophet particle from the prophet subgroup. j ( j ≠ i () as a reference neighbor, i.e., a reference prophet particle.

[0055] (2) Along the distance from the reference neighbor j Candidate positions are generated in the direction of:

[0056] The first random disturbance coefficient β 1 ~ U(0, 1). Calculate the position deviation vector between the current prophet particle and the reference prophet particle. Based on the position deviation vector and the first random perturbation coefficient, generate a first candidate position along the direction away from the reference neighbor j. Wherein the position deviation vector... From particles j Pointing to particles i The direction vector. Continue moving in this direction (plus...) (times this vector), causing the particle i away from companions j Thus exploring what has not yet been discovered j The space occupied. Random sampling within (0,1) makes the disturbance amplitude random.

[0057] (3) If the fitness F( x cand ) higher than the current position F( x i P If so, then accept the update. x i P = x cand Otherwise, the stag counter i Add 1.

[0058] Global optimum-guided refinement: This involves a structured exploration of the neighborhood of the global optimum. It's important to note that this doesn't involve directly approaching the global optimum, but rather using the global optimum as a reference point for a targeted and diversified search. The process includes: (1) Based on the updated current prophet particle position, calculate its position deviation vector from the current global optimal position. Then, based on the position deviation vector and the second random perturbation coefficient, generate a second candidate position along the direction away from the global optimal position. x gbest The direction generates the second candidate position:

[0059] The second random disturbance coefficient β 2 ~ U(0, 1). Searching with random step sizes in directions away from the global optimum complements the traditional strategy of searching in the direction towards the global optimum. This search away from the optimum is equivalent to exploring potential better solutions distributed along different dimensions near the global optimum—if the global optimum is a point on a locally optimal platform, then there may be better solutions outside that platform along some directions away from it.

[0060] (2) If the fitness of the second candidate position is higher than that of the current position, then accept the update.

[0061] Stagnation-triggered migration: Provides a soft restart capability for Seer particles that have not been improved for a long time—allowing them to jump out of their current search stagnation zone and be reseeded to a new random location without discarding valuable information accumulated by other Seer particles. The operation process includes: (1) Each prophet particle maintains a stag counter. i This records the number of consecutive failed iterations since the last successful update.

[0062] (2) If stag i >= S ( SIf the preset migration threshold is used (a value between 3 and 7 is recommended, with a typical value of 5), then migration is triggered: the position of the prophet particle i is randomly re-initialized to any position within the search space. (exist n Uniform random sampling within the deployment constraints of each sensor); clear the stag counter: stag i = 0 (3) Migration threshold S Managing the patient trade-off between exploration and development: S The smaller the size, the faster the prophet particle abandons the current area (aggressive exploration bias); S The larger the value, the longer the prophet particle remains refined in the current region (a conservative development bias). Experiments show... S Robust performance is available in the range of [3, 7] without the need for precise tuning.

[0063] The three-level operation of the prophet particle can be described by the following algorithmic pseudocode:

[0064] In this embodiment, by configuring the prophet particles, which undertake the global exploration function, with a three-layer progressive search rule of neighbor-guided local perturbation, global optimal guidance refinement, and stagnation-triggered migration, the search performance of the particle swarm can be simultaneously optimized from three dimensions: search breadth, search depth, and local optimal escape, thus overcoming the inherent limitations of homogeneous particle swarm algorithms in balancing exploration and development. The neighbor-guided distributed perturbation can drive prophet particles to diffuse into unsearched areas, maintaining the spatial search diversity of the swarm and avoiding the limitation of the search range caused by particle aggregation; the global optimal guidance targeted refinement can carry out neighborhood mining around the current optimal deployment scheme, improving the convergence efficiency of the search and strengthening the ability to deeply optimize high-quality solutions; the individual-level reset mechanism based on continuous failure counting can enable stagnant particles to autonomously leave the local optimal region without disrupting the overall search process of the swarm, reducing the computational resource consumption caused by ineffective iterations. Applying this mechanism to wireless sensor network coverage deployment scenarios with general and key monitoring areas, it can adapt to fitness landscapes with discontinuous gradients under a binary perception model. Within a limited number of iterations, it outputs a deployment scheme with superior overall coverage performance, simultaneously ensuring coverage accuracy in key areas and coverage breadth in general areas under conditions of limited sensor resources, thereby improving the overall perception efficiency of the wireless sensor network. In one embodiment, the step of selecting a random prophet particle as a social attraction target to update its own velocity and position in the foraging mode includes: randomly selecting a prophet particle from the prophet particle swarm as a social attraction target; generating an inertial velocity component based on the inertial weight and the historical velocity of the currently following particle; generating an individual cognitive velocity component based on the individual acceleration coefficient, a first random factor, and the deviation between the currently following particle and its own historical best position; generating a social guidance velocity component based on the social acceleration coefficient, a second random factor, and the positional deviation between the currently following particle and the selected prophet particle; superimposing the inertial velocity component, the individual cognitive velocity component, and the social guidance velocity component to obtain the updated velocity of the currently following particle; and adjusting the position of the currently following particle according to the updated velocity to complete the position update.

