User screening and incentive mechanism design method for distributed intelligent group perception

Abstract

The present invention discloses a user screening and incentive mechanism design method for distributed intelligent group perception, comprising: constructing an iterative framework based on a Steinberg model, initializing a price strategy through a requester for issuing tasks, setting a first iteration stop threshold for a user for processing tasks, and setting a second iteration stop threshold for a task requester; obtaining an expected return linear relationship of cooperation between users based on a zero determinant strategy and a prisoner's dilemma game; determining a user's utility function based on user contribution and user cost, and obtaining an optimal strategy based on the expected return linear relationship, the user's utility function, the user's task selection, and the calculation of the degree of participation in the task; based on the optimal strategy, judging whether the difference between two iterations of each user is greater than the first iteration stop threshold, and proceeding to the next step if the condition is met; based on the price strategy, judging whether the difference between two iterations of the requester is greater than the second iteration stop threshold, and obtaining the optimal price strategy if the condition is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of group intelligence perception mechanism design, and in particular to a user screening and incentive mechanism design method for distributed group intelligence perception. Background Art

[0002] With the rapid development of technology, users of smart terminal devices (e.g., smartphones, smart watches, etc.) have become ubiquitous, producing far-reaching positive impacts. These devices have significant computing power and rich sensors, which can not only collect valuable data but also complete certain computing tasks. Therefore, this has triggered a surge in researchers' interest in mobile crowdsensing (MCS) research. In a typical mobile crowdsensing scenario, there are two typical roles, namely requesters and workers (mobile users). The former publishes tasks and hopes to recruit workers to collaborate to complete them, while the latter selfishly makes decisions and selects tasks to obtain the maximum reward after completing specific tasks. Therefore, designing an appropriate incentive mechanism to maintain the interaction between these two roles and mobilize the enthusiasm of workers has become a valuable research area. However, the inherent heterogeneity of mobile devices and the complexity of actual scenarios have brought challenges to system modeling. However, many previous studies have deliberately avoided these challenges when designing the objective optimization function, such as ignoring the social-aware relationship between mobile users in idealized modeling. Therefore, there are still some challenges to be solved regarding the design of incentive mechanisms.

[0003] First, some researchers have focused on worker quality control in MCS. These studies can solve the problem of malicious users, but they are not effectively combined with incentive mechanisms. In addition, existing studies still have some limitations. They rarely consider the redundancy requirements of tasks. Even qualified users have a small probability of providing low-quality feedback due to dynamic environmental factors (such as packet loss in communication). Therefore, designing a suitable task load redundancy threshold is a practical problem, but how to evaluate the user's error rate is a challenge. In addition, many incentive mechanism designs ignore the social perception relationship between users. However, in most mobile computing scenarios, there are connections between these socially perceived users and rely on these relationships to assist in completing tasks. Some studies have taken social association effects into consideration in mobile MCS. However, the system models of these studies are relatively idealized. First, they only regard social effects as symmetric relationships. This means that they assume that collaborative relationships are equivalent. But in fact, in a pairwise collaborative relationship, Party A and Party B are usually in an unequal relationship, and there are challenges in how to deal with such an unequal collaborative relationship.

[0004] The above problems are common challenges in the design of incentive mechanisms for crowd-sensing systems. Existing research has not yet proposed widely accepted solutions on how to simultaneously consider incentive mechanisms and worker quality detection, reasonable evaluation of computing task redundancy, and modeling of asymmetric mobile user pairwise cooperation relationships. Related technical research is both challenging and particularly important. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a user screening and incentive mechanism design method for distributed intelligent group perception to solve the problems existing in the above-mentioned prior art.

[0006] To achieve the above object, the present invention provides a user screening method, comprising the following steps:

[0007] Obtain a task with a known result, where the requester who publishes the task divides the task into several subtasks and randomly distributes them to each user for processing the task, and obtains the work quality of each user completing the subtask; wherein the number of the subtasks is the same as the number of users;

[0008] Iteratively calculate the work quality of each user based on the random distribution of subtasks until an iteration exit condition is met, and obtain iteration data; wherein the iteration exit condition includes that the number of iterations reaches a preset number threshold, and the iteration data includes at least the work quality of each user during the iteration process;

[0009] A qualified work quality threshold is preset, and the average work quality of each user is calculated based on the iterative data. Users whose average work quality is greater than or equal to the qualified work quality threshold are regarded as qualified users, and unqualified users are eliminated.

