Mobile crowd sensing credible worker selection and task allocation method

Abstract

The invention relates to the technical field of mobile crowd sensing, and particularly provides a mobile crowd sensing credible worker selection and task allocation method, which comprises the following steps of: performing data sensing based on mobile intelligent equipment carried by a worker to obtain sensing data; carrying out truth value estimation on the sensing data based on the principle of minimizing the data variance; calculating the deviation of the mean value and the variance of the perception data of each worker relative to the standard data, and performing credible worker qualification examination; on the basis of the credible worker set, the contribution value of each worker is evaluated, and the actual performance of each worker in the sensing task is quantified; in a multi-round perception task allocation process, a reverse auction mechanism is adopted to dynamically adjust offers of workers, individual contribution values and virtual points of the workers are comprehensively considered, and effective allocation of tasks is realized. The feature distribution difference between malicious workers and credible workers is measured, the deep difference of data distribution is captured, the malicious worker recognition accuracy and reliability are improved, and quantitative evaluation is conducted on the quality of worker sensing data.

Description

Technical Field

[0001] The present invention relates to the field of mobile crowd intelligence sensing technology, and in particular to a mobile crowd intelligence sensing trusted worker selection and task allocation method. Background Art

[0002] With the widespread popularity of mobile terminal devices and the rapid development of sensing technology, mobile crowd sensing, as an emerging data collection paradigm, has gradually become a key data collection method in smart cities, environmental monitoring, traffic management and other fields.

[0003] Mobile crowd sensing utilizes a large number of mobile devices (such as smartphones and smartwatches) carried by ordinary users to collect environmental data (such as air quality, noise levels, and traffic conditions). Leveraging network communication technologies, this data is transmitted to the cloud for centralized analysis, processing, and application-specific applications. High-quality service applications rely on high-quality data. However, in mobile crowd sensing platforms, malicious workers often falsify or fabricate data to obtain higher compensation. Furthermore, some trusted workers may upload inaccurate data for profit during the sensing process, turning themselves into malicious workers. This severely impacts the data quality of mobile crowd sensing systems. Existing technologies have made some progress in improving the data quality and operational efficiency of mobile crowd sensing platforms through methods such as incentive mechanism design, data fusion algorithms, and task allocation optimization.

[0004] However, research on identifying and preventing malicious workers remains insufficient, and there is a significant disconnect between relevant findings and actual platform applications. Currently, much research focuses on improving worker engagement and data reliability, while limited attention is paid to systematic technical solutions for identifying and addressing malicious workers. This technological gap can lead to significant vulnerability of platforms to malicious behavior, hindering their overall stability and sustainable development. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for selecting and assigning trusted workers in mobile crowdsensing, aiming to curb malicious workers in mobile crowdsensing from submitting low-quality data.

[0006] To achieve the above objectives, the present invention provides a mobile crowd-sensing trusted worker selection and task allocation method, comprising the following steps:

[0007] The mobile crowd-intelligence perception platform senses data based on the mobile smart devices carried by workers and obtains perception data;

[0008] The mobile crowd-sensing platform estimates the true value of the sensing data based on the principle of minimizing data variance;

[0009] Calculating the deviation of the mean and variance of the perception data of each worker relative to the standard data based on the true value estimation, and using this difference to conduct a credible worker qualification review;

[0010] Based on the set of trusted workers, the mobile crowd-sensing platform further evaluates the contribution value of each worker and quantifies their actual performance in the sensing task;

[0011] In the multi-round perception task allocation process, a reverse auction mechanism is used to dynamically adjust workers' bids, comprehensively considering the individual contribution value of workers and the size of virtual points to achieve effective task allocation.

[0012] Among them, there are certain differences in the accuracy and consistency of the perception data. In order to quantify the data differences, the mobile crowd intelligence perception platform assumes that the data collected by workers obeys the normal distribution.

[0013] The specific method in which the mobile crowd-sensing platform estimates the true value of the perception data based on the principle of minimizing data variance is as follows:

[0014] The mobile crowd intelligence sensing platform assigns a weight to each worker based on the principle of minimizing data variance;

[0015] The mobile crowd-intelligence perception platform performs a weighted summation on the expected value of each worker's data to infer the true value of the current perception task.

[0016] Among them, the degree of deviation can effectively reflect the accuracy of the worker's data and serve as the basis for evaluating its credibility. If a worker's data deviation exceeds the accuracy threshold set by the platform, he or she will be judged as an untrustworthy worker and included in the unselected worker group, and will not participate in subsequent perception tasks for the time being; otherwise, he or she will be regarded as a trustworthy worker, included in the selected worker group, and continue to participate in the subsequent data collection and task allocation process.