[0065] In one embodiment, the fitness-adaptive position update performed in the vigilance mode towards the centroid of the following particle swarm includes: calculating the centroid position of the following particle swarm and the ratio of the current following particle's fitness to the total fitness of the swarm; determining a first step length coefficient for moving towards the centroid based on the ratio; generating a first displacement towards the centroid based on the first step length coefficient and the positional deviation between the current following particle and the centroid; randomly selecting a companion particle from the following particle swarm; determining a second step length coefficient based on the fitness difference between the current following particle and the companion particle; generating a second displacement along the historical best position direction of the companion particle based on the second step length coefficient and a random direction factor; and superimposing the first displacement and the second displacement onto the current position of the current following particle to complete the position update.

[0066] Specifically, follower particles update their positions through adaptive dual-mode motion. Let the size of the follower particle subpopulation be... m Each following particle i Maintenance location x i F ,speed v i F and individual historical best position p best i In each iteration step, each following particle independently has a foraging probability. p f Compare random numbers in the interval [0,1] r Use ~U(0,1) to select the behavior pattern: Prophet-guided foraging mode (trigger condition: r < p f Preferred p f= 0.8): Follower particles select a prophet particle as their social attraction target and move towards that prophet's search area, thus tracking and refining the exploration results. The key difference from standard PSO is that the social attraction target is replaced by a randomly selected prophet particle instead of the global optimum. The operation process includes: (1) Randomly select a prophet particle from the prophet particle subpopulation. k As a target for social attraction.

[0067] (2) Move according to the improved velocity-position update rule:

[0068]

[0069] in w = 0.5 is the inertia weight. c 1 = 1.4, which is the individual acceleration coefficient. c 2 = 1.7 is the social acceleration factor. r 1, r 2~U(0,1) represent the first random factor and the second random factor, respectively. c 1. r 1. The deviation between the current particle position and its historical best position constitutes the individual cognitive velocity component, reflecting the guiding role of the particle's own historical search experience. c 2. r 2. The combination of the two, the positional deviation between the current particle and the selected prophet particle, constitutes the social guidance velocity component, reflecting the guiding role of the prophet particle's exploration results.

[0070] It is worth noting that in standard PSO, all particles are uniformly attracted to a single global optimum position. g best The information propagation direction is one-to-many, which easily leads to premature homogeneous convergence of the group in geometric space. In the heterogeneous architecture of this invention, different follower particles are attracted to different prophet particles with equal probability. Since the prophet particles are distributed in different regions of the search space through local perturbations guided by their neighbors, they constitute a set of candidate attractors distributed throughout the search space. Follower particles randomly attach to different prophets, naturally achieving parallel following of multiple exploration directions. This many-to-many information propagation mode effectively decomposes the diversity of the search and avoids the risk of premature convergence of single attractor systems. From a mathematical point of view, this is equivalent to randomly selecting the social learning target in each iteration, increasing the random exploration component of the strategy.

[0071] Safe Gathering Alert Mode (Triggering Conditions:) r >=p f Probability 1- p f = 0.2): Simulating sentinel behavior in natural populations—when the population faces threats or fitness landscape uncertainty, individuals move towards the population center to reduce risk. Simultaneously, differentiated clustering is achieved through fitness-modulated step size, preserving the positional advantage of high-fitness individuals. The operation process includes: (1) Calculate the current centroid position of the following particle subpopulation:

[0072] (2) Randomly select a companion particle k ( k ≠ i (This can be used as a learning reference.)

