A device deployment optimization method for different road condition perception coverage problems

Abstract

The application provides a device deployment optimization method for different road condition perception coverage problems, comprising the following steps: analyzing different road types, establishing a grid map library and a roadside device model library; in order to ensure the rationality and robustness of the optimization result, a two-stage method is adopted to carry out deployment optimization for sensors and MEC key devices; in the first stage, the NSGA-Ⅱ algorithm is used to carry out deployment design on the sensors in the RSU, so as to ensure the optimal configuration of the sensors in different road scenes; in the second stage, a hybrid expert model of clustering and model predictive control (MPC) is used to carry out deployment on the MEC devices, and the topological relationship between the roadside devices is established to adapt to the communication demand of the actual road scene; based on the obtained roadside device deployment scheme, the deployment in the actual scene is carried out. The application can carry out optimization deployment for various different scenes and sensor demands based on the optimization algorithm, and has high data transmission robustness.

Description

TECHNICAL FIELD

[0001] The application provides a device deployment optimization method for different road condition perception coverage problems, and belongs to the technical field of traffic road detection. BACKGROUND

[0002] Current road side device deployment is mostly based on artificial experience. According to the existing devices, on the one hand, it may not be able to meet the actual road requirements, and cannot achieve efficient coverage of the road, on the other hand, too much redundancy and experience-based installation will lead to high deployment cost. In order to solve this problem, an automatic and high-robustness deployment method is urgently needed in the current vehicle-road cooperation field.

[0003] For the problem of perception coverage, scholars have analyzed the two-dimensional perception model of different sensor types, and further defined the coverage overlap as the intersection of the fields of view of multiple vision sensors. If a point is captured by the field of view of at least two vision sensors, it is considered to be covered, for example, the method based on the art gallery problem selects to minimize the number of sensors while avoiding blind spots.

[0004] Road Side Unit (RSU) is an important hardware for current vehicle-road cooperation to carry out perception and other functions. Global RSU device deployment involves many considerations.

[0005] The prior art such as (Zhao, Y., Wang, Z., Sun, C. (2024). Deployment Optimization of Roadside Sensing Units Based on NSGA-II for Vehicle Infrastructure Cooperated Autonomous Driving. In: Jia, L., Easa, S., Qin, Y. (eds) Developments and Applications in Smart Rail, Traffic, and Transportation Engineering. ICSTTE 2023. Lecture Notes in Electrical Engineering, vol 1209. Springer, Singapore. https: / / doi.org / 10.1007 / 978-981-97-3682-9_30) is based on the investigation and research of current perception devices, based on the self-built sensor performance model, on the two-dimensional grid map of typical roads such as expressways and crossroads, using genetic algorithm, improved genetic algorithm (NSGA-Ⅱ), while meeting the given coverage rate and redundancy rate, further multi-objective optimization is carried out for the three target values of coverage rate, redundancy rate and cost. In order to further test the model algorithm, actual road scene experiments are carried out, the input is designed as the type of perception sensor, two-dimensional grid map and optimization algorithm basic parameters, and the output is designed as the Pareto solution set of perception device optimization deployment. The technology has the following shortcomings:

[0006] 1. Only the deployment of roadside sensors is involved, without considering the deployment of computing devices and communication devices such as edge computing units (MEC), signal repeaters, etc. The overall architecture is not complete and does not conform to the actual scene.

[0007] 2. The application scenario of roadside sensor algorithm is poor and cannot completely cover the needs of current highway and urban scenarios.

[0008] 3. The topology relationship between the position of the sensor in the optimization result and the signal transmission is not further considered.

[0009] This paper (Tong, X.; Li, M.; Cui, Z. A Hybrid Traffic Sensor Deployment Model with Communication Consideration for Highways. Appl. Sci. 2024, 14, 536. https: / / doi.org / 10.3390 / app14020536) presents an optimization model for hybrid sensor deployment on highways, aiming to improve the accuracy of traffic monitoring and consider communication needs. The model uses 0-1 programming techniques to integrate sensor deployment and server placement problems together and is solved by a two-step search algorithm. The following shortcomings exist in this technology:

[0010] 1. The working conditions studied by this algorithm are concentrated on highways, lacking consideration of urban roads;

[0011] 2. The highway deployment algorithm focuses on one-dimensional coordinate deployment, only considering the spacing between sensors, and the algorithm framework formed cannot be applied to two-dimensional device deployment in urban environments;

[0012] 3. The sensors studied by this algorithm are limited to cameras and do not consider the differences in sensing performance and parameters of multiple sensors, lacking consideration of hybrid sensors, and the robustness for different engineering needs is weak. SUMMARY

[0013] The technical problems to be solved by the present application are:

[0014] Technical problem one: traditional roadside sensing device deployment methods result in high total cost. The present application provides a more cost-effective deployment scheme based on a multi-expert model optimization algorithm.

