A crowd-sourced task allocation method based on human-machine collaboration

Abstract

The present invention proposes a crowd intelligence task allocation method based on human-machine collaboration, belonging to the technical field of crowd sensing. This method introduces a large number of mobile users, vehicles and movable unmanned vehicles in the city to jointly complete the tasks submitted by users. By analyzing the historical routes of opportunistic users and opportunistic vehicles, it predicts the future movement routes, and designs a two-way expected value matching algorithm for task allocation; it conducts task allocation by auctioning the task bids of participatory users through a reverse auction mechanism; it clusters the remaining unallocated tasks through a clustering method, and allocates the unmanned vehicle to the nearest cluster, and sequentially allocates the tasks closest to the unmanned vehicle. The present invention makes full use of various movable resources in the city, effectively overcomes the limitations of single individual sensing, improves the completion rate of sensing tasks, saves the cost of completing sensing tasks, and can better adapt to the rich and diverse sensing task requests in the city.

Description

Technical Field

[0001] The present invention belongs to the technical field of crowd intelligence perception and relates to a crowd intelligence task allocation method based on human-machine collaboration. Background Art

[0002] Mobile crowd sensing has become a new paradigm for smart city perception. Unlike previous participatory environmental perception methods that used fixed sensors in cities, mobile crowd sensing emphasizes mobility, leveraging the vast number of mobile devices carried by the city's mobile population to achieve broader and more flexible environmental perception. In mobile crowd sensing, users have a flexible approach to completing tasks, primarily falling into two categories: opportunistic users who complete tasks by simply navigating their route, and participatory users who proactively complete tasks while on the move. However, because users apply for and accept sensing tasks based on their subjective will, while crowd sensing platforms implement incentive mechanisms to encourage users, practical experience shows that popular, lucrative, and low-cost sensing tasks are often prioritized, leaving many tasks unfinished on time. Consequently, mobile crowd sensing suffers from instability and uncontrollability, leading to a pressing challenge: how to improve the task completion rate of crowd sensing.

[0003] With the development of technologies such as cloud computing, autonomous driving, 5G communications, and urban vehicle networks, crowd-sensing technology based on mobile vehicles is becoming a new platform for urban perception. The large number of traditional and unmanned vehicles in a city can use various sensors to perceive the areas along their routes. After pre-processing, this perception data is transmitted to the crowd-sensing platform using communication technologies such as 5G and cellular networks to achieve city perception. Compared to mobile users, traditional vehicles can perform some perception tasks along the way, while mobile unmanned vehicles can use the crowd-sensing platform's instructions to perform less popular perception tasks and even complete relatively dangerous tasks on behalf of users. Furthermore, unmanned vehicles can travel long distances in the city for extended periods of time, enabling more cost-effective environmental perception. A large number of controllable mobile unmanned vehicles creates favorable conditions for achieving more extensive perception in cities.

[0004] Inspired by the above problems, the present invention proposes a method for allocating crowd intelligence tasks based on human-machine collaboration. Unlike previous mobile crowd intelligence perception models, the participants in the perception process are not limited to a single user or vehicle, but are composed of a large number of mobile users, vehicles, and mobile unmanned vehicles distributed throughout the city. By simulating real-world situations, mobile users can complete some perception tasks along the way or apply for perception tasks according to their own subjective wishes; vehicles moving in the city can also complete some perception tasks along the way; the remaining perception tasks that are unavailable, rarely accepted, or unsuitable for humans can be assigned to mobile unmanned vehicles by sending instructions from the crowd intelligence platform, thereby maximizing the completion of perception tasks through human-machine collaboration. Summary of the Invention

[0005] In response to the limitations of individual perception in existing crowd-intelligence perception methods, the present invention provides a crowd-intelligence task allocation method based on human-machine collaboration. It introduces a large number of mobile users, vehicles, and movable unmanned vehicles in the city to jointly complete tasks submitted by users. By analyzing the historical routes of opportunistic users and opportunistic vehicles to predict future movement routes, a two-way expected value matching algorithm is designed for task allocation; task allocation is carried out by bidding on the task bids of participating users through a reverse auction mechanism; the remaining unassigned tasks are clustered using a clustering method, and unmanned vehicles are assigned to the cluster closest to them, and then the tasks closest to the unmanned vehicles are assigned in turn.