[0073] (3) Calculate the total fitness .

[0074] (4) Move towards the centroid, with the step size adaptively adjusted by its own fitness. The update rule is:

[0075] in r 3 , r 4 ~ U(0,1), adaptive step size coefficient a 1 and a 2 is defined as:

[0076]

[0077] in ε It is a small positive constant used to prevent the denominator from being zero. a 1 represents the first step length coefficient. a 2 is the second step size coefficient. This represents the fitness difference between the currently following particle and the randomly selected companion particle.

[0078] The physical meaning of the fitness modulation step size is explained below: a 1. Coefficient: The higher the fitness F_i of particle i, the larger a1 is, and the stronger the particle is attracted to the population center of mass. This reflects the natural law that well-nourished individuals cluster more closely at the population core—superior individuals protect and consolidate their acquired superior resource positions through aggregation. Individuals with lower fitness... a A smaller value (1) indicates a slower movement towards the centroid, giving the user a greater chance to continue searching near the current location.

[0079] a 2. Coefficients: Included Items ( As a random orientation factor (taking values ​​between -1 and 1), the sign of peer learning depends on the relative fitness. F i < F k When (the companion is superior to oneself). a The sign of the coefficient 2 causes the particle to move towards its companion. p best k Directional learning; when F i > F k When (one is superior to one's companions) a 2. Sign reversal, the particle actually moves away p best k This two-way mechanism achieves implicit diversity maintenance, preventing low-fit individuals from unilaterally following high-fit individuals.

[0080] In this embodiment, the foraging mode, through a velocity-position update mechanism that uses randomly selected prophet particles as social attraction targets, drives following particles to keep up with global exploration results and perform local refinement, achieving parallel optimization of multiple potentially high-quality deployment areas and avoiding search convergence caused by a single global optimum. In the vigilance mode, a centroid-based adaptive adjustment mechanism prevents excessive particle dispersion and avoids the risk of capturing deceptive local optima in non-convex fitness landscapes. Combined with fitness differences and peer learning rules with random directions, it maintains internal search differences during group aggregation, implicitly preserving population diversity and preventing premature homogeneous convergence. The two behavioral modes switch according to preset probabilities, maintaining active search intensity in most iteration cycles while supplementing with phased group structure adjustments, thus maintaining the stability and robustness of the group while continuously advancing the optimization process. Applying this dual-mode adaptive mechanism to wireless sensor network coverage deployment scenarios containing general and key monitoring areas can adapt to the complex optimization characteristics under a binary perception model, improving the overall coverage efficiency of the deployment scheme with limited computing resources, and balancing the actual needs of dense sensing in key areas and wide-area coverage in general areas.

[0081] In one embodiment, the step of performing boundary constraint processing on all particles and recalculating fitness, updating individual historical optimal solutions and global optimal solutions, and synchronizing the position of the prophet particle with better fitness to the corresponding follower particle includes: performing boundary verification on the position components of all particles, adjusting position components that are outside the deployment range to within the constraint range; recalculating the fitness corresponding to the adjusted positions of all particles according to the objective function; comparing the fitness of each follower particle's current position with its own historical optimal position, and retaining the position with better fitness as the new individual historical optimal solution; comparing the fitness of all follower particles, and retaining the position with the highest fitness as the new global optimal solution; and comparing the fitness of each prophet particle with the corresponding index follower particle according to a preset one-to-one correspondence index relationship, and if the prophet particle has better fitness, replacing the current position of the corresponding follower particle with the position of the prophet particle.

[0082] In this embodiment, in each iteration, after the search for the prophet particle and the movement of the follower particle are completed, the following information exchange and optimal update are performed: S1 (Boundary Constraint Processing): Check the position components of all particles. If a component exceeds the boundary of the search space, it is either clipped back to the boundary value (boundary absorption) or bounced back into the search space (boundary reflection).

[0083] S2 (Fitness Reassessment): Based on the updated particle positions, recalculate the fitness value (i.e., the value of the objective function F for the coverage deployment problem) for each particle.

