Emergency degree sensitive opportunity type crowd sensing task allocation method and equipment

Abstract

The invention discloses an emergency degree sensitive opportunistic crowd sensing task allocation method and equipment, and belongs to the technical field of wireless sensor networks. A shortest path algorithm is used for correcting track deviation, user track data is matched to a main road network, then the main road network is divided into a group of sub-regions which are not overlapped with each other according to task requirements, user tracks and tasks are matched to corresponding sub-regions, a perception task optimization distribution model is established by considering the task emergency degree, and the task optimization distribution model is used for task optimization. The completion of emergency tasks is ensured while the utility of all sensing tasks is maximized; constructing a prediction model to predict a user track; and searching an optimal allocation scheme by using a genetic algorithm in combination with a user track prediction result and a perception task optimal allocation model considering the emergency degree. The method is simple in steps, effectively improves the utilization rate and task completion effectiveness of opportunity users, and provides an efficient and robust task allocation framework for scenes such as urban traffic real-time monitoring and environmental pollution dynamic perception.

Description

Technical Field

[0001] The present invention relates to a method and device for opportunistic crowd sensing task allocation sensitive to the degree of urgency, and belongs to the technical field of wireless sensor networks. Background Art

[0002] Opportunistic mobile crowd sensing uses widely used mobile intelligent terminals as sensing nodes to dynamically collect large-scale environmental data without the need for users to actively participate or change their daily behavior patterns. Through the characteristics of group device collaboration and data spatio-temporal complementarity, this technology provides decision-making support for urban management and public services with the advantages of low cost and high coverage.

[0003] There are still significant limitations in the task allocation level of existing opportunistic crowd sensing systems. First, in traditional task allocation models, all tasks are assigned the same or fixed priorities, lacking differential treatment of task urgency, importance, or timeliness, and no dynamic grading mechanism sensitive to task timeliness is established. When high-urgency tasks and low-urgency tasks compete for resources, high-urgency tasks are not given priority allocation; existing trajectory prediction models are mostly based on pure time-series data analysis, without integrating actual road network topology information, and the predicted paths are prone to deviate from the real traffic paths, resulting in low task allocation utility; for the problem of multi-constraint task allocation sensitive to the degree of urgency, existing algorithms have low search efficiency in considering the task allocation problem of the degree of urgency.

[0004] The prior art CN114546626A discloses an opportunistic crowd sensing online task allocation method based on user pre-grouping. Task requesters and users move randomly in the target environment. The allocation of sensing tasks and the return of sensing results can only be carried out when the task requesters and users directly meet spatially. The interaction of sensing tasks and results is carried out through short-range wireless communication methods, mainly including: a user pre-grouping method combining task load and encounter rules, and an in-group online task allocation method based on the first idle user first allocation and the maximum load task first allocation. The goal is to minimize the maximum expected completion time of all tasks of all task requesters. Its overall idea is simple and cannot handle complex situations. Summary of the Invention

[0005] Technical Problem: Aiming at the deficiencies in the prior art, a method and device for opportunistic crowd sensing task allocation sensitive to the degree of urgency are provided. The trajectories of users can be predicted in the real environment, and the task allocation plan can be planned in advance. By integrating trajectory prediction with road network constraints and an efficient global optimization algorithm, a "data-model-decision" closed loop is constructed to improve the execution success rate of key tasks and the utilization rate of system resources, and provide theoretical support for the deployment of urban-level crowd sensing systems.

[0006] Technical solution: For the above technical purpose, the present invention provides an opportunity-based crowd sensing task allocation method sensitive to the degree of urgency, including the following steps:

[0007] Obtain the historical trajectory data of the user and the backbone road data of the area where the user is located;

[0008] Use the shortest path algorithm to correct the trajectory deviation caused by GPS drift or signal interference, match the user trajectory data to the backbone road network, and restore the spatio-temporal continuity of the user's actual movement path;

[0009] Divide the backbone road network into a group of non-overlapping sub-regions according to the task requirements, and match the user trajectory and tasks to the corresponding sub-regions to reduce the complexity of task allocation;

[0010] For the difference in task timeliness, establish a perception task optimization allocation model considering the degree of task urgency, and ensure the completion of urgent tasks while maximizing the utility of all perception tasks;

[0011] Construct a prediction model based on the gated recurrent unit network to predict the user trajectory;

[0012] Combine the user trajectory prediction results with the perception task optimization allocation model considering the degree of urgency, and use the genetic algorithm to search for the optimal allocation scheme.