[0015] Technical problem two: the distribution range of existing deployment methods is unreasonable. The present application realizes a more reasonable device distribution strategy based on an improved genetic algorithm and a hybrid expert model.

[0016] Technical problem three: subjective factors have a large impact on traditional methods, and the robustness is poor. The present application aims to reduce the impact of subjective factors on the deployment scheme and performs well in scenarios such as roundabouts, ramps, and the like.

[0017] Technical problem four: coverage range is difficult to quantify. The present application aims to propose a deployment method that can quantify the coverage range.

[0018] Technical problem five: existing solutions rarely start from the performance of the sensor itself and combine edge computing units to carry out roadside device deployment research. The present application aims to develop a deployment algorithm based on a sensor performance model in a multi-layer architecture.

[0019] The specific technical solution is:

[0020] A device deployment optimization method for different road condition perception coverage problems, comprising the following steps:

[0021] S1, analyze different road types, establish a grid map library and a roadside device model library;

[0022] S2, in order to ensure the rationality and robustness of the optimization result, a two-stage method is adopted to carry out deployment optimization for sensors and MEC key devices;

[0023] In the first stage, the NSGA-II algorithm is used to deploy the sensors in the RSU to ensure the optimal configuration of the sensors in different road scenes;

[0024] In the second stage, a hybrid expert model using clustering and model predictive control (MPC) is used to deploy MEC devices, and the topological relationship between roadside devices is established to adapt to the communication needs of actual road scenes;

[0025] S3, based on the obtained roadside device deployment scheme, carry out deployment in actual scene.

[0026] Further:

[0027] In S1, planning is carried out for different road types, sensor needs and cost budget;

[0028] First, select the road type, the scenarios covered by the current deployment algorithm include expressway, crossroad, T-junction, ramp and roundabout; after determining which of the above scenarios the current road surface to be deployed belongs to, obtain the high-definition map of the current road and two-dimensional grid, which is used for the calculation of Occupany occupancy grid in the subsequent optimization process;

[0029] Then, based on the current sensor needs and cost needs, set the selection range of the sensor and the upper limit of the total deployment number.

[0030] In S2, the first stage uses RSU deployment optimizer;

[0031] The first stage uses NSGA-II algorithm to optimize the position coordinates, perception radius, FOV, deployment cost of the sensors in the RSU, and obtains the optimal scheme of RSU deployment;

[0032] The optimization variables are set as: maximum perception distance of the sensor, FOV (including horizontal and vertical), deployment cost, lens orientation and maximum communication bandwidth;

[0033] Then, set the constraint conditions of the optimization process:

[0034] a ≥ 90% (1)

[0035] b ≥ 30% (2)

[0036]

[0037] wherein a represents the coverage rate, b represents the redundancy rate, represents the distance between the ith and jth sensors;

[0038] Set optimization variables:

[0039] Coverage rate coverage:

[0040]

[0041] wherein, is the weight corresponding to the ith grid in the map Q, is the coverage score corresponding to the ith grid in the map Q;

[0042] Then, the improved NSGA-II algorithm is applied to carry out the deployment design of the roadside RSU device.

[0043] In the population initialization module, a plurality of deployment schemes each composed of N RSU individuals are randomly generated according to a preset population size N, each deployment scheme is a population, has N RSUs, each RSU has a plurality of sensors, and the fitness of the deployment scheme is equal to the sum of the fitness of each RSU.

[0044] The individual fitness function value is the basis for comparison and elimination among individuals in the genetic algorithm population and is also the most critical content of the genetic algorithm. Different individuals are sorted according to the size of their individual fitness, and those at the tail end of the sequence are eliminated, so as to ensure that the algorithm converges in the direction of high overall fitness. The fitness of the population individual is:

[0045]

[0046] wherein B n is the individual fitness, S n is the new grid score covered by the current individual, G n is the new grid score of redundancy of the current individual, R n is the current device detection radius; the higher the fitness value, the higher the new grid score covered by the individual, which means that the net increase in grid score per unit area is higher, and thus the coverage performance is relatively more excellent. Finally, the total score of the grid map is taken as the fitness score of the population to participate in the external competition of the population, and the internal competition of the population is carried out according to B n ​

[0047] The population selection part adopts a tournament-based selection method, randomly selects two deployment schemes for comparison, and selects the deployment scheme with high fitness to generate offspring;

[0048] The process of generating offspring deployment schemes includes crossover and mutation; the crossover operation adopts single-point crossover, and randomly exchanges the types, coordinates and parameters of part of the in-RSU sensors in the two deployment schemes; the mutation operation adopts single-point mutation, and randomly modifies the types, coordinates and parameters of the in-RSU sensors in the deployment scheme, wherein the lower the fitness of the individual, the more likely the RSU is to mutate;

[0049] Finally, according to the fitness score of the obtained deployment scheme, the deployment schemes are sorted, and only the top 50 deployment schemes are retained; when the iteration number reaches the requirement, the optimal deployment scheme is output.