[0017] Among them, the contribution value is positively correlated with the accuracy of the workers' perception data. The higher the accuracy, the greater the effect on improving the overall data quality of the platform, and the contribution value also increases accordingly.

[0018] The present invention provides a method for selecting and assigning trusted workers for mobile crowd-sensing. The mobile crowd-sensing platform uses the mobile smart devices carried by workers to perform data sensing and obtain perception data. The platform then estimates the true value of the perception data based on the principle of minimizing data variance. Based on the true value estimation, the platform calculates the deviation of the mean and variance of each worker's perception data relative to the standard data, and uses this difference to verify the trusted worker qualifications. Based on the trusted worker set, the platform further evaluates each worker's contribution to quantify their actual performance in the perception task. During multiple rounds of perception task allocation, a reverse auction mechanism is used to dynamically adjust worker bids, comprehensively considering individual worker contributions and virtual points to achieve efficient task allocation. This method uses the mobile crowd-sensing platform to perform data sensing based on the mobile smart devices carried by workers, thereby obtaining perception data. The platform then estimates the true value of the perception data based on the principle of minimizing data variance. Based on the true value estimation and the Wasserstein distance, the platform compares the difference between each worker's perception data and the standard data, and uses this difference to verify the trusted worker qualifications. Based on the set of trusted workers, each worker's contribution is further evaluated to quantify their actual performance in the perception task. During multiple rounds of perception task allocation, a reverse auction mechanism is used to dynamically adjust worker bids, comprehensively considering individual worker contributions and virtual credits to achieve efficient task allocation. Furthermore, a reasonable virtual credit is set to retain workers who fail tasks. This method effectively reduces the interference of noise and outliers, significantly improving the platform's stability and robustness in complex and dynamic data environments. The Wasserstein distance is introduced to measure the differences in feature distributions between malicious and trusted workers. Compared to traditional similarity metrics (such as Euclidean distance and KL divergence), this method more comprehensively captures the underlying differences in data distributions, significantly improving the accuracy and reliability of malicious worker identification and quantitatively assessing the quality of worker perception data. This model helps select the optimal worker combination that maximizes contribution under certain constraints, thereby improving the quality of selected workers. By integrating the reverse auction mechanism with the contribution value assessment method, the platform not only effectively reduces the cost of completing perception tasks but also drives workers to improve task completion quality through bidding, thereby achieving dual optimization of data quality and platform performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 It is an overall flowchart of worker recruitment and task allocation driven by high-quality data.

[0021] Figure 2 It is a flowchart of the truth value inference process.

[0022] Figure 3 It is a flowchart of the qualification review process.

[0023] Figure 4 It is a flowchart of the worker selection and reverse auction process.

[0024] Figure 5 It is a schematic diagram of the Wasserstein distance corresponding to different workers.

[0025] Figure 6 This is a comparison diagram of the total contribution values ​​of different methods.

[0026] Figure 7 This is a flowchart of a mobile crowd-sensing trusted worker selection and task allocation method provided by the present invention.

[0027] Figure 8 It is a flowchart of a specific method for the mobile crowd-intelligence perception platform to estimate the true value of the perception data based on the principle of minimizing data variance. DETAILED DESCRIPTION

[0028] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0029] See also Figures 1 to 8 The present invention provides a mobile crowd-sensing trusted worker selection and task allocation method, comprising the following steps:

[0030] The S1 mobile crowd-intelligence perception platform senses data based on the mobile smart devices carried by workers and obtains perception data;

[0031] In an embodiment of the present invention, the mobile crowd-intelligence perception platform relies on the mobile smart devices carried by workers to collect data. However, due to the heterogeneity of device types and sensor configurations, the collected data may vary in accuracy and consistency. Considering the high-frequency operating characteristics of mobile device sensors, workers typically collect multiple pieces of data at once when performing perception tasks, forming data subsets. The data subsets of all workers together constitute a complete data set. To quantitatively analyze the differences between the perception data of different workers, it can be assumed that their data follows a normal distribution. The platform calculates the mean and variance of each worker's data, and then constructs a preliminary statistical distribution of the perception data.

[0032] S2 The mobile crowd-sensing platform estimates the true value of the perception data based on the principle of minimizing data variance;

[0033] Specific method:

[0034] The mobile crowd intelligence perception platform described in S21 assigns a weight to each worker based on the principle of minimizing data variance;

[0035] In this embodiment of the present invention, the smaller the variance of a worker's perception data, the more stable and reliable the data. Therefore, the weight of the mean of their data in the expected value of the overall data set should be increased accordingly. A larger weight indicates more stable and reliable data. Therefore, workers with smaller variances are considered to provide more accurate perception information and contribute more to the final true value.