[0013] Furthermore, the process of obtaining the historical trajectory data of the user and the backbone road of the area where the user is located is as follows: Extract the backbone road network data in the sensing area, and call the intersections in the road network nodes; Perform uniform sampling of the user's position coordinates at equal time intervals to prepare for subsequent time series prediction;

[0014] Match the user trajectory data to the backbone road network. If the user's position coordinates are scattered around the backbone road network rather than on the backbone road network, then based on the distance between the user's position coordinates and the nodes of the backbone road network, map each position coordinate of the user to the nearest node in the backbone road network. When the straight line connecting the position nodes of the user at two adjacent moments cannot fit the backbone road network, use the Dijkstra algorithm to determine the shortest path between these two position nodes, and use this shortest path as the trajectory data of the user on the backbone road network.

[0015] Furthermore, the step of dividing the backbone road network into a group of non-overlapping sub-regions according to the task requirements is as follows: According to the extracted backbone road network data, determine the coverage range and regional boundaries of the backbone road network; Inside the backbone road network, divide the road network at equal intervals according to the task allocation requirements; Divide the sensing area L into a group of sub-regions l, denoted as L = {l1, l2, l3, …, l P}, where P is the total number of sub-regions, satisfying and (for all i ≠ j), and divide the users into the corresponding regions.

[0016] Furthermore, considering the differences in task timeliness, an optimization allocation model for sensing tasks is established by taking into account the urgency of tasks. While maximizing the utility of all sensing tasks, it meets the constraints and ensures the completion of urgent tasks. The specific steps are as follows:

[0017] Suppose there are M(t) sensing tasks to be allocated at time t, denoted as Ta(t) = {ta1, ta2, …, ta M(t)}, and the tolerable completion delay time for each task is different. The shorter the delay time, the higher the urgency of task completion. Let d i be the delay time of task ta i , then the urgency w i of task ta i is expressed as:

[0018]

[0019] When the delay time d i is within 5, the urgency w i takes 1; as d i increases, w i gradually decreases; when d i exceeds 15, w i tends to zero;

[0020] Suppose there are N(t) available users at time t, U(t) = {u1, u2,..., u N(t)}; denote the probability that user u k arrives in area l j at time t as pr jk (t). If pr jk (t) is greater than the threshold pr, then user uk 为 is an available mobile user in sub - area l j . The set of available mobile users in sub - area l j is Suppose the reward for user u k to execute task ta j in sub - area l i is r ijk = r b w i , r b is the compensation reward for task execution, and w i is the urgency value of task ta i . The higher w i , the greater the task compensation reward value;

[0021] When user u k is assigned multiple tasks, then the total reward obtained by u k is r k0 represents the basic reward that a user will have once hired;

[0022] The optimization objective of the sensing task optimal allocation model is to maximize the completion utility of all sensing tasks, which is defined as:

[0023]

[0024] where ut j (t) is the completion utility of all sensing tasks in sub-region l j :

[0025]

[0026] In the formula, pr jk (t) is the probability that user u k arrives at region l j at time t, predicted by a prediction model based on a gated recurrent unit network, w i represents the urgency of task ta i ; when task ta j in sub-region l i is assigned to the k-th mobile user u k , x ijk = 1, otherwise, x ijk = 0; when user u k is assigned to sub-region l j , then all tasks in l j will be executed by u k ;

[0027] The constraint conditions of the sensing task optimal allocation model include the following three points:

[0028] The sensing tasks in the j-th sub-region l j cannot be assigned to mobile users not belonging to its available mobile user set U j :

[0029]

[0030] The total execution cost of all tasks cannot exceed the budget B(t):

[0031]

[0032] The moving distance ds k of user u jk within a unit time period cannot exceed the distance threshold ds0:

[0033] ds jk x ijk ≤ ds0;

[0034] where ds jk is the moving distance of user u k going to sub-region l j .