[0050] In S2, the first stage adopts the MEC deployment optimizer;

[0051] The second stage is based on a hybrid expert model. First, the weighted sum of the positions and communication bandwidths of the RSUs is used as the topological distance, and the K-mean clustering algorithm is used to carry out clustering to find the potential topological relationship between the RSUs. After setting the maximum number of clusters according to the number of RSUs, the loop starts. In the loop, the position coordinates, maximum communication bandwidth and communication coverage radius of the MEC in the current cluster are optimized according to the constraint conditions and initial values, the deployment cost is calculated, and the RSUs contained are indexed to form a topological structure. Finally, the deployment costs of different numbers of clusters are compared, and the cluster with the minimum cost is selected as the optimal solution output. This part is based on the MPC model predictive control algorithm, adds the K-mean clustering algorithm, designs the optimization target, limits the constraint conditions, and gives the initial conditions to carry out rolling optimization on the deployment algorithm. The algorithm inputs the optimal deployment scheme obtained in the last step, and the coordinates and maximum communication bandwidth of each RSU;

[0052] First, the K-means clustering scheme is used to find the relationship between each RSU; K-means is to find the centers of K clusters through iteration, so that the sum of the distances from each data point in the cluster to the cluster center is minimized; let V be a set of n observations, where each observation vi is a d-dimensional vector; the optimization target of the K-means algorithm is to minimize the following objective function:

[0053]

[0054] Where: k is the number of clusters; Z i is the i-th cluster, which contains all observations belonging to the cluster; u i is the center of the i-th cluster, calculated as the mean of all observations in the cluster; represents the square of the Euclidean distance between the observation and the cluster center;

[0055] For the coordinates and parameters of the RSU obtained in the last stage, the preset cluster class number is traversed from 1 to the maximum class number to find the internal topological relationship between RSUs; in each iteration, the optimal deployment scheme of MEC under the current class number is found; after traversing all solutions, the scheme with the minimum deployment cost is selected as the optimal solution output;

[0056] Next, a roadside MEC library is established; the maximum carrying bandwidth of MEC is set as the following array:

[0057] W s =[w s1 ,w s2 ,...] (7)

[0058] Next, the optimal solution (x, y, bandwith) obtained in the previous stage is taken as input:

[0059] P=[p1,p2,...,p n ] (8)

[0060] The deployment scheme of MEC when there are k RSU clusters is Zk:

[0061] Z k =[z k1 ,z k2 ,...,z kk ] (9)

[0062]

[0063] (1) Constraint condition

[0064] ① The total data amount of the RSU combination transmission in the current area should be lower than the upper limit of the MEC processing:

[0065]

[0066] Where i represents the ith cycle, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is an MEC at the unit u, F u represents the maximum carrying bandwidth of the MEC server at the unit u, and index represents the topological relationship between the RSU at position i and the MEC server at the unit u, where index contains one or more MEC indexes;

[0067] ② The MEC located in the jth cluster should satisfy:

[0068]

[0069] When the unit i is contained in the coverage of the server MEC on the jth unit u, its index is set to the id of the current RSU:

[0070]

[0071] where d is the distance between the RSU and the MEC server, R u is the coverage radius of the jth server MEC on the unit u;

[0072] ③The total cost of the RSU in a single area is lower than a fixed upper limit:

[0073]

[0074] where c k is the cost of each deployment scheme, and W is the cost of the MEC server deployment scheme;

[0075] ④The spatial deployment position of the RSU should be limited in a certain area to adapt to the actual scene needs:

[0076]

[0077] where i represents the ith loop, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is MEC at the unit u, F u represents the maximum carrying bandwidth of the MEC server at the unit u, and index represents the topological relationship between the RSU at position i and the MEC server at the unit u, where the index can contain one or more MEC indexes;

[0078] (3) Optimization goal

[0079] An optimal solution is obtained, which satisfies the constraint condition while minimizing the total cost of the RSU:

[0080] Q opt = min{Q(Z k )|Z k ∈ Z} (16)

[0081] where i represents the ith loop, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is MEC at the unit u, F u represents the maximum carrying bandwidth of the MEC server at the unit u, and index represents the topological relationship between the RSU at position i and the MEC server at the unit u, where the index can contain one or more MEC indexes;

[0082] (4) Output the properties of the solution of each cycle:

[0083] R k1 , w k1 , c k1 ;

[0084] …

[0085] R kj , w kj , c kj ;

[0086] …

[0087] R kk , w kk , c k

[0088] Wherein the first two columns are the position coordinates of the RSU in the jth class at the kth iteration, R kj is the communication coverage, w kj is the maximum carrying bandwidth, c kj is the deployment cost; for the kth cycle, there are Qk:

[0089]

[0090] According to the size of Qk, arrange in order, select the optimal solution corresponding to the smallest Qk in the iteration process as the final solution.