[0006] The technical solutions of the present invention are as follows:

[0007] A crowd-based task allocation method based on human-machine collaboration consists of three stages: in the first stage, the crowd-based perception platform analyzes the historical routes of opportunistic users and vehicles to predict the possibility of future mobile destinations and perform task matching; in the second stage, the crowd-based perception platform issues task notifications, and participating users actively apply for tasks through task bids. The crowd-based perception platform auctions the bids to complete task allocation for participating users; in the third stage, the crowd-based perception platform clusters the remaining unassigned tasks, assigns the unmanned vehicle to the closest cluster, and then assigns the tasks closest to the unmanned vehicle in turn. The specific steps are as follows:

[0008] Step 1: The crowd-sensing platform analyzes the customer's task request and extracts specific information such as the task set T, task location, task time, task type, and task budget.

[0009] Step 2: In the first phase, the crowd-sensing platform assigns tasks to the opportunistic user set OPP and the opportunistic vehicle set OPV. Assuming that the historical routes of opportunistic users follow a Poisson distribution and that opportunistic vehicles have GPS history, the platform analyzes the past travel locations of opportunistic users and vehicles to calculate the probability of them appearing at the task location in the future. Task assignment is then performed using a designed bidirectional expected value matching algorithm. The specific process is as follows:

[0010] Step 2.1: Assume that an opportunistic user passes through a task t in one day. i Location i According to the Poisson distribution:

[0011]

[0012] Among them, s is the number of times passed, λ i The task location loc i The coefficient of .

[0013] Evaluate the opportunistic users passing through the task location loc within a day i The probability of

[0014]

[0015] Count each opportunistic user opp in task t i The past week before the start of the task location loc i The average probability of appearance is used to predict the opportunistic user's future appearance at the task location loc i The possibility of opp (loc i ).

[0016] Step 2.2: Use the historical GPS data of the unmanned vehicle to count the historical driving records of each opportunistic vehicle (opv) within one week and predict the opportunistic vehicle to appear at a certain task location (loc) i The probability is calculated as follows:

[0017]

[0018] Among them, N represents the number of tasks, count(loc i ) indicates that an opportunistic vehicle appears at the mission location loc i times.

[0019] Step 2.3: For each task t i There is an expectation value Will appear at the mission location loc i Possibility opportunistic users or Opportunistic vehicles are considered to be able to successfully complete tasks. Assuming that the maximum number of tasks that opportunistic users and opportunistic vehicles can match is both q, the opportunistic user set OPP and the opportunistic vehicle set OPV are traversed in sequence. According to E opp (loc i ) and E opv (loc i ), the tasks t i are sorted from large to small and then marked.

[0020] Step 2.4: During the traversal, if a marked task appears, then backtrack to see if there are unmarked tasks among the previously traversed opportunistic users or opportunistic vehicles. If an unmarked task is successfully found during the backtracking, the previously traversed opportunistic user or opportunistic vehicle marks the unmarked task, cancels the mark of the current task by the previously traversed opportunistic user or opportunistic vehicle, and at the same time the currently traversed opportunistic user or opportunistic vehicle marks the current task.

[0021] Step 2.5: If no unmarked task is found during the backtracking, the currently traversed opportunistic user or opportunistic vehicle skips marking the current task and selects the remaining unmarked tasks for marking.

[0022] Step 2.6: Repeat Step 2.3 to Step 2.5 until both the opportunistic users and opportunistic vehicles have been traversed.

[0023] Step 2.7: The crowdsensing platform assigns the marked tasks to the opportunistic users and opportunistic vehicles, and waits for the opportunistic users and opportunistic vehicles to upload the task results.

[0024] Step 3: The crowdsensing platform publishes the remaining unassigned task set RT in the first stage. The set of participatory users PAT in the city actively uploads the target task set set and the bid price price to the crowdsensing platform to apply for task execution. After receiving the applications from the participatory users, the crowdsensing platform filters out the applications with bids exceeding the task budget top, and uses the reverse auction mechanism to match tasks and participants. At the same time, a certain amount of points are given to the participatory users who participate in the auction but do not succeed as a cost compensation for the next auction to incentivize the participatory users. The specific process is as follows:

[0025] Step 3.1: Filter out the task bid prices h of the participatory user pat that exceed the budget top(set(pat h )) of the task set set(pat h ).