[0035] Furthermore, the specific steps for constructing the prediction model based on the gated recurrent unit network include: constructing the user location data into a time series format, where each time series data point includes a historical timestamp and the corresponding node identification ID; constructing a three-layer neural network architecture including an embedding layer, a gated recurrent unit network layer, and a fully connected layer, where the first layer processing module contains a trainable embedding matrix, the second layer gated recurrent unit GRU structure captures context-related features, and the third layer processing module constructs a fully connected network with a non-linear activation function, ultimately realizing an end-to-end non-linear mapping from the original input space to the target prediction domain; using the trained prediction model to predict the test set data, and the output is the probability distribution of the user arriving at each sub-region in the time series.

[0036] Furthermore, based on the user trajectory prediction results, a genetic algorithm is used to search for the optimal allocation scheme. The specific steps include:

[0037] First, initialize the population Pop. At time t, sort all tasks according to the task urgency w i . For the task ta i with the highest urgency, confirm its location region l j . Based on the probability pr j (t) of each user at time t in region l jk output by the prediction model, assign ta i to the user u k with the highest probability; remove ta i from the task set, and repeat the previous step until all tasks are assigned. Then, an initial allocation scheme is obtained; to ensure the diversity of the initial population, the remaining initial allocation schemes are generated by randomly assigning users to each task; for the population Pop composed of the above initial allocation schemes, calculate the fitness value of each individual, i.e., the allocation scheme, according to the objective function ;

[0038] Perform selection, crossover, and mutation evolution operations on the population Pop:

[0039] 1) Adopt the tournament selection operation. Randomly select several individuals from the current population Pop to form a group, and select the individual with the optimal fitness as the parent. Repeat the above operation until enough parent individuals are selected from the population Pop to form a new parent population Pop l ;

[0040] 2) The crossover operation first selects from the parental population Pop l two individuals, denoted respectively as and Then, multiple crossover points are randomly selected, and a multi-point crossover method is used to exchange the allocation schemes of and to generate two new offspring individuals, denoted as off c and off d ;

[0041] 3) The mutation operation randomly selects a sub-region of allocated users from the offspring individual off c and replaces it with a user having a higher predicted probability and cost ratio. The same operation is performed on the offspring individual off d . The offspring individuals off c and off d are added to the offspring population Off;

[0042] Update the population Pop using Off based on the fitness value; determine whether the termination criterion is met. If not, continue to execute 1) - 3); otherwise, the algorithm terminates, and the feasible individual with the maximum fitness value in the population is output as the optimal task allocation scheme.

[0043] A computer device includes a processor and a memory. The processor is electrically connected to the memory. The memory is used to store instructions and data, and the processor is used to execute the emergency-degree-sensitive opportunistic crowd-sensing task allocation method.

[0044] A computer-readable storage medium stores a computer program, and this computer program is suitable for being loaded and executed by the processor to perform the emergency-degree-sensitive opportunistic crowd-sensing task allocation method.

[0045] Beneficial effects: This method proposes a task grading mechanism based on the delay sensitivity function, which improves the allocation priority of traffic accident tasks, shortens the response time, and effectively avoids the missed detection of emergency events caused by resource competition. This method integrates the gated recurrent unit prediction model of the backbone road network structure, and through the embedding layer, it integrates the connectivity of road network nodes to ensure that the task pre-allocation scheme conforms to the actual traffic logic. This method optimizes the search strategy, greatly accelerating the search efficiency of the allocation scheme; emergency tasks are processed first, and important tasks are directly allocated to the most suitable users at the start of the algorithm to prevent key tasks from being missed due to random allocation. This method significantly improves the utility of traffic event detection during peak hours by dynamically allocating road condition collection tasks to neighboring vehicles, providing minute-level decision support for the traffic management department. In addition, this framework can be extended to the fields of environmental pollution monitoring and public safety patrol, and can be quickly migrated only by adjusting the road network data and task weight parameters, having broad engineering application value. Description of the Drawings

[0046] Figure 1 It is a schematic flowchart of an embodiment of the present invention.