[0091] In S3, based on the RSU optimization deployment scheme obtained in the first stage and the second stage, joint deployment verification is carried out.

[0092] The deployment scheme selection carries out simulation in CARLA and MATLAB software, selects the sensor in CARLA as the sensor library to be deployed, selects the map and converts it into a grid map, and completes the input pretreatment.

[0093] Then, in MATLAB, deployment optimization is carried out, which successively passes through the RSU deployment optimizer and the MEC deployment optimizer; finally, the MEC deployment design is obtained.

[0094] The technical scheme of the application has the following technical effects:

[0095] 1. The application can carry out optimization deployment based on optimization algorithm for various scenes and sensor requirements, and the algorithm covers scenes such as highway, crossroads, T-junction, ramp and roundabout, and five road levels, and the sensor library contains various popular products in the market such as lidar, camera and millimeter wave radar.

[0096] 2. The application integrates the sensing characteristics of multiple sensors to carry out RSU deployment, and the coverage rate of the scene can be maintained above 90%;

[0097] 3. The application can establish a communication topology architecture between RSU and MEC devices end-to-end while designing the physical architecture of RSU and MEC, avoiding the excessive load of RSU at the transmission peak caused by a single architecture, and having high robustness of data transmission. BRIEF DESCRIPTION OF DRAWINGS

[0098] Figure 1 It is a schematic diagram of the overall framework of the deployment method of the application;

[0099] Figure 2 It is a schematic diagram of the overall architecture design of the application;

[0100] Figure 3 It is a simulation result diagram of the RSU deployment scheme in multiple scenarios of the embodiment;

[0101] Figure 4 It is a schematic diagram of the intersection device and architecture deployment of the embodiment

[0102] Figure 5 It is a simulation result diagram in multiple road scenarios of the embodiment. DETAILED DESCRIPTION

[0103] The specific technical solutions of the application are described in combination with the embodiments.

[0104] The overall framework of the deployment method of the application is shown in Figure 1 The overall architecture design is shown in Figure 2 Based on a two-stage hybrid expert optimization model, first, based on the current road environment and the actual needs of sensors, costs, etc., a two-dimensional grid map of the current road is generated, and the range of the currently used sensors and the range of the cost are set; second, input the above contents into the RSU deployment optimizer as known quantities to obtain the RSU deployment scheme of the current road; third, input the RSU deployment scheme into the MEC deployment optimizer to obtain the MEC deployment scheme of the current road; finally, based on the obtained RSU and MEC, carry out joint deployment and verify the optimization target.

[0105] During the research process, a variety of research methods and technical means were used:

[0106] S1, different road types are analyzed, and a grid map library and a roadside device model library are established;

[0107] S2, in order to ensure the rationality and robustness of the optimization results, a two-stage method is designed to carry out deployment optimization for key devices such as sensors and MECs. In the first stage, the NSGA-II algorithm is used to deploy sensors in RSU to ensure the optimal configuration of sensors in different road scenarios; in the second stage, a hybrid expert model combining clustering and model predictive control (MPC) is used to deploy MEC devices, and the topological relationship between roadside devices is established to adapt to the communication needs of actual road scenarios.

[0108] S3, based on the obtained roadside device deployment scheme, the deployment in the actual scene is carried out.

[0109] Specifically:

[0110] In S1, input preprocessing includes:

[0111] To design the device deployment scheme, it is necessary to plan for different road types, sensor needs and budget.

[0112] First, select the road type. The scenarios covered by the current deployment algorithm include expressway, crossroad, T-junction, ramp and roundabout. After determining which of the above scenarios the current road belongs to, the high-definition map of the current road is obtained through OpenStreet software or Gaode map and is two-dimensionally rasterized, which is used for the calculation of Occupany occupancy grid in the subsequent optimization process. The grid map can reveal the current characteristics and give the potential deployment position of the sensor and the area to be covered.

[0113] Then, based on the current sensor needs and cost needs, the selection range of the sensor and the upper limit of the total number of deployments are set. The most important part of RSU is the sensor. The sensor library established by the present application can flexibly add and shield the types of sensors according to the needs of the actual engineering scene, and can further find the target sensor with appropriate maximum sensing radius and field of view (FOV) according to the needs. The cost limit can be used as a constraint condition for the optimization process, which can ensure that the optimization process will not output a scheme that exceeds the actual engineering needs. The types of sensors and their parameters are shown in Table 1:

[0114] Table 1 Device types and related parameters

[0115]

[0116] For example, for the intersection scene, the pure vision-based algorithm has a higher demand for visual cameras, fisheye cameras, etc. For the highway, the millimeter wave radar has higher advantages in speed measurement. The upper limit of the cost limits the current deployment scale, ensures that the cost is within a reasonable range and can be further optimized. Based on the current road situation, the application regenerates a new sensor device library for the next optimizer to carry out optimization design, and the upper limit of the cost will be used as a constraint condition in the optimization process.