[0010] Optionally, in the process of iteratively calculating the work quality of each user based on the random distribution of subtasks, the requester rearranges the order of subtasks in each iteration before distributing them to the users, and obtains the work quality of each user based on the rearranged subtasks.

[0011] Optionally, the process of obtaining the work quality of each user completing the subtask includes:

[0012] Obtain the completion status of each subtask completed by the user, including correct completion and incorrect completion;

[0013] The user's accuracy rate is obtained based on the quotient of all correct completion times and the total number of completed subtasks, and the accuracy rate is used as the work quality.

[0014] The above process is integrated into the following incentive mechanism design. The present invention also discloses a distributed intelligent group perception incentive mechanism design method, including:

[0015] An iteration framework is constructed based on the Steinberg model, a price policy is initialized by a requester for publishing a task, a first iteration stop threshold is set for a user for processing a task, and a second iteration stop threshold is set for a task requester;

[0016] Start iterative calculation in the iterative framework:

[0017] S1, based on the zero-determinant strategy and the prisoner’s dilemma game, obtains the expected linear relationship of cooperation between users;

[0018] S2, determining the user's utility function based on the user's contribution and the user's cost, and calculating and obtaining the optimal strategy based on the expected return linear relationship, the user's utility function, the user's task selection, and the degree of participation in the task;

[0019] S3, based on the optimal strategy, determine whether the difference between two iterations of each user is greater than the first iteration stop threshold, and proceed to the next step if the condition is met;

[0020] S4, based on the price strategy, determine whether the difference between the requester's two iterations is greater than the second iteration stop threshold. If the condition is met, the optimal price strategy is obtained, otherwise, enter the iterative process of S1-S4 until the condition is met.

[0021] Optionally, before starting S1 in the iteration framework, the method further includes: when the number of iterations is the number of user screening iterations, performing user screening using the user screening method according to any one of claims 1 to 3 to determine a qualified user set;

[0022] The user screening iteration number includes the first iteration calculation and any other iteration calculation numbers.

[0023] Optionally, in S1, in the process of obtaining the expected linear relationship of returns for cooperation between users based on the zero-determinant strategy and the prisoner's dilemma game, the Markov matrix is ​​used to solve the optimal strategy of short-term memory between the last iteration and the current iteration in the prisoner's dilemma game situation, and the zero-determinant strategy is used for calculation based on the optimal strategy to obtain the expected linear relationship of returns for cooperation between users.

[0024] Optionally, in S2, in the process of determining the user's utility function based on the user contribution and the user cost:

[0025] The user contribution includes the user's own contribution and the contribution between users. The expression of the user contribution is:

[0026] C i (π i ,π -i )=f i (π i )+SCi (π i ,π -i )

[0027] Among them, C i (π i ,π -i ) is user contribution, f i (π i ) is the user's own contribution, SC i (π i ,π -i ) refers to the contribution between the users through social cooperation, π i =(t i ,x i ) i represents the task selection strategy of user i, where x i represents the workload of user i, Indicates the specific task that user i participates in;

[0028] The user cost expression is:

[0029] c i (π i )=∈·(1+E)·c i x i

[0030] Among them, c i (π i ) is the cost of user i, ∈ represents a tuning parameter greater than 1, E represents the overall error rate of users completing tasks calculated by the user screening mechanism, c i represents the unit cost of user i completing the task, x i represents the workload of user i;

[0031] The expression of the utility function is:

[0032]

[0033] Among them, p t represents the unit value of task t, P t represents the overall value of task t, M t represents the number of users participating in task t, a i represents a non-negative parameter related to user i itself, b i Indicates that the non-negative parameter is related to user i itself, The indicator parameter for obtaining rewards through positive cooperation from i to j is expressed as, The indicator parameter for obtaining rewards through reverse cooperation from j to i is expressed as, x j represents the workload of user j, π -irepresents the task selection strategy of all users except user i, π -i ={π1,…,π i-1 ,π i+1 ,…,π N}.

[0034] Optionally, in S2, the optimal strategy is obtained based on the expected return linear relationship, the user's utility function, the user's task selection, and the degree of participation in the task;

[0035] Obtaining each user's task selection based on the expected return linear relationship, the participation degree of each task and the utility function, thereby obtaining an optimal strategy;

[0036] Among them, the calculation expression of the participation of each task is:

[0037]

[0038] in, represents the participation of user i in task t in the nth iteration, represents the participation of user j in task t in the n-1th iteration, C i Represents ∈·(1+E)·c i , Represents the overall value of task t in the n-1th iteration.

[0039] Optionally, in S3, based on the optimal strategy, when determining whether the difference between two iterations of each user is greater than the first iteration stop threshold, if the condition is not met, iterate the process of S1-S3 until the condition is met.