[0084] S3 (Individual Historical Best Update): If the fitness value of the current following particle position is higher than its individual historical best fitness value, then replace it with the current position.

[0085] S4 (Global Optimal Update): If the fitness value of the current following particle position is higher than the global optimal fitness value, then replace it with the current position.

[0086] S5 (Prophet → Follower Information Injection): For each prophet particle, if its current fitness value is higher than that of the follower particle at the corresponding index, the prophet particle's position is copied to replace the follower particle's position. This operation ensures that high-quality solutions discovered independently by the prophet particle are not wasted, but are effectively injected into the follower population for refinement in subsequent iterations. From an information system perspective, this is a one-to-one feedforward information pathway from the exploration subsystem to the development subsystem.

[0087] The closed-loop structure of information flow: Iteration t In Chinese: Prophet search → Follow movement → Individual / global optimal update → Information injection (Prophet → Follower); Iteration t+1 In: The updated global optimumx gbest Used by the second-level operation of the prophet particle (global optimal guided refinement) → prophet search → ...

[0088] The globally optimal solution updated by the follow-up group can be fed back to the next iteration's prophet particle-oriented refinement stage, thus forming a closed-loop optimization chain of exploration and discovery, development and refinement, and feedback updates, continuously driving performance improvements in the deployment scheme. This forms the process of exploration and discovery (prophet) → development and refinement (follower) → feedback updates (optimal updates) → guiding a new round of exploration (prophet second layer uses new...) g best A complete information loop.

[0089] In one embodiment, the sum of the weights corresponding to the general monitoring area coverage rate, the key monitoring area coverage rate, and the wireless sensor resource utilization rate in the objective function is 1, and the weight of the key monitoring area coverage rate is higher than the weights of the general monitoring area coverage rate and the wireless sensor resource utilization rate.

[0090] In one embodiment, in the initial particle swarm, the number of prophet particles and follower particles are equal, and the total particle size is proportional to the decision dimension corresponding to the wireless sensor deployment.

[0091] In one embodiment, initializing the particle swarm containing prophet particles and follower particles in the search space includes: randomly generating the initial position of each particle within the deployment range, initializing the initial velocity of all follower particles to zero, and initializing the number of consecutive search failures of all prophet particles to zero; dividing the particle swarm into a set of prophet particles and a set of follower particles on an even basis, and assigning a corresponding role identifier to each particle; calculating the fitness corresponding to the initial position of all particles according to the objective function, setting the initial position of each follower particle as its individual historical best position, and setting the position of the follower particle with the highest fitness as the initial global best position.

[0092] In one specific embodiment, such as Figure 2 The diagram illustrates a coverage deployment framework based on networked heterogeneous particle swarm optimization. The core steps are as follows: S101: Population Initialization and Role Decomposition - Decompose the population into two types of roles: prophet particles and follower particles, and randomly initialize the position and parameters of all particles.

[0093] S102: Prophet Particle Multi-Layer Search – sequentially executes neighbor-guided local perturbation, global optimal guided refinement, and stagnation-triggered migration.

[0094] S103: Follows the particle's adaptive dual-mode movement - enters the prophet-guided foraging mode with probability, and enters the safe gathering and alert mode with probability 1-pf.

[0095] S104: Boundary Constraints and Optimal Updates—Boundary Handling, Fitness Re-evaluation, Individual and Global Optimal Updates, and Information Injection from Seer to Follower.

[0096] S105: Iterative Evolution - Determine if the maximum number of iterations has been reached. If not, return to S102; otherwise, output the global optimal solution.

[0097] The algorithm performance can be optimized in a balanced way through the following exemplary parameter configurations. Each parameter can be flexibly adjusted according to the needs of the actual wireless sensor deployment scenario.

[0098] Group size m The size of the subpopulations of prophet particles and follower particles can be configured according to the decision dimensions of the deployment problem. Both subpopulations are equal in size, and the total population size is the sum of the sizes of the two subpopulations. For a deployment scenario involving 15 wireless sensors and corresponding to 30 decision dimensions, an exemplary configuration is a single subpopulation size of 100 and a total population size of 200, achieving a balance between solution quality and computational efficiency. For larger deployment problems, the ratio of single subpopulation size to decision dimensions can be linearly adjusted within a range of 3:1 to 5:1.