[0117] In S2, the RSU deployment optimizer is used in the first stage, and the specific method is:

[0118] The first stage is based on the NSGA-II algorithm, and the coverage rate and cost are used as optimization objectives. The position coordinates of the sensors in the RSU, the sensing radius, the FOV, and the deployment cost are used for deployment and optimization design, and the optimal scheme of the RSU deployment is obtained.

[0119] In view of some parameters of the sensors, such as the number of point clouds generated per second, the reflection intensity, or the image capture period, the industry often sets the parameters according to the inherent parameters, and therefore the optimization variable settings of the application are: the maximum sensing distance of the sensor, the FOV (including the horizontal and vertical), the deployment cost, the lens orientation, and the maximum communication bandwidth, as shown in Table 2:

[0120] Table 2 Optimal solution data structure

[0121]

[0122] Then, the constraint conditions of the optimization process are set:

[0123] α≥90% (1)

[0124] β≥30% (2)

[0125]

[0126] Wherein α represents the coverage rate, and β represents the redundancy rate, represents the distance between the i th and j th sensors

[0127] The optimization variables are set:

[0128] The coverage rate coverage is:

[0129]

[0130] Wherein, is the weight corresponding to the i th grid in the map Q, is the coverage score corresponding to the i th grid in the map Q, and the coverage score is calculated as shown in Table 3:

[0131] Table 3 Grid score algorithm

[0132]

[0133] Then the improved NSGA-II algorithm is applied to carry out the deployment design of the roadside RSU device. The corresponding road grid information has been obtained through the processing of the map information. The improved NSGA-II algorithm is developed from the genetic algorithm, and has population initialization, fitness function setting, population selection operation, population crossover operation, population mutation operation and population elimination operation.

[0134] In the population initialization module, according to the preset population size N, a plurality of deployment schemes composed of N RSU individuals with the structure of table 3 are randomly generated, each deployment scheme is a population, has N RSUs, each RSU has a plurality of sensors, and the fitness of the deployment scheme is equal to the sum of the fitness of each RSU.

[0135] The individual fitness function value is the basis for comparison and elimination among individuals in the genetic algorithm population, and is also the most critical content of the genetic algorithm. Different individuals are sorted according to the size of their individual fitness, and those located at the tail end of the sequence are eliminated, so as to ensure that the algorithm converges to the direction with higher overall fitness. Here, the fitness of the population individual designed by the application is:

[0136]

[0137] Wherein B n is the individual fitness, S n is the grid score newly covered by the current individual, G n is the grid score of the current individual new redundancy, R n is the current device detection radius. The individual with higher fitness has higher grid score newly covered, which means that the net increase of grid score per unit area is higher, and thus the coverage performance is relatively more excellent. Finally, the total score of the grid map is taken as the fitness score of the population, participating in the external competition of the population, and the population is competed internally according to B n .

[0138] Similar to the traditional genetic algorithm, the application carries out the operations of selection, crossover, mutation and elimination. The population selection part adopts the selection method based on the tournament, randomly selects two deployment schemes for comparison, and the deployment scheme with higher fitness is selected to generate offspring.

[0139] The process of generating offspring deployment schemes includes crossover and mutation. The crossover operation adopts single-point crossover, randomly exchanges the types, coordinates and parameters of the sensors in some RSUs in two deployment schemes; the mutation operation adopts single-point mutation, randomly modifies the types, coordinates and parameters of the sensors in the RSUs in the deployment scheme, and the lower the fitness of an individual RSU, the more likely it is to mutate. The above operations can discard the RSUs with poor deployment performance in the deployment scheme, and achieve relatively fast convergence. In addition, the patent selection introduces the concept of "foreign population" to avoid falling into the local optimum trap in the optimization process.