[0040] Compared with the prior art, the present invention has the following advantages and technical effects:

[0041] The present invention discloses a user screening and incentive mechanism design method for distributed intelligent group perception. The Stackelberg Game (SG) is used to ensure the fairness and feasibility of the task allocation scheduling algorithm. The requester (Requester) motivates the participation of mobile users (Mobile Users) through rewards. Under the conditions of ensuring efficiency and credibility, the optimization method maximizes the interests of all parties. The invention is combined with the Iterated Prisoner's Dilemma (IPD) game framework based on the Zero-Determinant (ZD) strategy to theoretically ensure the cooperation relationship between users. The invention can be applied to the task allocation and optimization problems of mobile group perception systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0043] Figure 1 A schematic diagram of a user screening and task redundancy requirement assessment process according to an embodiment of the present invention;

[0044] Figure 2 The figure is an overall flow chart of the iterative algorithm of the Steinberg model game according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0046] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0047] Embodiment 1

[0048] like Figure 1 As shown, this embodiment provides a user screening method, including:

[0049] Obtain a task with a known result, where the requester who publishes the task divides the task into several subtasks and randomly distributes them to each user for processing the task, and obtains the work quality of each user completing the subtask; wherein the number of the subtasks is the same as the number of users;

[0050] Iteratively calculate the work quality of each user based on the random distribution of subtasks until an iteration exit condition is met, and obtain iteration data; wherein the iteration exit condition includes that the number of iterations reaches a preset number threshold, and the iteration data includes at least the work quality of each user during the iteration process;

[0051] A qualified work quality threshold is preset, and the average work quality of each user is calculated based on the iterative data. Users whose average work quality is greater than or equal to the qualified work quality threshold are regarded as qualified users, and unqualified users are eliminated. Specifically, it includes:

[0052] User Screening and Task Redundancy Requirements Assessment: A Plug-and-Play Module (Screening Process & Redundancy Design):

[0053] Before the iterative game begins, potential users need to be screened and the overall error rate of valid users needs to be evaluated. The specific process is as follows.

[0054] In order to effectively filter malicious users and low-quality users, and to efficiently evaluate task redundancy requirements, a sub-algorithm for user screening and task redundancy requirement evaluation is designed (Algorithm 1). It should be pointed out that this sub-algorithm can not only be used to initialize parameter values ​​in the main framework of Algorithm 2, but can also be used as a plug-and-play module to monitor user quality and evaluate redundancy thresholds in real time during the iteration process. The main steps of the algorithm are as follows:

[0055] The requester has a toy task Ψ with known real results, and the potential user set is defined as: Divide this task into N' subtasks, represented as {τ1,…,τ i ,…,τ N'}.

[0056] The requester randomly distributes these N' subtasks to N' users as baits, and then checks the quality of each user's work and records their results.

[0057] As an implementation method of this embodiment, in the process package of iteratively calculating the work quality of each user based on the random distribution of subtasks, the requester rearranges the order of subtasks in each iteration and then distributes them to the users, and obtains the work quality of each user based on the rearranged subtasks. The specific methods include:

[0058] Rearrange the order of the subtasks and repeat the above two steps k times (eg k=10) to test the quality of the user.

[0059] As an implementation method of this embodiment, the process of obtaining the work quality of each user completing a subtask includes: obtaining the completion status of each subtask completed by the user, the completion status including correct completion and incorrect completion; obtaining the user's accuracy rate based on the quotient of all correct completion times and the total number of completed subtasks, and using the accuracy rate as the work quality. Specifically including:

[0060] By calculating the average of each user's completion quality, a qualified user set can be selected based on the quality criteria. For example, the average accuracy of individual users is >95%.

[0061] Assuming that the overall error rate of qualified users follows a certain statistical distribution, we can calculate the error rate E and set the actual workload of completing the task to ∈·(1+E) times the theoretical value (i.e., x i ), where ∈≥1 is a security parameter.

[0062] Embodiment 2

[0063] like Figure 2 As shown, this embodiment provides a distributed intelligent group perception incentive mechanism design method, including:

[0064] An iteration framework is constructed based on the Steinberg model, a price strategy is initialized by a requester for issuing a task, a first iteration stop threshold is set for the user, and a second iteration stop threshold is set for a requester for processing the task; specifically, the following steps are included:

[0065] During the initialization process, the requester provides the initial price strategy and the iteration stop threshold:

[0066] As a utility-driven system, requesters have an incentive to adjust the price of each task: This adjustment has two purposes: one is to encourage users to participate in the task, and the other is to maximize the utility of the requester. t The number of participants in is considered a strong indicator of the popularity of the task. Based on this, the unit task value is defined as follows:

[0067] p t =P t / |M t |.