[0099] Foraging mode trigger probability p f An exemplary value of 0.8 for the foraging mode trigger probability is used, which controls the execution ratio of the two behavior modes: on average, in each iteration step, approximately 80% of the follower particles enter the oracle-guided foraging mode to actively search, while approximately 20% of the follower particles enter the safe gathering vigilance mode to adjust the group structure. Practical applications show that when this probability is 0.5, an excessively high group vigilance ratio leads to a 12.7% decrease in overall performance; when this probability is 1.0, completely canceling the vigilance mode leads to a 3.8% decrease in overall performance. When this probability is between 0.7 and 0.9, the algorithm performance is stable, requiring no high-precision parameter tuning.

[0100] Migration Threshold S: The exemplary range of the migration threshold is 3 to 7, with a typical value of 5. A migration threshold that is too small will cause particles to reset frequently, failing to complete an effective local refinement search; a migration threshold that is too large will cause particles to stagnate for too long, failing to escape the local optimum in time. When the migration threshold is 5, the prophet particle performs a position reset only after several consecutive rounds of searching without achieving fitness improvement, ensuring sufficient local search time while also allowing it to escape the stagnation region in time.

[0101] Inertia weight w With acceleration coefficient c 1 、c2: The exemplary values ​​for inertia weight, individual acceleration coefficient, and social acceleration coefficient are 0.5, 1.4, and 1.7, respectively. Compared with the typical parameter configuration of conventional particle swarm optimization algorithms, this scheme uses a lower inertia weight and an asymmetric acceleration coefficient. The lower inertia weight allows the movement of follower particles to rely more on the guidance of the prophet and their own historical best experience; the asymmetric setting of the social acceleration coefficient being greater than the individual acceleration coefficient allows the social guidance weight of the prophet particles to be higher than the cognitive weight of the particles themselves, which fits the architectural design of heterogeneous role division and strengthens the exploration and guidance role of the prophet group.

[0102] Coverage indicator weights a, b, c The weights for general monitoring area coverage, key monitoring area coverage, and sensor resource utilization are 0.2, 0.5, and 0.3 respectively, with the sum of all weights being 1. This corresponds to a priority configuration where tracking accuracy in key areas > resource utilization efficiency > monitoring breadth in general areas. The weight allocation can be flexibly adjusted according to the coverage priority of the actual task. For example, in a wide-area monitoring scenario, the weight corresponding to general monitoring area coverage can be increased.

[0103] like Figure 3 As shown, a schematic diagram of the area division and sensor deployment results for a coverage deployment scenario is provided, intuitively presenting the spatial distribution characteristics of the final deployment scheme output by this optimization method. The X and Y axes together constitute a two-dimensional monitoring area with coordinates ranging from [-10, 10]. The Sensor position represents the final deployment coordinates of each wireless sensor, i.e., the sensor position corresponding to the globally optimal solution output by the algorithm. The sensors are evenly distributed throughout the monitoring area, yet relatively concentrated in the central area, reflecting the optimized distribution characteristics that balance wide-area coverage and key area enhancement. Sensingrange, r=3 represents the circular sensing range of a single wireless sensor, with a sensing radius r=3, completely corresponding to the binary sensing model: the area covered by the circle is the range within which the sensor can effectively detect targets, and the overlapping area of ​​the circles is the area repeatedly covered by multiple sensors. Overlapping coverage can improve the detection reliability of key areas, while excessive overlap is constrained by the resource utilization index. The Key area represents the key monitoring area with higher monitoring priority, located at the center of the entire monitoring area, corresponding to the key monitoring area K in the scheme. As can be seen from the figure, this area is covered by the overlapping sensing ranges of multiple sensors, demonstrating that the optimization model guides sensors towards key areas with higher weights, ensuring the coverage accuracy of the core area.