[0047] Figure 2 It is a schematic diagram of road network matching in an embodiment of the present invention.

[0048] Figure 3 It is a function graph of task urgency in an embodiment of the present invention.

[0049] Figure 4 It is a schematic diagram of a gated recurrent unit prediction model in an embodiment of the present invention.

[0050] Figure 5 It is a graph of the training result of the gated recurrent unit prediction model in an embodiment of the present invention.

[0051] Figure 6 It is a flowchart of an emergency - level - sensitive opportunistic crowd - sourced sensing task allocation in an embodiment of the present invention. Detailed implementation manners

[0052] The following further describes the embodiments of the present invention with reference to the accompanying drawings.

[0053] As Figure 1 shown, the present invention discloses an emergency - level - sensitive opportunistic crowd - sourced sensing task allocation method. This embodiment provides historical trajectory data of users and backbone road network data of the area where the users are located. The shortest path algorithm is used to match the user trajectory data to the backbone road network. The specific steps include:

[0054] Extract the backbone road network data in the sensing area, and call the intersections in the road network nodes, as Figure 2 shown.

[0055] Select an area in the northern part of the urban area of Beijing with a latitude of 39.936693, a southern latitude of 39.868402, an eastern longitude of 116.34499, and a western longitude of 116.258994, and extract the position coordinates of 50 users in chronological order.

[0056] If the user position coordinates are scattered around the backbone road network rather than on the road network, map the user position coordinates to the nearest node in the road network.

[0057] When the straight line connecting the position nodes of the user at two adjacent moments cannot fit the road network, use the Dijkstra algorithm to determine the shortest path that fits the road network between these two position nodes, and use this path as the corrected user trajectory data.

[0058] In the embodiment, the step S2 divides the backbone road network into regions. The specific steps include:

[0059] Define the boundary of the traffic monitoring area according to the coverage of the backbone road network.

[0060] According to the task scenario of traffic monitoring, divide the road network at intervals of 200 meters to obtain a number of sub-regions, denoted as L = {l1, l2, l3, …, l P}, where P is the total number of sub-regions, satisfying and (for all i ≠ j), and divide the users into the corresponding sub-regions.

[0061] In the embodiment, an optimization allocation model of sensing tasks is established considering the urgency level to maximize the completion utility of all sensing tasks. The specific steps include:

[0062] The tolerable completion delay time of each task is different. The shorter the delay time, the higher the urgency level of task completion. Denote d i as the delay time of task ta i , then the urgency level w i of task ta i is expressed as:

[0063]

[0064] where Figure 3 represents the function graph of the task urgency level. When the delay time d i is within 5, it is defined as a traffic accident task, and the urgency level w i takes 1; as d i increases, 5 < d i ≤ 10 is defined as a sudden congestion task, and w i gradually decreases; when d i exceeds 15, it is defined as a conventional traffic flow monitoring task, and w i tends to zero.

[0065] Suppose there are N(t) available users at time t, U(t) = {u1, u2,..., u N(t)}; denote the probability that user u k arrives at region l j at time t as pr jk (t). If pr jk (t) is greater than the threshold pr, then user u k is an available mobile user in sub-region l j , and the set of available mobile users in sub-region l j is Suppose the reward for user u k to execute task ta j in sub-region l i is r ijk = r b wi , r b is the compensation reward for task execution, w i for task ta i urgency value, w i The higher it is, the greater the task compensation reward value; when user u k is assigned multiple tasks, then u k total reward obtained r k0 represents the basic reward that a user has once hired;

[0066] The optimization objective of the mathematical model is to maximize the completion utility of all sensing tasks, which is defined as:

[0067]

[0068] where ut j (t) is the completion utility of all sensing tasks in sub-region l j :

[0069]

[0070] In the formula, pr jk (t) is the probability that user u k arrives at region l j at time t, predicted by a prediction model based on a gated recurrent unit network, w i represents the urgency of task ta i ; when the task ta j in sub-region l i is assigned to the k-th mobile user u k , x ijk = 1, otherwise, x ijk = 0; when user u k is assigned to sub-region l j , then all tasks in l j will be executed by u k ;

[0071] The constraint conditions include the following three points:

[0072] The sensing tasks in the j-th sub-region l j cannot be assigned to mobile users who do not belong to its available mobile user set U j :

[0073]

[0074] The total execution cost of all tasks shall not exceed the budget B(t):

[0075]

[0076] User u k The moving distance ds within a unit time period jk shall not exceed the distance threshold ds0:

[0077] ds jk x ijk ≤ds0

[0078] Wherein, ds jk is the moving distance of user u k going to sub-region l j .

[0079] Figure 4 This is a schematic diagram of the trajectory prediction model of the present invention. In this embodiment, the constructed prediction model based on the gated recurrent unit network predicts the user trajectory. The specific steps include:

[0080] Divide the data set into a training set, a validation set and a test set.

[0081] Construct a prediction model including an embedding layer, a gated recurrent unit network layer (GRU layer) and a fully connected layer. The embedding layer maps the input node ID and the corresponding longitude and latitude of the position into an embedding vector with a fixed dimension, and passes it together with the corresponding timestamp and user ID to the GRU layer. The GRU layer combines the above input with the hidden state of the previous moment, and obtains the hidden state h of the current moment through the update gate z t and the reset gate r t . The specific calculation formula is: t Specifically, the calculation formula is:

[0082] r t = sigmoid(w r ·[h t-1 , x t )

[0083] z t = sigmoid(w z ·[h t-1 , x t )

[0084]

[0085] Wherein, h t-1 , h t represent the hidden layer states at times t-1 and t respectively, is the candidate hidden state, and w r , w z , w o represent weights, sigmoid and tanh are activation functions, and x t is the input at time t.

[0086] h t Convert it into the predicted output y through the fully connected layer t , that is, the probability of the user reaching each sub-region

[0087] y t = softmax(w out ·h T + b out )

[0088] where w out , b out are the weight and bias terms of the fully connected layer respectively, and softmax is the activation function

[0089] To further implement the above technical solution, the training stage of the model includes the following steps, and finally it is tested on the test set

[0090] 1) Obtain training data

[0091] Collect the coordinates of 50 users, sample once every 1 minute, from 7:00 in the morning to 21:00 in the evening, and each user is sampled for 5 days, with a total of 206045 pieces of data. Among them, the first three days are used as the training set, and the last two days are used as the validation and test sets respectively

[0092] 2) Data preprocessing

[0093] Let the latitudes of the user location and the nodes in the sub-region be lat1 and lat2 respectively, and the longitudes be lon1 and lon2 respectively. Among them, both the latitude and longitude are converted to radians. The radius of the earth is R (usually taking the average value of 6371000 meters), and the Haversine formula is used to calculate the distance d between two points as

[0094]

[0095] When d < 200, the user is classified into this sub-region

[0096] 3) Evaluation index

[0097] Cross-entropy loss

[0098]

[0099] where P is the number of categories, and the predicted output of the model is a probability distribution pr j = (pr1, pr2, pr3,..., pr P ), pr i is the probability that the model predicts the sample belongs to the j-th category. y = (y1, y2, y3,..., y n ) is the true label of the sample, where y j= 1 indicates that the real sample belongs to the j-th class, and the rest of y j = 0.

[0100] 4) Result analysis.

[0101] To verify the feasibility of the proposed model, iterative training is performed 300 times. From Figure 5 the model training result graph, it can be seen that the training loss and validation loss gradually decrease, and with the increase of the number of iterations, the accuracy gradually improves. Finally, it is tested on the test set, and the accuracy rate reaches 73.89%.