[0026] Step 3.2: Calculate the average bid price of the participating users pat h for the tasks submitted:

[0027]

[0028] Step 3.3: Sort in ascending order according to the size of , and evaluate the bids according to the following utility function:

[0029]

[0030] where RT is the set of tasks remaining unassigned after the task assignment of the first-stage opportunistic users and opportunistic vehicles, and CT is the set of candidate tasks for which the bids have been selected.

[0031] Step 3.4: Select the bids for which the utility function Unity(pat h ) > 0, and discard the bids with duplicate task sets.

[0032] Step 3.5: Continuously iterate, repeating Step 3.3 and Step 3.4 until all the bids submitted by the participating users are traversed.

[0033] Step 3.6: The participating users upload their initial positions, the crowd-sensing platform calculates the moving cost of the participating users, and after the participating users return the sensing results, the crowd-sensing platform rewards the participating users:

[0034]

[0035] where γ1 and γ2 respectively represent the price coefficients; dis(pat h , set(pat h )) represents the moving distance of the participating user pat h to complete the task set set(pat h ); represents the points obtained by the participating user pat h for losing the auction, and the points will accumulate after each failed auction and will be cleared after use.

[0036] Step 4: The crowd-sensing platform assigns the remaining tasks to the dedicated unmanned vehicles in the third stage. The crowd-sensing platform clusters the set of tasks RTF remaining unassigned in the second stage according to the task geographical coordinates and task types. After the task clustering is completed, the unmanned vehicles respectively select the cluster that is currently the closest, and then sequentially search for the task that is the closest in the cluster to complete. The specific steps are as follows:

[0037] Step 4.1: Select the number K of unmanned vehicles as the number of clusters after clustering, and use the positions of the unmanned vehicles as the initial cluster centers, denoted as

[0038] Step 4.2: Calculate the sensing task t in the remaining task set RTF in the second stage w the distance to the cluster center, and assign the sensing task to the cluster to which the nearest cluster center belongs.

[0039] Step 4.3: Recalculate the mean of the task points in each cluster, and use the mean as the new cluster center, denoted as where l = 1, 2, … is the number of iterations.

[0040] Step 4.4: Use the sum of the squared errors between the distance of each sensing task and the cluster center to which it belongs as the loss function, and calculate it in turn:

[0041]

[0042] where, c w is the cluster to which the task t w belongs, represents the center point of the cluster c w and M is the number of tasks in the remaining task set RTF.

[0043] Step 4.5: Repeat Step 4.2 to Step 4.4 until the loss function converges, and output the stable cluster center, denoted as

[0044] Step 4.6: Select the cluster that is currently the closest to the unmanned vehicle car v in turn:

[0045]

[0046] where, represents the Manhattan distance between the participatory unmanned vehicle car v and the cluster center output in Step 4.5 and j = 1, 2, …, K.

[0047] Step 4.7: The unmanned vehicle sequentially selects the currently closest cluster center, assigns the unmanned vehicle to the cluster, and sequentially selects the closest task in the cluster to complete.

[0048] Advantages of the present invention: The method for crowdsourcing task allocation based on human-machine collaboration proposed by the present invention makes full use of various mobile resources in the city, effectively overcomes the limitations of single-agent sensing, improves the completion rate of sensing tasks, saves the cost of completing sensing tasks, and can better adapt to the rich and diverse sensing task requests in the city. Brief Description of the Drawings

[0049] Figure 1 is the hierarchical structure diagram provided by the present invention based on human-machine collaboration.

[0050] Figure 2 It is a scenario diagram of task assignment based on human-machine collaboration provided by the present invention.

[0051] Figure 3 It is a flowchart of the task matching algorithm provided by the present invention.

[0052] Figure 4 It is a specific case diagram of the task matching algorithm provided by the present invention.

[0053] Figure 5 It is a flowchart of the user bid selection algorithm provided by the present invention.

[0054] Figure 6 It is a flowchart of the task clustering algorithm provided by the present invention. Specific embodiments [[ID=;]]

[0055] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the following will describe in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0056] The technical solution of the present invention is a crowdsourcing task assignment method based on human-machine collaboration. After the crowdsourcing platform analyzes the task requests uploaded by customers, as Figure 1 shown, the following task assignment method mainly consists of four task assignment sub-methods: opportunistic user task assignment, opportunistic vehicle task assignment, participatory user task assignment, and dedicated unmanned vehicle task assignment.