[0036] The mobile crowd-intelligence perception platform described in S22 performs a weighted summation on the expected value of each worker's data to infer the true value of the current perception task.

[0037] In the embodiment of the present invention, the true value serves as the optimal estimate of the overall data, which is not only used for subsequent data analysis and decision support, but also serves as an important basis for perception task allocation.

[0038] S3 calculates the deviation of the mean and variance of the perception data of each worker relative to the standard data based on the true value estimation, and uses this difference to conduct a credible worker qualification review;

[0039] In this embodiment of the present invention, after completing the truth inference, the platform further compares the deviation of the mean and variance of each worker's perception data relative to the standard data. This deviation effectively reflects the accuracy of the worker's data and serves as a basis for assessing its credibility. If a worker's data deviation exceeds the accuracy threshold set by the platform, they are deemed untrustworthy and placed in the unselected worker group, temporarily excluding them from participating in subsequent perception tasks. Otherwise, they are deemed trustworthy and included in the selected worker group, continuing to participate in the subsequent data collection and task assignment process.

[0040] S4 Based on the trusted worker set, the mobile crowd-intelligence perception platform further evaluates the contribution value of each worker and quantifies their actual performance in the perception task;

[0041] In this embodiment of the present invention, after completing the screening of trusted workers, the platform evaluates the contribution value of each worker in the selected worker group to quantify their actual performance in the perception task. The contribution value is positively correlated with the accuracy of the worker's perception data. Higher accuracy contributes more to the platform's overall data quality, and the contribution value increases accordingly. Taking into account constraints such as budget, the platform selects the subgroup of workers with the highest total contribution value from the trusted worker group as the final subgroup to perform the perception task.

[0042] During the multi-round perception task allocation process, S5 uses a reverse auction mechanism to dynamically adjust workers' bids, comprehensively considering the workers' individual contribution values ​​and the size of virtual points to achieve effective task allocation.

[0043] In this embodiment of the present invention, the platform uses a reverse auction mechanism to dynamically adjust workers' bids and virtual points during multiple rounds of sensing tasks. If a worker loses a bid, the platform allocates virtual points to improve their competitiveness in subsequent bidding rounds, effectively retaining them. During the auction, successful bidders have their virtual points reset to zero and their bids adjusted accordingly. Unsuccessful bidders receive a certain amount of virtual points, which reduces their bids and increases their chances of winning in the next round.

[0044] To better understand the present technical solution, the following examples are provided for further explanation:

[0045] The symbols used in the present invention are shown in Table 1.

[0046] Table 1 Symbols

[0047]

[0048]

[0049] Task model: This paper models the i-th task in the perception task set as a triplet t i ={B i ,W i ,D i}, where B i represents the budget of the task, W i represents the set of workers assigned to the task, D i Represents the data set submitted by all workers in the task.

[0050] Worker model: For any perception task, you need to first define an initial set of workers Each of the workers It can be specifically represented as a quintuple Indicates that the worker is in task t i The data set collected in represents the weight of its perception data, Represents the virtual points allocated by the platform to workers, represents the worker's bid, Represents the cost for a worker to perform a task.

[0051] This paper focuses on identifying and preventing malicious workers in a mobile crowd-sensing platform, which requires a detailed classification of workers. Workers are mainly divided into two categories: trustworthy workers and malicious workers. Trustworthy workers use heterogeneous sensors to collect data of different precisions. Due to the influence of equipment precision and environmental factors, their perception data has certain errors, which can usually be approximated as obeying a normal distribution ò~N(d,σ 2 ), where d is the true value of the perception task, σ 2 is the variance. Trusted workers are further categorized into excellent, average, and poor workers based on device accuracy, with their device accuracy decreasing. Malicious workers intentionally falsify data or exploit platform vulnerabilities for profit, often deviating from the true value. Malicious workers can be further categorized into random and cautious based on their behavioral characteristics.

[0052] Problem definition: In a platform environment containing multiple tasks, assume that each task t i There is a worker set W containing m workers i , the minimum number of workers required for the reverse auction of this task is s n , the budget of the task is B i Any worker w in the set j Provide perception data, and its data quality is expressed in contribution value measure, represents the worker's bid, Represents the cost of the worker performing the task. We use the binary variable x j To indicate whether the worker is selected, where x j =1 means selected, x j =0 means not selected. The present invention expects to maximize the quality of data collected by the platform H i , the optimization objective function is expressed as:

[0053]

[0054] The first constraint ensures that each worker's bid is not less than the cost of executing the task; the second constraint limits the total bid of the selected workers to not exceed the task budget B. i; The third constraint ensures that the number of workers participating in the auction is not less than the minimum bidding requirement s n .