[0102] See Figure 6 is the flow chart of the urgency-sensitive opportunistic crowd-sensing task allocation of the present invention. Based on the user trajectory prediction result, a genetic algorithm is used to search for the optimal allocation scheme. The specific steps include:

[0103] Individual encoding: At time t, there are M(t) sensing tasks to be allocated, located in P sub-regions. In addition, there are N(t) available users located in the P sub-regions. Then an individual can be represented by a [x ijk M(t)×N(t)×P three-dimensional binary matrix encoding matrix. When the task ta j in the sub-region l i is allocated to the k-th mobile user u k , x ijk = 1, otherwise, x ijk = 0.

[0104] Population Pop initialization: At time t, all tasks are sorted according to the task urgency w i . For the task ta i with the highest urgency, confirm the region l j where it is located. Based on the probability pr j (t) of each user in the region l jk at time t output by the prediction model, allocate ta i to the user u k with the highest probability. Remove ta i from the task set, and repeat the previous step until all tasks are allocated. Then an initial individual is obtained. In addition, to ensure the diversity of the initial population, the remaining individuals are generated by randomly allocating users to each task.

[0105] Calculate the fitness value of each individual in the population Pop according to the objective function .

[0106] Perform selection, crossover, and mutation evolution operations on the population Pop, that is, execute 1) - 3):

[0107] ​1) Tournament selection operation is adopted. Several individuals are randomly selected from the current population Pop to form a group, and the individual with the best fitness is selected as the parent. The above operation is repeated until enough parent individuals are selected from the population Pop to form a new parent population Pop l ;

[0108] 2) For the crossover operation, first two individuals are selected from the parent population Pop l , denoted as and respectively. Then multiple crossover points are randomly selected, and the allocation schemes of and are exchanged using the multi-point crossover method to generate two new offspring individuals, denoted as off c and off d .

[0109] 3) For the mutation operation, a user in a randomly selected sub-region of the offspring individual off c is replaced with a user having a higher ratio of predicted probability to cost. The same operation is performed on the offspring individual off d . The offspring individuals off c and off d are added to the offspring population Off;

[0110] The population Pop is updated based on the fitness value using Off.

[0111] It is judged whether the termination criterion is satisfied. If not, return to steps 1) - 3); otherwise, the algorithm terminates, and the feasible individual with the largest fitness value in the population is output as the optimal task allocation scheme.

[0112] The effects of the present invention are illustrated by the following simulation experiments:

[0113] For the example 50 - 1900 consisting of 50 users and 1900 tasks, the proposed genetic algorithm is used for solution, and the traditional genetic algorithm and the random allocation method are used as comparison algorithms. The population size of the genetic algorithm is set to 100, and the algorithm termination criterion is that the number of iterations reaches 300.

[0114] Table 1 Allocation Results lists the perceived utilities of the allocation schemes found by the three methods. It can be seen that the method proposed in the present invention obtains the largest perceived task completion utility in the example 50_1900, proving the effectiveness of the algorithm of the present invention.

[0115] Table 1

[0116] Optimized model Sample 50_1900 emergency task completion rate Urgency level 57.41％ Without considering the urgency level 48.51％

[0117] Table 2 Allocation Results lists the consideration of the delay time di In the case of traffic accident tasks within 5 and sudden congestion tasks of 5 < d i ≤ 10, when comparing the task completion rate with the case of not considering the urgency of the task, the model method proposed by the present invention obtains a relatively high task completion rate in the sample 50_1900, which proves the effectiveness of the model of the present invention.

[0118] Table 2

[0119] Algorithm Sample 50_1900 perceived utility Improved genetic algorithm proposed by the present invention 375.8752 Traditional genetic algorithm 310.6932 Random assignment 261.6345 。

[0120] A computer device includes a processor and a memory. The processor is electrically connected to the memory. The memory is used to store instructions and data, and the processor is used to execute the above-mentioned urgency-sensitive opportunistic crowd-sensing task allocation method.