[0057] Figure 2 Shown as the task assignment scenario, after receiving the task requests uploaded by customers, the crowdsourcing perception platform divides the task assignment into three stages and finally returns the perception results to the customers. The task assignment work mainly consists of three stages. In the first stage, the historical routes of opportunistic users and opportunistic vehicles are analyzed to predict the possibility of future moving destinations for task matching; in the second stage, the crowdsourcing perception platform issues task notifications, and participatory users actively apply for tasks through bids, and the crowdsourcing perception platform conducts auctions on the bids to complete the task assignment of participatory users; in the third stage, the crowdsourcing perception platform clusters the remaining unassigned tasks, assigns the unmanned vehicles to the nearest clusters, and sequentially assigns the tasks closest to the unmanned vehicles. The specific steps are as follows:

[0058] Step 1: The crowdsourcing perception platform analyzes the task requests submitted by customers, extracts the task set T, and each task t i has attributes such as task location loc i and task budget top i and so on. Then, the task assignment work is carried out through the following three stages in sequence, and the results are returned one by one after the tasks are completed.

[0059] Step 2: The crowd-sensing platform performs task allocation for the opportunistic user set OPP and the opportunistic vehicle set OPV in the first stage. Assume that the historical routes of opportunistic users follow a Poisson distribution, and opportunistic vehicles have GPS historical driving records. Analyze the past driving locations of opportunistic users and opportunistic vehicles to calculate the probability of different opportunistic users appearing at the task location, and perform task allocation through the designed two-way expected value matching algorithm. The matching process is as Figure 3 shown; the specific steps are as follows:

[0060] Step 2.1: Assume that an opportunistic user passes through a certain task t i at the location loc i satisfies a Poisson distribution:

[0061]

[0062] where s is the number of passes, and λ i is the coefficient of the task location loc i .

[0063] The probability that an opportunistic user passes through the task location loc i in one day is

[0064]

[0065] Statistically analyze the average probability of each opportunistic participant opp appearing at the task location loc i in the past week before the start of task t i to predict the likelihood of an opportunistic user appearing at the task location loc i , denoted as E opp (loc i ).

[0066] Step 2.2: Through the historical GPS data of driverless vehicles, statistically analyze the historical driving records of each opportunistic vehicle opv within a week, and predict the probability of an opportunistic vehicle appearing at a certain task location loc i , and the calculation is as follows:

[0067]

[0068] where N represents the number of tasks, and count(loc i ) represents the number of times an opportunistic vehicle appears at the task location loc i .

[0069] Step 2.3: Each task t i has an expected value The likelihood of appearing at the task location loc i ​ opportunistic users or opportunistic vehicles are considered to be able to successfully complete the task. First, filter out the tasks that do not meet the task expectations For the tasks, assume that the maximum number of tasks that opportunistic users and opportunistic vehicles can match is both q. Traverse the opportunistic user set OPP and the opportunistic vehicle set OPV in turn, and respectively according to E opp (loc i ) and E opv (loc i ) sort the tasks t i in descending order and then mark the tasks.

[0070] Step 2.4: During the traversal process, if a task that has already been marked appears, then perform backtracking to see if there are unmarked tasks among the previously traversed opportunistic users or opportunistic vehicles. If an unmarked task is successfully found during the backtracking, the previously traversed opportunistic user or opportunistic vehicle marks the unmarked task, cancels the marking of the current task, and at the same time the currently traversed opportunistic user or opportunistic vehicle marks the current task.

[0071] Step 2.5: If no unmarked task is found during the backtracking, then the currently traversed opportunistic user or opportunistic vehicle skips marking this task and instead selects the remaining unmarked tasks for marking.

[0072] Step 2.6: Repeat Step 2.3 to Step 2.5 until the traversal of opportunistic users and opportunistic vehicles is completed.

[0073] Step 2.7: The crowd sensing platform assigns the marked tasks to opportunistic users and opportunistic vehicles and waits for the opportunistic users and opportunistic vehicles to upload the task results.