[0104] In this embodiment, the particle swarm is functionally divided according to its search function, constructing a heterogeneous role system of two types of particles with differentiated behavioral rules: prophet particles and follower particles. This breaks the homogeneous behavior limitation of the standard particle swarm algorithm and achieves structural decoupling between global exploration and local development. Prophet particles undertake the function of guiding global exploration. Through a three-layer progressive search mechanism of neighbor-guided local perturbation, global optimal guidance refinement, and stagnation-triggered migration, they continuously search for potential high-quality regions in the solution space. This mechanism can maintain the diversity of the search space and conduct neighborhood mining around elite solutions through distributed perturbation, and can also achieve precise local optimal escape through an individual-level position reset mechanism. It can jump out of the stagnation region without destroying the search information accumulated by the swarm. Experimental data shows that canceling this migration mechanism will lead to a significant performance decline, and this mechanism can maintain stable optimization effects within a wide range of parameter values.

[0105] The following particles undertake the function of local adaptive refinement, and adopt a dual-mode movement rule of probabilistically switchable prophet-guided foraging and safe gathering vigilance. In the foraging mode, the randomly selected prophet particles perform position updates as multiple attractors, which can promote the deep optimization of multiple exploration directions in parallel and fundamentally reduce the risk of premature homogenization convergence of the group. In the vigilance mode, the particles perform fitness-adaptive gathering movement towards the center of mass of the group, which can maintain the stability of the group structure while preserving the internal search differences. The two modes work together to ensure the robustness of the group while continuously advancing the search process. The overall optimization benefit generated by the dual-mode collaboration is better than the sum of the effects of the two modes acting independently.

[0106] Based on this, the solution constructs a coverage optimization model that weights three indicators: general area coverage rate, key area coverage rate, and resource utilization rate. Combined with three types of constraints, namely, upper limit of the number of sensors, full connectivity, and deployment range, it can transform the coverage requirements with differentiated priorities into a mathematical problem that can be quantified and solved. Under limited wireless sensor resources, it can automatically tilt the deployment towards high-priority areas while ensuring network communication connectivity and data interaction capabilities.

[0107] Overall, the method of this invention can adapt to fitness landscapes with discontinuous gradients under binary perception models. In wireless sensor network coverage deployment scenarios with multi-level monitoring priorities, it can simultaneously improve the overall coverage quality and optimization efficiency of the deployment scheme. Each core design has a clear positive gain on the algorithm performance.

[0108] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0109] In one embodiment, a coverage deployment apparatus based on networked heterogeneous particle swarm optimization is provided, comprising: The parameter acquisition module is used to acquire the task parameters corresponding to the wireless sensor coverage deployment task; the task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of wireless sensors, and the sensing radius of each wireless sensor. The model building module is used to build a coverage deployment optimization model based on the task parameters. The coverage deployment optimization model takes the weighted sum of the coverage rate of general monitoring areas, the coverage rate of key monitoring areas and the utilization rate of wireless sensor resources as the objective function, and the upper limit constraint of the number of wireless sensors, the full connectivity constraint and the deployment range constraint as the constraints. An initialization module is used to initialize a particle swarm containing prophet particles and follower particles in the search space, where the position of each particle represents the distribution position of each wireless sensor in the monitored area. The particle search module is used to perform a neighborhood perturbation search along the direction away from the randomly selected reference prophet particle for each prophet particle, and to perform a directional refinement search along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches the preset migration threshold, the position of the current prophet particle is randomly reset. The position update module is used to switch the behavior mode of each of the following particles according to a preset probability. In the foraging mode, it selects a random prophet particle as a social attraction target to update its own speed and position. In the vigilance mode, it performs a fitness-adaptive position update towards the centroid of the following particle group. The fitness is calculated by the objective function. The results output module is used to perform boundary constraint processing on all particles and recalculate their fitness, update the individual historical optimal solution and the global optimal solution, synchronize the position of the prophet particle with better fitness to the corresponding follower particle, iteratively update the particle population, stop the iteration when the preset iteration termination condition is met, and output the wireless sensor coverage deployment scheme corresponding to the global optimal solution.