[0121] A computer-readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to execute the above-mentioned urgency-sensitive opportunistic crowd-sensing task allocation method.

Claims

1. An opportunistic crowd sensing task allocation method sensitive to the urgency level, characterized in that It includes the following steps: Obtain the historical trajectory data of the user and the arterial road data of the area where the user is located; Use the shortest path algorithm to correct the trajectory deviation caused by GPS drift or signal interference, match the user trajectory data to the arterial road network, and restore the spatio-temporal continuity of the user's actual movement path; Divide the arterial road network into a group of non-overlapping sub-areas according to the task requirements, and match the user trajectory and tasks to the corresponding sub-areas to reduce the complexity of task allocation; In view of the difference in task timeliness, establish a perception task optimization allocation model considering the urgency of the task, and ensure the completion of urgent tasks while maximizing the utility of all perception tasks; Construct a prediction model based on the gated recurrent unit network to predict the user trajectory; Combine the user trajectory prediction results with the perception task optimization allocation model considering the urgency, and use the genetic algorithm to search for the optimal allocation scheme.

2. The method for opportunistic crowd sensing task allocation sensitive to the urgency level according to claim 1, wherein The process of obtaining the historical trajectory data of the user and the arterial road in the area where the user is located is as follows: extract the arterial road network data in the perception area, and call the intersections in the road network nodes; perform uniform sampling of the user's position coordinates at equal time intervals to prepare for subsequent time series prediction; Match the user trajectory data to the arterial road network. If the user's position coordinates are scattered around the arterial road network rather than on the arterial road network, then based on the distance between the user's position coordinates and the nodes of the arterial road network, map each position coordinate of the user to the nearest node in the arterial road network. When the straight line connecting the position nodes of the user at two adjacent moments cannot fit the arterial road network, use the Dijkstra algorithm to determine the shortest path between these two position nodes, and use this shortest path as the trajectory data of the user on the arterial road network.

3. The method for opportunistic crowd sensing task allocation sensitive to the urgency level according to claim 1, wherein: The steps for dividing the backbone road network into a non-overlapping set of sub-regions according to task requirements are as follows: Based on the extracted backbone road network data, determine the coverage range and regional boundaries of the backbone road network; within the backbone road network, equally divide the road network according to task allocation requirements; divide the sensing area L into a set of sub-regions l, denoted as L = {l1, l2, l3, …, l P}; P is the total number of sub-regions, satisfying and (for all i ≠ j), and divide the users into their respective regions.

4. The method for opportunistic crowd sensing task allocation sensitive to the urgency level according to claim 1, wherein: In view of the difference in task timeliness, establish a perception task optimization allocation model considering the urgency of the task, and satisfy the constraints and ensure the completion of urgent tasks while maximizing the utility of all perception tasks. The specific steps include: Let there be M(t) sensing tasks to be allocated at time t, denoted as Ta(t) = {ta1, ta2, …, ta M(t)}, and the tolerable completion delay times of each task are different. The shorter the delay time, the higher the urgency of task completion. Let d i be the delay time of task ta i , then the urgency w i of task ta i is expressed as: When the delay time d i is within 5, the urgency level w i is set to 1; as d i increases, w i gradually decreases; when d i exceeds 15, then w i tends to zero; At time t, there are N(t) available users U(t) = {u1, u2,..., u N(t)}; Denote the probability that user u k arrives in area l j at time t as pr jk (t). If pr jk (t) is greater than the threshold pr, then user u k is an available mobile user in sub - area l j . The set of available mobile users in sub - area l j is Suppose the reward for user u k to execute task ta j in sub - area l i is r ijk = r b w i . r b is the compensation reward for executing the task, and w i is the urgency value of task ta i . The higher w i , the greater the task compensation reward value; When user u k is assigned multiple tasks, then u k the total reward obtained r k0 represents the base reward that the user has once hired; The optimization objective of the perception task optimization allocation model is to maximize the completion utility of all perception tasks, which is defined as: where, ut j (t) is the completion utility of all sensing tasks in sub-region l j : where pr jk (t) is the probability that user u k reaches area l j at time t, predicted by a prediction model based on a gated recurrent unit network, w i represents the urgency level of task ta i ; when task ta j in sub - area l i is assigned to the k - th mobile user u k , x ijk = 1, otherwise, x ijk = 0; when user u k is assigned to sub - area l j , then all tasks in l j will be executed by u k . The constraints of the perception task optimization allocation model include the following three points: The j-th sub-region l j The sensing task in it cannot be assigned to the mobile users not belonging to its available mobile user set U j : among the mobile users The total execution cost of all tasks shall not exceed the budget B(t): User u k Moving distance ds within a unit time period jk Must not exceed the distance threshold ds0: ds jk x ijk ≤ds0; where ds jk is the moving distance of user u k going to sub-region l j .