[0140] Finally, according to the fitness scores of the obtained deployment schemes, the deployment schemes are sorted, and only the top 50 deployment schemes are retained. When the iteration number reaches the requirement, the optimal deployment scheme is output. The simulation results are shown in Figure 3

[0141] In S2, the second stage adopts the MEC deployment optimizer, and the specific method is as follows:

[0142] The second stage is based on a hybrid expert model. First, the weighted sum of the positions and communication bandwidths of the RSUs is used as the topological distance, and the K-mean clustering algorithm is used to carry out clustering to find the potential topological relationship between the RSUs. After setting the maximum number of clusters according to the number of RSUs, the loop starts. In the loop, the position coordinates, maximum communication bandwidth and communication coverage radius of the MEC in the current cluster are optimized according to the constraint conditions and initial values, the deployment cost is calculated, the RSUs contained are indexed, and the topological structure is formed. Finally, the deployment costs of different numbers of clusters are compared, and the cluster with the minimum cost is selected as the optimal solution output. This part is based on the MPC model predictive control algorithm, adds the K-mean clustering algorithm, designs the optimization target, limits the constraint conditions, and gives the initial conditions to carry out rolling optimization on the deployment algorithm. The algorithm inputs the optimal deployment scheme obtained in the last step, mainly the coordinates and maximum communication bandwidth of each RSU.

[0143] First, the K-means clustering scheme is used to find the relationship between each RSU. The core idea of K-means is to find the centers of K clusters through iteration, so that the sum of the distances from each data point in a cluster to the center of the cluster is minimized. Let V be a set of n observations, where each observation vi is a d-dimensional vector. The optimization objective of the K-means algorithm is to minimize the following objective function:

[0144]

[0145] where: k is the number of clusters. Zi is the ith cluster, containing all observations belonging to the cluster. u i ​is the center of the ith cluster, usually calculated as the mean of all observations within the cluster. represents the squared Euclidean distance between an observation and the cluster center.

[0146] Here, the application aims to obtain the coordinates and parameters of RSU in the last stage, and find the inherent topological relationship between RSUs by presetting the number of clusters, from 1 to the maximum number of clusters. In each iteration, find the optimal deployment scheme of MEC under the current number of clusters. After traversing all solutions, the application selects the scheme with the minimum deployment cost as the optimal solution output.

[0147] Next, the roadside MEC library is established; the maximum carrying bandwidth of MEC is set to the following array:

[0148] W s =[w s1 ,w s2 ,...] (7)

[0149] Next, the optimal solution (x, y, bandwith) obtained in the previous stage is taken as input:

[0150] P=[p1,p2,...,p n ] (8)

[0151] The deployment scheme of MEC when there are k clusters of RSU is Zk:

[0152] Z k =[z k1 ,z k2 ,...,z kk ] (9)

[0153]

[0154] (1) Constraint condition

[0155] ① The total data amount of RSU combination transmission in the current area should be lower than the upper limit of MEC processing:

[0156]

[0157] Where i represents the ith cycle, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is a MEC at unit u, F u represents the maximum carrying bandwidth of the MEC server located at unit u, index represents the topological relationship between the RSU at position i and the MEC server located at unit u, here index contains one or more MEC indexes;

[0158] ② The MEC located in the jth cluster should satisfy:

[0159]

[0160] When the unit i is contained in the coverage of the server MEC on the jth unit u, its index is set to the id of the current RSU:

[0161]

[0162] Where d is the distance between the RSU and the MEC server, R u is the coverage radius of the jth server MEC on the unit u;

[0163] ③The total cost of RSU in a single area is lower than a fixed upper limit:

[0164]

[0165] Where c k is the cost of each deployment scheme, and W is the cost of the MEC server deployment scheme;

[0166] ④The spatial deployment position of the RSU should be limited in a certain area to adapt to the actual scene needs:

[0167]

[0168] Where i represents the ith loop, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is MEC at the unit u, F u represents the maximum carrying bandwidth of the MEC server at the unit u, and index represents the topological relationship between the RSU at position i and the MEC server at the unit u, where index contains one or more MEC indexes;

[0169] (3) Optimization goal

[0170] An optimal solution is obtained, which satisfies the constraint condition while minimizing the total cost of the RSU:

[0171] Q opt = min{Q(Z k )|Z k ∈ Z} (16)

[0172] Where i represents the ith loop, j represents the jth cluster, a k represents the bandwidth required by the kth RSU, Z u represents that there is MEC at the unit u, F urepresents the maximum carrying bandwidth of the MEC server located at the unit u, and index represents the topological relationship between the RSU at position i and the MEC server located at the unit u, where the index can contain one or more MEC indexes.

[0173] (4) The attributes of the solution of each loop are shown in Table 4:

[0174] Table 4 MEC optimal solution data structure

[0175]

[0176] where the first two columns are the position coordinates of the RSU in the jth class at the kth iteration, R kj is the communication coverage, w kj is the maximum carrying bandwidth, and c kj is the deployment cost. For the kth loop, there are Qk:

[0177]

[0178] According to the size of Qk, the optimal solution corresponding to the smallest Qk in the iteration process is selected as the final solution.