[0068] Note that this equation intuitively shows that p t and |M t | In addition, the requester can dynamically set P t =p t ×|M t |To adjust these two interrelated indicators p t and |M t |.

[0069] When the system is initialized, in order to attract users to participate in the crowd intelligence perception task, the request will initialize a pricing strategy At the same time, according to the actual situation, a threshold α1 and α2 for stopping the iterative calculation will be set for the user and the requester respectively. If both the user and the requester reach the threshold, it is considered that both parties have achieved the optimal strategy, that is, the equilibrium state. After initialization, through s∈{1,2,3,…} rounds of iterative game, the requester and the user will eventually reach the optimal strategy at the same time. The first iterative stopping threshold is α1, and the second iterative stopping threshold is α2.

[0070] Start iterative calculation in the iterative framework:

[0071] S1, based on the zero-determinant strategy and the prisoner’s dilemma game, obtains the expected linear relationship of cooperation between users;

[0072] As an implementation method in this embodiment, in S1, in the process of obtaining the expected return linear relationship of cooperation between users based on the zero determinant strategy and the prisoner's dilemma game, the Markov matrix is ​​used to solve the optimal strategy of short-term memory between the last iteration and the current iteration in the prisoner's dilemma game, and the zero determinant strategy is used to calculate based on the optimal strategy to obtain the expected return linear relationship of cooperation between users. The specific process includes:

[0073] Based on the zero determinant strategy for asymmetric cooperation modeling, we construct the cooperation reward parameters:

[0074] Due to the social perception connection between users, users can increase each other's benefits through pairwise cooperation. The benefits generated by pairwise cooperation between users are defined as additional benefits, which are defined as:

[0075] SC i (π,π -i )=PSC i (π i ,π -i )+RSC i (π i ,π -i ) (1)

[0076] PSC i (π i ,π -i ) represents the positive social effect of user i on other users, and RSC i (π i ,π -i ) represents the reverse social effects of other users on user i. Based on the revised zero decisiveness (ZD) strategy (see below for details), a mutual cooperation enforcement mechanism is constructed to keep these socially aware connections stable. Under the ZD strategy, the positive and reverse social effects can be expressed as:

[0077]

[0078] Among them, x i ,x j are the participation levels of user i and user j in the task, and are two different rewards for socially aware connections between pairs of users in different directions. Note that Its actual value depends on the IPD game selection, Follow the same rules.

[0079] The following is a detailed introduction to the IPD game process based on the ZD strategy, as well as improvements based on the classic IPD game.

[0080] Here, we focus on analyzing the positive social effects PSC i (π i ,π -i ), because RSC i (π i ,π -i ) can produce the same result. IPD (Iterated Prisoner's Dilemma) is a classic two-way game in which each user, user i or user j, can selfishly choose to cooperate (abbreviated as C) or betray (i.e., not cooperate, abbreviated as D) at each stage. Therefore, based on the strategies of the two users, the possible combinations can be expressed as IJ = {CC, CD, DC, DD}. Due to different results, the rewards paid to users follow the following three rules:

[0081] If both users choose to cooperate (CC), each will receive a reward R ij x i x j to encourage such cooperation.

[0082] If both users choose to defect (DD), each will receive a small reward P ij x i x j . Due to R ij x i x j Much larger than P ij x i x j , cooperation is more beneficial to both parties.

[0083] If CD / DC occurs, the cooperator (C) will receive the minimum reward S ij x i x j , and the traitor (D) will receive the highest reward T ij x i x j , which is greater than R ij x i x j This reward setting makes sense because the traitor only gains benefits from the opponent without spending his own resources, while the cooperator does not gain any benefits except for making contributions.

[0084] Due to the order of IJ = {CC, CD, DC, DD}, the reward indicator vector of user i and user j can be expressed as S I =(R ij ,S ij ,T ij ,P ij ) and S J =(R ij ,T ij ,S ij ,P ij ). It should be noted that the game must satisfy two inequalities: T ij >R ij >P ij >S ij and 2R ij >T ij +S ij These conditions clearly show that users can maximize their own rewards by defecting, or achieve the global optimal reward by cooperating with each other.