[0110] Specific limitations regarding the coverage deployment device based on networked heterogeneous particle swarm optimization can be found in the limitations of the coverage deployment method based on networked heterogeneous particle swarm optimization mentioned above, and will not be repeated here. Each module in the aforementioned coverage deployment device based on networked heterogeneous particle swarm optimization can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A coverage deployment method based on networked heterogeneous particle swarm optimization, characterized in that, The method includes: Obtain the task parameters corresponding to the wireless sensor coverage deployment task; the task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of wireless sensors, and the sensing radius of each wireless sensor; Based on the task parameters, a coverage deployment optimization model is constructed. The coverage deployment optimization model takes the weighted sum of the coverage rate of general monitoring areas, the coverage rate of key monitoring areas, and the utilization rate of wireless sensor resources as the objective function, and the upper limit constraint of the number of wireless sensors, the full connectivity constraint, and the deployment range constraint as the constraint conditions. A particle swarm containing prophet particles and follower particles is initialized in the search space, and the position of each particle represents the distribution position of each wireless sensor in the monitored area. For each prophet particle, a neighborhood perturbation search is performed along the direction away from the randomly selected reference prophet particle, and a directional refinement search is performed along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches a preset migration threshold, the position of the current prophet particle is randomly reset. For each of the following particles, the behavior mode is switched according to a preset probability. In the foraging mode, a random prophet particle is selected as a social attraction target to update its own speed and position. In the vigilance mode, a fitness-adaptive position update is performed towards the centroid of the following particle group. The fitness is calculated by the objective function. Boundary constraints are applied to all particles and their fitness is recalculated. The individual historical optimal solution and the global optimal solution are updated. The position of the prophet particle with better fitness is synchronized to the corresponding follower particle. The particle population is iteratively updated. The iteration stops when the preset iteration termination condition is met. The wireless sensor coverage deployment scheme corresponding to the global optimal solution is output.

2. The method according to claim 1, characterized in that, For each of the prophet particles, the process of performing a neighborhood perturbation search along a direction away from the randomly selected reference prophet particle, and a directional refinement search along a direction away from the current global optimal position, includes: Randomly select a reference prophet particle from the prophet particle swarm that is different from the current particle, and calculate the position deviation vector between the current prophet particle and the reference prophet particle; Based on the position deviation vector and the first random perturbation coefficient, a first candidate position is generated along the direction away from the reference prophet particle; Compare the fitness of the first candidate position with the current prophet particle position. If the first candidate position has a better fitness, then update the current prophet particle position. Calculate the updated position deviation vector between the current prophet particle and the current global optimal position, and generate a second candidate position along the direction away from the current global optimal position based on the position deviation vector and the second random perturbation coefficient; Compare the fitness of the second candidate position with the current prophet particle position. If the second candidate position has a better fitness, then update the current prophet particle position.

3. The method according to claim 1, characterized in that, If the fitness of the searched position does not improve, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches a preset migration threshold, the position of the current prophet particle is randomly reset, including: After a single iteration of neighborhood perturbation search and directional refinement search is completed, the fitness of the current prophet particle before and after the update is compared. If the fitness of the updated position is not higher than that of the original position, then the consecutive search failure count of the current prophet particle will be incremented by one. The number of consecutive search failures is compared with a preset migration threshold. When the number of consecutive search failures reaches the preset migration threshold, a new position is randomly generated within the deployment range to replace the current position of the prophet particle, and the number of consecutive search failures is reset to zero.

4. The method according to claim 1, characterized in that, The step of selecting random prophet particles as social attraction targets and updating one's own speed and position in foraging mode includes: Randomly select a prophet particle from the prophet particle swarm as a social attraction target; Based on the inertial weight and the historical velocity of the currently following particle, generate inertial velocity components; The individual cognitive velocity component is generated based on the individual acceleration coefficient, the first random factor, and the deviation between the current following particle and its own historical best position. The social guidance velocity component is generated based on the social acceleration coefficient, the second random factor, and the positional deviation between the current following particle and the selected prophet particle. The inertial velocity component, the individual cognitive velocity component, and the social guidance velocity component are superimposed to obtain the updated velocity of the current following particle; Adjust the position of the currently following particle based on the updated velocity to complete the position update.

5. The method according to claim 1, characterized in that, The step of performing fitness-adaptive position updates towards the centroid of the following particle population in vigilance mode includes: Calculate the centroid position of the following particle swarm, and the ratio of the current following particle's fitness to the total fitness of the swarm; The first step length coefficient for moving towards the center of mass is determined based on the ratio. Based on the first step length coefficient and the positional deviation between the current following particle and the center of mass, the first displacement in the direction of the center of mass is generated. Randomly select a companion particle from the following particle swarm, determine the second step size coefficient based on the fitness difference between the current following particle and the companion particle, and generate a second displacement along the direction of the companion particle's historical best position based on the second step size coefficient and the random direction factor. The first displacement and the second displacement are superimposed on the current position of the currently following particle to complete the position update.