[0074] To better understand the task matching algorithm provided in the first stage, a specific example is provided below for illustration, such as Figure 4As shown in the figure, assume that the current tasks have been sorted, and the maximum number of task matches q is 2. According to step 2.3 of the above algorithm, we first traverse the opportunistic user opp1, mark the tasks {t1, t2}, and then find that task t2 has been marked by the opportunistic user opp1 during the traversal of user opp2. According to step 2.4 above, we perform backtracking and find that task t3 has not been marked by the opportunistic user opp1. Therefore, the opportunistic user opp1 deletes the mark on task t2 and adds a mark on task t3. At this time, the task sequence marked by the opportunistic user opp1 is {t1, t3}, and the opportunistic user opp2 adds a mark on task t2. Then the task sequence marked by the opportunistic user opp2 is {t2}. According to step 2.6, we perform subsequent traversal, and the opportunistic user opp3 marks task t4. When traversing the opportunistic vehicle opv1, it is found that task t4 has been marked by the opportunistic user opp3, and in the backtracking, it is found that the user opp3 has no unmarked tasks. Therefore, the opportunistic vehicle opv1 has no tasks to mark. Finally, the opportunistic vehicle opv2 marks task t5. At this point, the traversal has ended, and the crowd sensing platform assigns tasks to opportunistic users and vehicles and waits for the subsequent return of task results.

[0075] Step 3: The crowd sensing platform publishes the remaining unassigned task set RT in the first phase. The set of participatory users PAT in the city actively uploads the target task set set and the bid price price to the platform to apply for task execution. As Figure 5 shown, after receiving the applications from participatory users, the crowd sensing platform filters out the applications with bid prices exceeding the task budget top, and uses the reverse auction mechanism to match tasks and participants. At the same time, a certain amount of points are given to the participatory users who participate in the auction but do not succeed as a cost compensation for the next auction to incentivize participatory users.

[0076] Step 3.1: First, the crowd sensing platform filters out the task bid price price h submitted by the participatory user pat path that exceeds the budget top(set(pat h )) of the task set set(pat h ).

[0077] Step 3.2: Calculate the average bid price of the tasks submitted by the participatory user pat h :

[0078]

[0079] Step 3.3: Sort in ascending order according to and evaluate the bids according to the following utility function:

[0080]

[0081] Among them, RT is the set of tasks that remain unassigned after the first-stage opportunistic user and opportunistic vehicle task allocation, and CT is the set of candidate tasks that have won the bids.

[0082] Step 3.4: Select the bid with the utility function Unity(pat i ) > 0. In other cases, it is regarded as a failed auction. Discard the tasks that duplicate the candidate task set CT, and assign the remaining tasks to the participatory users. Update the remaining task set RT and the candidate task set CT. And issue a certain amount of integral card to the users who fail in the auction as compensation.

[0083] Step 3.5: Continuously iterate and repeat Steps 3.3 and 3.4 until all the bids submitted by the participatory users are traversed.

[0084] Step 3.6: The participatory users upload their initial positions, and the crowd sensing platform calculates the movement cost of the participatory users. After the participatory users return the task results, the crowd sensing platform rewards the participatory users:

[0085]

[0086] Among them, γ1 and γ2 represent the price coefficients respectively; dis(pat h , set(pat h )) represents the movement distance of the participatory user pat h to complete the task set set(pat h ); represents the integral obtained by the participatory user pat h when the auction fails. The integral will accumulate after each failed auction and will be cleared after use.

[0087] Step 4: The crowd sensing platform allocates the remaining tasks to the dedicated unmanned vehicles in the third stage. As Figure 6 shown, the crowd sensing platform first clusters the remaining task set RTF in the second stage according to the task geographical coordinates and task types. After the task clustering is completed, the unmanned vehicles respectively select the nearest cluster currently, and then sequentially search for the nearest task in the cluster to complete.

[0088] Step 4.1: First, select the number K of unmanned vehicles as the number of clusters after clustering, and use the positions of the unmanned vehicles as the initial cluster centers, denoted as

[0089] Step 4.2: Calculate the sensing task t in the unassigned task set RTF remaining in the second stage wThe distance to the cluster center, and allocate the sensing tasks to the cluster to which the nearest cluster center belongs.

[0090] Step 4.3: Recalculate the mean of the task points in each cluster, and use the mean as the new cluster center, denoted as where l = 1, 2, … is the number of iterations.

[0091] Step 4.4: For each sensing task t w Take the sum of the squared errors of the distance from the task to the cluster center as the loss function, and calculate it successively as follows:

[0092]

[0093] where, c w is the cluster to which task t w belongs, represents the center point of cluster c w and M is the number of tasks in the task set RTF.