[0085] In an iterative game, two players engage in a series of interactions and adjust their strategies based on their opponent's previous choices. In order to explore hidden relationships, William Press and Freeman Dyson used Markov matrices as an effective analytical tool in IPD games. The Markov matrix (also called the transition matrix) is a matrix used to describe the probability of state transitions in a Markov chain. A Markov chain is a random process in which transitions occur between states between discrete time steps, and the transition probability depends only on the current state, not on past states. Based on this assumption, long-term memory cannot help users increase their winning advantage. Therefore, the process from stage t-1 to stage t is usually considered. The strategy of user i in stage t is defined as the probability vector P I =(I1,I2,I3,I4), where each item represents the probability of user i cooperating under each possible outcome of CC, CD, Dc and DD in stage t-1. For example, I1 is the probability of cooperating with user j under the CC outcome in stage t-1. Similarly, the strategy of user j in stage t can be defined as P J =(J1,J2,J3,J4), according to each possible outcome of CC,DC,CD and OD in stage t-1. Then, between any two consecutive stages, the Markov matrix for IJ={CC,CD,DC,DD} can be expressed as M:

[0086]

[0087] Note that each element represents the transition probability from stage t-1 to stage t. For example, the elements in the first row represent the probability of state IJ = {CC, CD, DC, DD} in stage t based on state CC in stage t-1.

[0088] Let v = (v1, v2, v3, v4) be the steady-state distribution of IJ on the Markov matrix M. For any 4-dimensional vector f = (f1, f2, f3, f4), the dot product v·f can be derived as follows:

[0089]

[0090] To utilize equation (8), the expected rewards of users can be calculated separately as follows:

[0091]

[0092] Note that 1 is a vector with 4 values ​​of 1, and the denominator is used for normalization.

[0093] In addition, a striking feature of equation (4) is the second column of the determinant:

[0094]

[0095] It only interacts with P under the control of user i I =(I1,I2,I3,I4) related. In decoupling P I and P J Afterwards, it provides a simple way for user i to unilaterally control the iterated game. For user j, the same conclusion is reached when considering the third column.

[0096] Similar to equations (4) and (5), the linear combination of their returns can be defined as:

[0097]

[0098] where α, β and γ are constants.

[0099] Note that if user i uses This means that the numerator in equation (11) will be zero. This strategy is called a zero deterministic (ZD) strategy, and it enforces a linear relationship between its expected returns:

[0100] αR I +βR J +γ=0 (8)

[0101] Therefore, the ZD strategy allows user i to unilaterally utilize its own strategy so that the expected returns of users i and j satisfy the linear constraint relationship.

[0102] To be consistent with real-life situations, positive and negative social influences were considered, and the IPD game was adjusted to fit the specific scenario. i (π i ,π -i ), user i plays a dominant role, indicating a greater willingness to cooperate. For example, user i may need the location information data of the opponent to complete navigation-related tasks. Therefore, it is sufficient to maintain the ZD strategy capabilities of the dominant user and unilaterally force cooperation. In the system, inspired by previous studies, the classic setting is modified to better adapt to the asymmetric social perception relationship between users.

[0103] Considering that the dominant user i takes action first in each round, then its strategy is still P' I =(I'1,I'2,I'3,I'4), similar to the previous analysis of P I As for subordinate user j, let P' J =(J'1,J'2) represents its strategy in stage t, where J'1 and J'2 are the probabilities that user j chooses to cooperate (C) when observing the action (C / D) of user i in stage t. The revised transition matrix can be defined as M':

[0104]

[0105] Similarly, define v'=(v'1,v'2,v'3,v'4) as the stationary distribution of IJ on the Markov matrix M', and f'=(f'1,f'2,f'3,f'4) as a 4-dimensional vector. The dot product expression of v'·f' is as follows:

[0106]

[0107] Note that since the equivalent states of the two users are modified, only the dominant user i can Use ZD strategy.

[0108] Similar to equation (9), the expected rewards of users can be defined as follows:

[0109]

[0110] In addition, we want to ensure mutual cooperation among socially aware users. According to previous studies, mutual cooperation can be enforced if the dominant user i satisfies the following conditions:

[0111]

[0112] Where χ is greater than 1, I'1 = 1 means that user i always chooses cooperation when the last stage is CC.

[0113] Considering that user j’s reward is limited and cannot exceed R ij x i x j Therefore, by cooperating, users will eventually achieve a mutually beneficial outcome in which each user will receive R ij x i x j The reward (i.e. ). Note that the above analysis is performed on the RSC with reverse social effect i (π i ,π -i ) occurs naturally, and the dominant user is j, and the mutual cooperation reward is R ji x i x j (Right now ). Therefore, as long as users are rational, that is, in accordance with the IPD game setting, they will eventually choose to cooperate to maximize their benefits.