6. The method according to claim 1, characterized in that, The process of applying boundary constraints to all particles and recalculating their fitness, updating the individual historical optimal solution and the global optimal solution, and synchronizing the position of the prophet particle with the better fitness to the corresponding follower particle includes: Perform boundary checks on the position components of all particles, and adjust the position components that are outside the deployment range to the constraint range; Based on the objective function, recalculate the fitness of all particles at their adjusted positions; Compare the fitness of each follower particle’s current position with its own historical best position, and retain the position with the better fitness as the new individual historical best solution; Compare the fitness of all following particles and retain the position with the highest fitness as the new global optimum; According to the preset one-to-one correspondence index relationship, the fitness of each prophet particle is compared with the fitness of the corresponding index follower particle. If the fitness of the prophet particle is better, the position of the prophet particle replaces the current position of the corresponding follower particle.

7. The method according to claim 1, characterized in that, In the objective function, the sum of the weights corresponding to the general monitoring area coverage rate, the key monitoring area coverage rate, and the wireless sensor resource utilization rate is 1, and the weight of the key monitoring area coverage rate is higher than the weights of the general monitoring area coverage rate and the wireless sensor resource utilization rate.

8. The method according to claim 1, characterized in that, In the initial particle swarm, the number of prophet particles and follower particles are equal, and the total particle size is proportional to the decision dimension corresponding to the wireless sensor deployment.

9. The method according to claim 1, characterized in that, The initialization of a particle swarm containing prophet particles and follower particles in the search space includes: The initial position of each particle is randomly generated within the deployment range, the initial velocity of all following particles is initialized to zero, and the number of consecutive search failures of all prophet particles is initialized to zero. The particle swarm is divided into a set of prophet particles and a set of follower particles, and each particle is assigned a corresponding role identifier. The fitness of all particles at their initial positions is calculated based on the objective function. The initial position of each following particle is set as its individual historical best position, and the position of the following particle with the highest fitness is set as the initial global best position.

10. A coverage deployment device based on networked heterogeneous particle swarm optimization, characterized in that, The device includes: The parameter acquisition module is used to acquire the task parameters corresponding to the wireless sensor coverage deployment task; the task parameters include the area range of the general monitoring area and the key monitoring area in the monitoring area, the number of wireless sensors, and the sensing radius of each wireless sensor. The model building module is used to build a coverage deployment optimization model based on the task parameters. The coverage deployment optimization model takes the weighted sum of the coverage rate of general monitoring areas, the coverage rate of key monitoring areas and the utilization rate of wireless sensor resources as the objective function, and the upper limit constraint of the number of wireless sensors, the full connectivity constraint and the deployment range constraint as the constraints. An initialization module is used to initialize a particle swarm containing prophet particles and follower particles in the search space, where the position of each particle represents the distribution position of each wireless sensor in the monitored area. The particle search module is used to perform a neighborhood perturbation search along the direction away from the randomly selected reference prophet particle for each prophet particle, and to perform a directional refinement search along the direction away from the current global optimal position. If the fitness of the position does not improve after the search, the number of consecutive search failures is accumulated. When the number of consecutive search failures reaches the preset migration threshold, the position of the current prophet particle is randomly reset. The position update module is used to switch the behavior mode of each of the following particles according to a preset probability. In the foraging mode, it selects a random prophet particle as a social attraction target to update its own speed and position. In the vigilance mode, it performs a fitness-adaptive position update towards the centroid of the following particle group. The fitness is calculated by the objective function. The results output module is used to perform boundary constraint processing on all particles and recalculate their fitness, update the individual historical optimal solution and the global optimal solution, synchronize the position of the prophet particle with better fitness to the corresponding follower particle, iteratively update the particle population, stop the iteration when the preset iteration termination condition is met, and output the wireless sensor coverage deployment scheme corresponding to the global optimal solution.

Images