[0094] Step 4.5: Repeat Step 4.2 to Step 4.4 until the above loss function converges, and output the stable cluster center, denoted as

[0095] Step 4.6: Successively select the cluster that is the nearest to the current driverless vehicle car v as follows:

[0096]

[0097] where, represents the Manhattan distance between the participatory driverless vehicle car v and the above output cluster center

[0098] Step 4.7: The driverless vehicles respectively select the currently nearest cluster center, allocate the driverless vehicles to the clusters, and the driverless vehicles sort the tasks in the clusters according to the distance from the driverless vehicles to the tasks from near to far, and successively select the task that is the nearest to the current position in the clusters to complete.

[0099] In summary, the crowdsourcing task allocation method based on human-machine collaboration proposed by the present invention makes full use of various types of mobile resources in the city, effectively overcomes the limitations of single individual sensing, improves the completion rate of sensing tasks, saves the cost of task completion, and can better adapt to the rich and diverse sensing task requests in the city.​

Claims

1. A crowd intelligence task allocation method based on human-machine collaboration, characterized in that, The method includes the following steps: Step 1: The crowd-sensing platform analyzes the task requests of customers, and extracts the task set T, task location, task time, task type, and task budget; Step 2: The crowd-sensing platform conducts task allocation for the opportunistic user set OPP and the opportunistic vehicle set OPV in the first stage; assume that the historical routes of opportunistic users follow a Poisson distribution, and opportunistic vehicles have GPS historical records. Analyze the past driving locations of opportunistic users and opportunistic vehicles to calculate the possibility of appearing at the task location in the future, and perform task allocation through the designed two-way expected value matching algorithm; Step 3: The crowd-sensing platform publishes the remaining unallocated task set RT after the first stage. The participatory user set PAT in the city actively uploads the target task set set and the bid price price to the crowd-sensing platform to apply for task execution; after receiving the applications from participatory users, the crowd-sensing platform filters out the applications with bids exceeding the task budget top, and uses a reverse auction mechanism to match tasks and participants; at the same time, it issues points to the participatory users who participate in the auction but do not succeed as a cost compensation for the next auction to incentivize participatory users; Step 4: The crowd-sensing platform clusters the remaining unallocated task set RTF in the second stage according to the task geographical coordinates and task types. After the task clustering is completed, the unmanned vehicles respectively select the nearest cluster currently, and then sequentially search for the nearest task in the cluster to complete.

2. The method for crowdsourcing task allocation based on human-machine collaboration according to claim 1, wherein The specific process of step 2 is as follows: Step 2.1: Assume that an opportunistic user passes a task t in one day. i Location i According to the Poisson distribution: where s is the number of passes, λ i is the coefficient of the task position loc i ; Evaluate the probability that opportunistic users pass through the task location loc within one day i is Count the average probability of each opportunistic user opp appearing at the task location loc in the past week before the start of task t i to predict the likelihood that the opportunistic user will appear at the task location loc in the future i , denoted as E i (loc opp ); i ) Step 2.2: Through the historical GPS data of the driverless vehicle, count the historical driving records of each opportunistic vehicle (opv) within a week, and predict the possibility that the opportunistic vehicle appears at a certain task location (loc), which is calculated as follows: i The possibility is calculated as follows: Among them, N represents the number of tasks, and count(loc i ) represents the number of times that the opportunistic vehicle appears at the task location loc i . Step 2.3: For each task t i has an expected value to appear at the task location loc i with a probability of an opportunistic user or an opportunistic vehicle is considered to be able to successfully complete the task. Let the maximum number of tasks that can be matched by opportunistic users and opportunistic vehicles be q. Traverse the opportunistic user set OPP and the opportunistic vehicle set OPV in turn, and respectively according to E opp (loc i ) and E opv (loc i ) sort the tasks t i from largest to smallest and then mark the tasks; Step 2.4: During the traversal process, if a marked task appears, then backtrack to see if there are unmarked tasks for the previously traversed opportunistic users or opportunistic vehicles; if an unmarked task is successfully found during the backtracking, the previously traversed opportunistic user or opportunistic vehicle marks the unmarked task, cancels the mark of the current task, and at the same time the currently traversed opportunistic user or opportunistic vehicle marks the current task; Step 2.5: If no unmarked task is found during the backtracking, the currently traversed opportunistic user or opportunistic vehicle skips marking the current task and selects the remaining unmarked tasks for marking; Step 2.6: Repeat steps 2.3 to 2.5 until the traversal of both opportunistic users and opportunistic vehicles is completed; Step 2.7: The crowd-sensing platform allocates the marked tasks to opportunistic users and opportunistic vehicles, and waits for opportunistic users and opportunistic vehicles to upload task results.