[0114] In S1, in the process of obtaining the expected linear relationship of returns for cooperation between users based on the zero-determinant strategy and the prisoner's dilemma game, the Markov matrix is ​​used to solve the optimal strategy of short-term memory between the last iteration and the current iteration in the prisoner's dilemma game situation, and the zero-determinant strategy is used for calculation based on the optimal strategy to obtain the expected linear relationship of returns for cooperation between users.

[0115] The specific process of users selecting the optimal strategy includes:

[0116] In the iterative framework of s∈{1,2,3,…} rounds, mobile users also need to perform multiple rounds of iterations so that each user can obtain the optimal strategy. Here, the iteration indicator of the mobile user is set: n=0, and the strategy of each user is initialized: in

[0117] S2, determining the user's utility function based on the user's contribution and the user's cost, and obtaining the optimal strategy based on the expected return linear relationship, the user's utility function, the user's task selection, and the degree of participation in the task; the specific process includes:

[0118] In each iteration, each user will calculate his or her optimal strategy for this round, that is, the optimal task and the optimal task participation, based on the utility function and the information about other users (task selection and task participation, etc.) calculated in the previous iteration (that is, [n-1]).

[0119] The mobile user's utility function is defined as follows.

[0120] For each user Configure π based on each user's own policy i =(t i ,x i ), and define its contribution as follows:

[0121] C i (π i ,π -i )=f i (π i )+SC i (π i ,π -i ) (13)

[0122] Among them, C i (π i ,π -i ) is user contribution, f i (π i ) is the user's own contribution, SC i (π i ,π -i ) refers to the contribution between the users through social cooperation, π i =(t i ,x i ) i represents the task selection strategy of user i, where x i represents the workload of user i, Indicates the specific task that user i participates in. In this equation, the first term represents the user’s independent contribution, which is independent of other users. The second term represents the additional contribution due to the social connection relationship with other users.

[0123] For the first item, the user's own contribution can be quantified concisely as:

[0124]

[0125] The coefficient a i and b i is a non-negative number related to the user's intrinsic efficacy and heterogeneity-based differences.

[0126] The second term describes the additional contribution brought by the social interaction between user i and other users, as shown in formulas (1)-(2).

[0127] Now, we are ready to define the cost per mobile user, which is determined by the fee paid by the requester and the cost of performing the task. Let c irepresents the unit cost of completing the task of user i. As mentioned above, in order to ensure the quality of the task, it is necessary to consider the redundant calculation caused by uncontrollable factors. Therefore, the total cost of user i can be expressed as:

[0128] c i (π i )=∈·(1+E)·c i x i (15)

[0129] Among them, c i (π i ) is the cost of user i, ∈ represents a tuning parameter greater than 1, E represents the overall error rate of users completing tasks calculated by the user screening mechanism, c i represents the unit cost of user i completing the task, x i represents the workload of user i;

[0130] Based on formulas (13)-(15), the utility of each user i is Can be expanded to:

[0131]

[0132] Among them, p t represents the unit value of task t, P t represents the overall value of task t, M t represents the number of users participating in task t, a i represents a non-negative parameter related to user i itself, b i Indicates that the non-negative parameter is related to user i itself, The indicator parameter for obtaining rewards through positive cooperation from i to j is expressed as, The indicator parameter for obtaining rewards through reverse cooperation from j to i is expressed as, x j represents the workload of user j, π -i represents the task selection strategy of all users except user i, π -i ={π1,…,π i-1 ,π i+1 ,…,π N}, which does not include π i .

[0133] As shown in Algorithm 2,

[0134] According to the optimal task participation formula, calculate the optimal participation of each task:

[0135]

[0136] The superscript [n-1] represents the mobile user information in the previous round, and the strategy for this round is calculated based on the information in the previous round (i.e., the parameter represented by the superscript [n]). represents the participation of user i in task t in the nth iteration, represents the participation of user j in task t in the n-1th iteration, C i Represents ∈·(1+E)·c i , Represents the overall value of task t in the n-1th iteration.

[0137] S3, based on the optimal strategy, determine whether the difference between two iterations of each user is greater than the first iteration stop threshold, and proceed to the next step if the condition is met;

[0138] According to formula (17), we can use formula (16) to calculate the optimal task selection for each user in this round, thus obtaining the optimal strategy That is, the optimal task and optimal participation in this round.