[0179] In S3, joint deployment is adopted:

[0180] Based on the RSU optimization deployment scheme obtained from the second part and the third part, joint deployment verification is carried out. The deployment scheme is selected to carry out simulation in CARLA and MATLAB software, the sensors in CARLA are selected as the sensor library to be deployed, and the map is selected and converted into a grid map to complete the input preprocessing. Then, the deployment optimization is carried out in MATLAB, which passes through the RSU deployment optimizer and the MEC deployment optimizer in turn. Finally, the MEC deployment design obtained by the application is as shown in Figure 4 .

[0181] The simulation results are shown in Figure 5 .

[0182] Among them, the green area is the covered area, the darker the color, the higher the coverage rate, the red point is the RSU deployment point, and the yellow is the MEC deployment position. It can be seen that under the RSU deployment, the coverage rate is higher than 90%, and the distance between the MEC and the RSU with topological relationship with itself is moderate, and the position is reasonable.

[0183] Table 5 Simulation results of joint deployment scheme

[0184] Type Coverage Fees Four-lane intersection 93.211％ 19.2189 Two-lane T-junction 93.603％ 14.5447 Two-lane roundabout 91.054％ 14.1350 Two-lane ramp 94.198％ 12.4341

[0185] The total cost is shown in Table 5.

Claims

1. A method for optimizing equipment deployment to address sensing coverage issues under different road conditions, characterized in that, Includes the following steps: S1, Analyze different road types and establish a grid map library and a roadside equipment model library; S2. To ensure the rationality and robustness of the optimization results, a two-stage approach is adopted to optimize the deployment of sensors and key MEC devices. In the first phase, the deployment design of sensors in the RSU was carried out based on the NSGA-II algorithm to ensure the optimal configuration of sensors in different road scenarios; The first phase uses the RSU deployment optimizer; The first stage is based on the NSGA-II algorithm, with coverage and cost as optimization objectives. It focuses on the deployment and optimization design of the sensor location coordinates, sensing radius, FOV and deployment cost in the RSU to obtain the optimal RSU deployment scheme. The optimization variables are set as follows: maximum sensing distance of the sensor, FOV, deployment cost, lens orientation, and maximum communication bandwidth; Next, set the constraints for the optimization process: （1） （2） （3） in Represents coverage. Represents redundancy rate. Indicates the first The and the first The spacing between the sensors; Set optimization variables: coverage: （4） in, For map The weight corresponding to the i-th grid cell, For map The coverage score corresponding to the i-th grid cell; Then, the improved NSGA-II algorithm was applied to carry out the deployment design of roadside RSU equipment; In the population initialization module, based on the preset population size N, several deployment schemes consisting of N RSU individuals are randomly generated. Each deployment scheme is a population with N RSUs. Each RSU has several sensors. The fitness of the deployment scheme is equal to the sum of the fitness of each RSU. The individual fitness function value is the basis for comparison and elimination among individuals in the genetic algorithm population, and it is also the most crucial aspect of the genetic algorithm. Different individuals are sorted according to their fitness values, and those at the end of the sequence are eliminated, thus ensuring that the algorithm converges towards the direction with the highest overall fitness. The fitness of an individual in the population is: （5） Among them B n For individual fitness, S n G represents the newly covered raster score for the current individual. n R represents the raster score of the current individual's new redundancy. n The current detection radius of the device is used; individuals with higher fitness values ​​have higher newly covered raster scores, indicating a higher net increase in raster scores per unit area, thus demonstrating superior coverage performance; finally, the total score of the raster map is taken as the fitness score of the population, participating in external competition, while internal competition is based on B. n Competition occurs within the population; The population selection process employs a tournament-based selection method, randomly selecting two deployment schemes for comparison, and choosing the deployment scheme with higher fitness to produce offspring. The process of generating offspring deployment schemes includes crossover and mutation. The crossover operation uses single-point crossover, randomly swapping the types, coordinates, and parameters of some sensors within RSUs in two deployment schemes. The mutation operation uses single-point mutation, randomly modifying the types, coordinates, and parameters of sensors within RSUs in the deployment schemes. RSUs with lower individual fitness are more likely to mutate. Finally, based on the fitness scores of the obtained deployment schemes, the deployment schemes are sorted, and only the top 50 deployment schemes are retained; when the required number of iterations is reached, the optimal deployment scheme is output. In the second phase, a hybrid expert model of clustering and model predictive control MPC is used to deploy MEC devices and establish the topological relationship between roadside devices to adapt to the communication needs of actual road scenarios. S3, based on the obtained roadside equipment deployment plan, will be deployed in real-world scenarios.

2. The method for optimizing equipment deployment to address different road condition sensing coverage issues according to claim 1, characterized in that, S1 includes planning for different road types, sensor requirements, and cost budgets; First, the road type is selected. The scenarios covered by the current deployment algorithm include highways, intersections, T-junctions, ramps, and roundabouts. After determining which of the above scenarios the road to be deployed belongs to, a high-resolution map of the current road is obtained and rasterized in two dimensions, which is used to calculate the occupancy of the grid in the subsequent optimization process. Then, based on the current sensor requirements and cost constraints, set the range of sensor selection and the upper limit of the total number of sensors to be deployed.