3. A method for crowd intelligence task allocation based on human-machine collaboration according to claim 1 or 2, characterized in that, The specific process of step 3 is as follows: Step 3.1: Filter out the task bid quotes price of the participatory user pat h that exceed the budget top(set(pat path )) of the task set set(pat h ); h ​ Step 3.2: Calculate the average quote of the participating user pat h for the submitted tasks: Step 3.3: Sort in ascending order according to the size of , and evaluate the bid according to the following utility function: Among them, RT is the remaining unallocated task set after the task allocation of opportunistic users and opportunistic vehicles in the first stage, and CT is the candidate task set that has won the bid; Step 3.4: Select the utility function Unity9pat h )>0 bids, discard bids with duplicate task sets; Step 3.5: Continuously iterate, repeat steps 3.3 and 3.4 until all the bids submitted by participatory users are traversed; Step 3.6: Participatory users upload their initial positions, the crowd-sensing platform calculates the movement cost of participatory users, and after participatory users return the sensing results, the crowd-sensing platform issues rewards to participatory users: Among them, γ1 and γ2 respectively represent price coefficients; dis(pat h , set(pat h )) represents the moving distance of the participatory user pat h to complete the task set set(pat h ); represents the points obtained by the participatory user pat h for failing the auction. The points will accumulate after each failed auction and will be cleared after use.

4. A method for crowd intelligence task allocation based on human-machine collaboration according to claim 1 or 2, characterized in that The specific process of step 4 is as follows: Step 4.1: Select the number K of driverless vehicles as the number of clusters after clustering, and use the positions of the driverless vehicles as the initial cluster centers, denoted as Step 4.2: Calculate the distance from the sensing task t in the remaining task set RTF in the second stage w to the cluster center, and assign the sensing task to the cluster to which the nearest cluster center belongs; Step 4.3: Recalculate the mean of the task points in each cluster, and use the mean as the new cluster center, denoted as where l = 1, 2, … is the number of iterations; Step 4.4: Take the sum of the squared errors between each perception task distance and the center of its belonging cluster as the loss function, and calculate them sequentially: Among them, c w is the cluster to which task t w belongs, represents the center point of cluster c w , and M is the number of tasks in the remaining task set RTF; Step 4.5: Repeat steps 4.2 to 4.4 until the loss function converges, and output the stable cluster centers, denoted as Step 4.6: Select the cluster that is closest to the current driverless vehicle car in sequence v in turn: Among them, represents the participatory driverless vehicle car v and the cluster center output in step 4.5 of the Manhattan distance, j = 1, 2, …, K; Step 4.7: The driverless vehicle sequentially selects the currently nearest cluster center, assigns the driverless vehicle to the cluster, and sequentially selects the nearest task in the cluster to complete.

5. A method for crowdsourcing task allocation based on human-machine collaboration according to claim 3, characterized in that, The specific process of Step 4 is as follows: Step 4.1: Select the number K of driverless vehicles as the number of clusters after clustering, and use the positions of the driverless vehicles as the initial cluster centers, denoted as Step 4.2: Calculate the distance of the sensing task t in the remaining task set RTF in the second stage to the cluster center, and assign the sensing task to the cluster to which the nearest cluster center belongs; w ​ Step 4.3: Recalculate the mean of the task points in each cluster, and use the mean as the new cluster center, denoted as where l = 1, 2, … is the number of iterations; Step 4.4: Take the sum of the squared errors between each perception task distance and the center of its belonging cluster as the loss function, and calculate them sequentially: Among them, c w is the cluster to which task t w belongs. represents the center point of cluster c w , and M is the number of tasks in the remaining task set RTF. Step 4.5: Repeat steps 4.2 to 4.4 until the loss function converges, and output the stable cluster centers, denoted as Step 4.6: Select the cluster that is closest to the current driverless vehicle car in sequence v in turn: Among them, represents the participatory autonomous vehicle car v and the cluster center output in step 4.5 The Manhattan distance, where j = 1, 2, …, K; Step 4.7: The driverless vehicle sequentially selects the currently nearest cluster center, assigns the driverless vehicle to the cluster, and sequentially selects the nearest task in the cluster to complete.

Images