[0139] Determine whether each user meets the threshold |x i [n] -x i [n-1] |>α1, if it is satisfied, then output the optimal strategy for all users Give it to the requester; if not satisfied, repeat the above process.

[0140] S4, based on the price strategy, determine whether the difference between the requester's two iterations is greater than the second iteration stop threshold. If the condition is met, the optimal price strategy is obtained, otherwise, enter the iterative process of S1-S4 until the condition is met.

[0141] The requester makes price strategy adjustments:

[0142] First, define the requester's utility function. Due to the inherent heterogeneity among users, the quality of each user's reception completion is different. i Represents each user Therefore, the actual contribution provided by user i to the requester can be expressed as q i x i Then, the utility function It can be defined as the difference between the total contributions and total payments of all users. That is:

[0143]

[0144] according to If the inequality is satisfied, then the price strategy of the requester is considered to be optimal. The inequality indicates that the utility function U in rounds s-1 and s is R The difference.

[0145] 1) Obtain the optimal price strategy and user strategy / continue iterative game:

[0146] If the requester also obtains the optimal price strategy, the optimal strategy of the requester and the user is output at the same time π * ; If the requester does not obtain the optimal price strategy, set s=s+1 for a new round of iteration.

[0147] The overall algorithm flow of specific implementation steps: Two-stage Stackelberg Game Algorithm Framework:

[0148] This section summarizes the algorithm flow of the above specific implementation steps.

[0149] An iterative optimization framework is proposed, which consists of two sequential parts: follower-side and leader-side. The framework aims to converge to the equilibrium point of the SG game. Before the SG game, the requester broadcasts the initial price strategy to the users, and two early stopping thresholds are set to help stop the iteration. Subsequently, the iterative process will begin. At each iteration, it is first necessary to decide whether to execute the plug-and-play block such as Figure 1 As shown. Figure 2 As shown, except for the first round, some rounds (i.e., represented as a set of random positive integers: l) are randomly selected to execute this nested block, which can ensure the timeliness of user detection and error rate evaluation.

[0150] On the follower side, each user can observe the strategies of other users in the previous round. Then, each user establishes a stable social perception connection with other users who joined the same task in the previous round through the ZD strategy. Based on these actions, each user calculates its optimal participation level on each task and then selects the optimal strategy for the current round. Users continue to update their strategies until the difference in the optimal task participation level of all users between two iterations is less than a threshold α1. Once all users' strategies contradict the criterion α1 (i.e., the absolute value < α1), it means that the users have reached acceptable convergence. Then, the strategy vector, as the optimal solution π * , shared with the requester, who will adjust their pricing strategy in the next subsection

[0151] On the leader side, the requester checks the difference in its utility between two iterations and decides whether to continue adjusting its price strategy until the convergence criterion α2 is reached. When the requester adjusts its price strategy, all users will re-evaluate their own strategies and roll back the cycle to obtain a new equilibrium. The whole framework allows obtaining an approximate equilibrium point of the SG game, such as Figure 2 shown.

[0152] The present invention uses Stackelberg Game (SG) to ensure the fairness and feasibility of the task allocation and scheduling algorithm. The requester motivates the participation of mobile users through rewards, and maximizes the interests of all parties under the conditions of ensuring efficiency and credibility. The invention is combined with the Iterated Prisoner's Dilemma (IPD) game framework based on the Zero-Determinant (ZD) strategy to theoretically ensure the cooperation between users. The invention can be applied to the task allocation and optimization problems of mobile crowd intelligence perception systems.

[0153] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A distributed intelligent group perception incentive mechanism design method, characterized in that: include: An iteration framework is constructed based on the Steinberg model, a price policy is initialized by a requester for publishing a task, a first iteration stop threshold is set for a user for processing a task, and a second iteration stop threshold is set for a task requester; Start iterative calculation in the iterative framework: S1, the prisoner's dilemma game based on the zero-determinant strategy obtains the expected linear relationship of cooperation between users; S2, determining the user's utility function based on the user's contribution and the user's cost, and calculating and obtaining the optimal strategy based on the expected return linear relationship, the user's utility function, the user's task selection, and the degree of participation in the task; S3, based on the optimal strategy, determine whether the difference between two iterations of each user is greater than the first iteration stop threshold, and proceed to the next step if the condition is met; S4, based on the price strategy, determine whether the difference between the requester's two iterations is greater than the second iteration stop threshold. If the condition is met, the optimal price strategy is obtained, otherwise, enter the iterative process of S1-S4 until the condition is met.