3. The method for optimizing equipment deployment to address different road condition sensing coverage issues according to claim 1, characterized in that, In S2, the first phase uses the MEC deployment optimizer; The second stage, based on a hybrid expert model, first uses a K-means clustering algorithm to cluster RSUs based on the weighted sum of their location and communication bandwidth as the topological distance, identifying potential topological relationships between RSUs. After setting the maximum number of clusters based on the number of RSUs, a loop is initiated. During the loop, the location coordinates, maximum communication bandwidth, and communication coverage radius of the MECs in the current cluster are optimized according to constraints and initial values. Deployment costs are calculated, and the included RSUs are indexed to form the topology. Finally, the deployment costs of different numbers of clusters are compared, and the cluster with the lowest cost is output as the optimal solution. This part uses the MPC model predictive control algorithm as a foundation, incorporating K-means clustering. By designing optimization objectives, limiting constraints, and providing initial conditions, the deployment algorithm undergoes rolling optimization. The algorithm inputs the optimal deployment scheme obtained in the previous step, which consists of the coordinates of each RSU and the maximum communication bandwidth. First, the K-means clustering scheme is used to find the relationships between the various RSUs. K-means iteratively finds the centers of K clusters such that the sum of the distances from each data point in a cluster to the cluster center is minimized. Let V be a set containing n observations, where each observation vi is a d-dimensional vector. The optimization objective of the K-means algorithm is to minimize the following objective function: （6） Where: k is the number of clusters; Z i It is the i-th cluster, containing all observations belonging to that cluster; u i is the center of the i-th cluster, calculated as the mean of all observations within that cluster; represents the square of the Euclidean distance between the observation and the cluster center; Based on the coordinates and parameters of the RSU obtained in the previous stage, the number of clusters is preset and iterated from 1 to the maximum number of clusters to find the inherent topological relationship between the RSUs; in each iteration, the optimal deployment scheme of MEC under the current number of clusters is found; after traversing all solutions, the scheme with the minimum deployment cost is selected as the optimal solution and output. Next, establish the roadside MEC library; set the maximum carrying bandwidth of the MEC to the following array: （7） Next, the optimal solution (x, y, bandwith) obtained in the previous stage is used as input: （8） The deployment scheme of MEC with k types of RSU clusters is Z. k: （9） （10） (1) Constraints ① The total amount of data transmitted by RSUs in the current area must be lower than the upper limit that the MEC can handle: （11） Where i represents the i-th iteration and j represents the j-th cluster. Z represents the bandwidth required for the k-th RSU. u This indicates that an MEC, F exists at cell u. u This indicates the maximum bandwidth carried by the MEC server located at cell u. The index indicates the topological relationship between the RSU at location i and the MEC server located at cell u. Here, the index contains one or more MEC indices. ② The MEC located in the j-th cluster should satisfy: （12） When unit i is included within the coverage of the server MEC on the j-th unit u, its index is set to the id of the current RSU: （13） Where d is the distance between the RSU and the MEC server, R u Let be the coverage radius of the server MEC on the j-th middle unit u; ③ The total cost of RSUs within a single region must be below a fixed upper limit: （14） Among them, c k The cost for each deployment scheme, where W is the cost of the MEC server deployment scheme; ④ The spatial deployment location of RSUs should be restricted to a certain area to adapt to the needs of the actual scenario: （15） (2) Optimization Objective Find an optimal solution that satisfies the constraints while minimizing the total cost of RSU: （16） (3) Output the properties of the solution for each iteration: 、 、 、 、 ； … 、 、 、 、 ； … 、 、 、 、 ； The first two columns represent the position coordinates of the RSU in the j-th class during the k-th iteration. kj For communication coverage, w kj For the maximum carrying bandwidth, c kj For deployment costs; for the k-th iteration, we have Qk: （17） （18） （19） According to Q k Arrange the values ​​in order of size, and select the smallest Q during the iteration process. k The corresponding optimal solution is taken as the final solution.

4. A method for optimizing equipment deployment to address different road condition sensing coverage issues, as described in any one of claims 1 to 3, characterized in that, In S3, joint deployment verification is carried out based on the RSU optimized deployment scheme obtained in the first and second phases; The deployment scheme was simulated in CARLA and MATLAB software. The sensors in CARLA were selected as the sensor library to be deployed, and the map was selected and converted into a raster map to complete the input preprocessing. Next, deployment optimization was carried out in MATLAB, going through the RSU deployment optimizer and the MEC deployment optimizer; finally, the MEC deployment design was obtained.

Images