5. The method for opportunistic crowd sensing task allocation sensitive to urgency according to claim 4, characterized in that: The specific steps of constructing a prediction model based on the gated recurrent unit network include: constructing the user position data into a time series format, where each time series data point includes the historical timestamp and the corresponding node identification ID; constructing a three-layer neural network architecture including an embedding layer, a gated recurrent unit network layer, and a fully connected layer. The first-layer processing module contains a trainable embedding matrix, the second-layer gated recurrent unit GRU structure captures context-related features, and the third-layer processing module constructs a fully connected network with a non-linear activation function to finally realize the end-to-end non-linear mapping from the original input space to the target prediction domain; use the trained prediction model to predict the test set data, and the output is the probability distribution of the user arriving at each sub-area under the time series.

6. The method for opportunistic crowd sensing task allocation sensitive to the urgency level according to claim 1, wherein: Based on the user trajectory prediction results, use the genetic algorithm to search for the optimal allocation scheme. The specific steps include: First, initialize the population Pop. At time t, according to the task urgency level w i Sort all tasks. For the task ta with the highest urgency level i , confirm its location area l j . Based on the probability pr j of each user in area l at time t output by the prediction model jk (t), assign ta i to the user u with the highest probability k ; Remove ta from the task set i , and repeat the previous step until all tasks are assigned. Then an initial assignment plan is obtained. To ensure the diversity of the initial population, generate the remaining initial assignment plans by randomly assigning users to each task. For the population Pop composed of the above initial assignment plans, calculate the fitness value of each individual, that is, the assignment plan, according to the objective function . Perform selection, crossover, and mutation evolution operations on the population Pop: 1) The tournament selection operation is adopted to randomly select several individuals from the current population Pop to form a group, and the individual with the optimal fitness is selected as the parent. The above operation is repeated until a sufficient number of parent individuals are selected from the population Pop to form a new parent population Pop l ; 2) The crossover operation first selects two individuals from the parent population Pop l , denoted as and respectively. Then, multiple crossover points are randomly selected, and a multi-point crossover method is used to exchange the allocation schemes of and to generate two new offspring individuals, denoted as off c and off d ; 3) The mutation operation randomly selects an assigned user in a sub-region from the offspring individual off c and replaces it with a user having a higher predicted probability and cost ratio. The same operation is performed on the offspring individual off d . The offspring individuals off c and off d are added to the offspring population Off; Update the population Pop based on the fitness value using Off; determine whether the termination criterion is met. If not, continue to execute 1) - 3); otherwise, terminate the algorithm and output the feasible individual with the largest fitness value in the population as the optimal task allocation scheme.

7. A computer device, characterized in that, It includes a processor and a memory, the processor is electrically connected to the memory, the memory is used to store instructions and data, and the processor is used to execute the method for allocating emergency - degree - sensitive opportunistic crowd - sourced sensing tasks according to any one of claims 1 - 6.

8. A computer-readable storage medium, characterized in that, A computer program is stored in the computer - readable storage medium, and this computer program is adapted to be loaded and executed by the processor to execute the method for allocating emergency - degree - sensitive opportunistic crowd - sourced sensing tasks according to any one of claims 1 - 6.

Images