2. The distributed intelligent group perception incentive mechanism design method according to claim 1 is characterized in that: Before starting S1 in the iteration framework, the method further includes: when the number of iterations is the number of user screening iterations, performing user screening using a user screening method to determine a qualified user set; The user screening iteration number includes the first iteration calculation and any other iteration calculation numbers.

3. The distributed intelligent group perception incentive mechanism design method according to claim 2 is characterized in that: The following steps are involved: Obtain a task with a known result, where the requester who publishes the task divides the task into several subtasks and randomly distributes them to each user for processing the task, and obtains the work quality of each user completing the subtask; wherein the number of the subtasks is the same as the number of users; Iteratively calculate the work quality of each user by randomly distributing subtasks until the iteration exits the condition and obtains the iteration data; The iteration exit condition includes that the number of iterations reaches a preset number threshold, and the iteration data includes at least the work quality of each user during the iteration process; A qualified work quality threshold is preset, and the average work quality of each user is calculated based on the iterative data. Users whose average work quality is greater than or equal to the qualified work quality threshold are regarded as qualified users, and unqualified users are eliminated.

4. The distributed intelligent group perception incentive mechanism design method according to claim 3 is characterized in that: In the process of iteratively calculating the work quality of each user based on the random distribution of subtasks, the requester rearranges the order of subtasks in each iteration before distributing them to the users, and obtains the work quality of each user based on the rearranged subtasks.

5. The distributed intelligent group perception incentive mechanism design method according to claim 3 is characterized in that: The process of obtaining the work quality of each user completing a subtask includes: Obtain the completion status of each subtask completed by the user, including correct completion and incorrect completion; The user's accuracy rate is obtained based on the quotient of all correct completion times and the total number of completed subtasks, and the accuracy rate is used as the work quality.

6. The distributed intelligent group perception incentive mechanism design method according to claim 1 is characterized in that: In S1, in the process of obtaining the expected linear relationship of returns for cooperation between users based on the zero-determinant strategy and the prisoner's dilemma game, the Markov matrix is ​​used to solve the optimal strategy of short-term memory between the last iteration and the current iteration in the prisoner's dilemma game situation, and the zero-determinant strategy is used for calculation based on the optimal strategy to obtain the expected linear relationship of returns for cooperation between users.

7. The distributed intelligent group perception incentive mechanism design method according to claim 1 is characterized in that: In S2, in the process of determining the user's utility function based on the user's contribution and the user's cost: The user contribution includes the user's own contribution and the contribution between users. The expression of the user contribution is: C i (p i ,p -i )=f i (p i )+SC i (p i ,p -i ) Among them, C i (π i , π -i ) is user contribution, f i (π i ) is the user's own contribution, SC i (π i , π -i ) refers to the contribution between the users through social cooperation, π i =(t i , x i ) i represents the task selection strategy of user i, where x i represents the workload of user i, Indicates the specific task that user i participates in; The user cost expression is: c i (π i )=∈·(1+E)·c i x i Among them, c i (π i ) is the cost of user i, ∈ represents a tuning parameter greater than 1, E represents the overall error rate of users completing tasks calculated by the user screening mechanism, c i represents the unit cost of user i completing the task, x i represents the workload of user i; The expression of the utility function is: Among them, p t represents the unit value of task t, P t represents the overall value of task t, M t represents the number of users participating in task t, a i represents a non-negative parameter related to user i itself, b i Indicates that the non-negative parameter is related to user i itself, The indicator parameter for obtaining rewards through positive cooperation from i to j is expressed as, The indicator parameter for obtaining rewards through reverse cooperation from j to i is expressed as, x j represents the workload of user j, π -i represents the task selection strategy of all users except user i, π -i ={π1,…,π i-1 , π i+1 , …, π N }.

8. The distributed intelligent group perception incentive mechanism design method according to claim 1 is characterized in that: In S2, the optimal strategy is obtained based on the expected return linear relationship, the user's utility function, the user's task selection, and the task participation calculation; Obtaining each user's task selection based on the expected return linear relationship, the participation degree of each task and the utility function, thereby obtaining an optimal strategy; Among them, the calculation expression of the participation of each task is: in, represents the participation of user i in task t in the nth iteration, represents the participation of user j in task t in the n-1th iteration, C i Represents ∈·(1+E)·c i , Represents the overall value of task t in the n-1th iteration.

9. The distributed intelligent group perception incentive mechanism design method according to claim 1 is characterized in that: In S3, based on the optimal strategy, it is determined whether the difference between two iterations of each user is greater than the first iteration stop threshold. If the condition is not met, the process of S1-S3 is iterated until the condition is met.

Images