[0055] The high-quality data-driven worker recruitment and task assignment method consists of five steps:

[0056] Step 1: Perception Data Collection and Statistics: Multiple workers use various heterogeneous devices to perform perception tasks and upload their perception data to the platform. Due to differences in device performance, the collected data exhibits a certain degree of heterogeneity. To quantify this data variability, the platform assumes that the data collected by workers follows a normal distribution and calculates the mean and variance of each worker's data set accordingly.

[0057] Step 2: Truth Value Inference: The platform assigns a weight to each worker based on the principle of minimizing data variance. A larger weight indicates more stable and reliable data. The platform then takes a weighted sum of the expected values ​​of each worker's data to infer the true value of the current perception task.

[0058] Specifically, for each worker w ji , the data set collected is D ji ={d j1i ,d j2i ,…,d jλi}. Calculate the mean of the worker data The formula is as follows:

[0059]

[0060] The mean reflects the central location of the worker data and is used to preliminarily assess the central tendency of the data.

[0061] Based on the mean Calculate the variance of worker data The formula is as follows:

[0062]

[0063] Variance reflects the degree of fluctuation of the data. The smaller the variance, the more stable the data and the higher the credibility of the workers.

[0064] In order to minimize the overall variance between the true value and the worker data, we use the Lagrange multiplier method to calculate the weight k of each worker in the task ji , the formula is as follows:

[0065]

[0066] Weight k ji and variance Inversely proportional to , the worker with smaller variance has a higher weight, indicating that its data contributes more to the true value estimation.

[0067] Based on the weight k ji , the mean of the worker data Perform weighted summation to obtain the true value The formula is as follows:

[0068]

[0069] At the same time, calculate the variance corresponding to the true value The formula is as follows:

[0070]

[0071] The above truth value and the corresponding variance It serves as standard data and provides a basis for subsequent distribution difference measurement and malicious worker detection.

[0072] Step 3: Trusted Worker Screening: The platform compares each worker's submitted data with the standard data. If the difference is less than a threshold set by the platform, the worker's data is deemed trustworthy and included in the trusted worker pool; otherwise, it is removed. This step completes the initial screening of workers' perception capabilities.

[0073] Specifically, the difference between each worker's perception data and the standard data is calculated using the Wasserstein distance calculation formula:

[0074]

[0075] Among them, (x) + =max(0,x) represents the ReLU function. The formula can be divided into two parts: one is the absolute value of the difference between the mean of the worker data and the true value, reflecting the difference between the data collected by the worker and the true data; the other is the ReLU function of the variance between the two, reflecting whether the data collected by the worker is within the tolerable error range. If the error in the data collected by the worker is less than the acceptable error, the function value is zero. The smaller the function value, the closer the characteristics of the collected raw data are to those of the standard data.

[0076] The worker set is filtered based on the difference value obtained in the previous step and the threshold set by the platform. Workers with a difference value greater than the threshold are added to the unselected worker group, and vice versa.

[0077] Step 4: Contribution Value Evaluation and Worker Subset Optimization: Based on the set of trusted workers, the platform further evaluates each worker's contribution value, which is related to their data quality and historical performance. Within the constraints of budget constraints and a virtual points mechanism, the platform selects the worker subgroup with the highest total contribution value.

[0078] Specifically, the platform checks whether the number of workers in the selected worker group meets the auction opening conditions. If the number is insufficient, some workers from the unselected worker groups will be selected to participate in the auction.

[0079] Based on the Wasserstein distance between each worker's perception data and the standard data, the contribution value is used to quantitatively evaluate the performance of workers participating in the perception task. The contribution value calculation formula is as follows:

[0080]

[0081] The smaller the Wasserstein distance between the worker's perception data and the standard data, the greater the corresponding contribution value. For multi-round tasks, the worker's contribution value depends on all its perception task records, and the calculation formula is as follows:

[0082]

[0083] The platform uses a dynamic programming algorithm to select the worker subgroup with the largest total contribution value under the budget constraint. Assume that P i,B It represents the maximum total contribution value when the platform selects the first i workers and the current total bid does not exceed B. The dynamic programming transfer equation is as follows:

[0084]

[0085] The algorithm can determine the corresponding worker subgroup while obtaining the maximum total contribution value, and assign tasks to the workers of the subgroup for execution.

[0086] Step 5: Task Assignment and Reverse Auction Incentive Mechanism: The platform assigns the current task to the optimal subgroup of workers selected in Step 4. For unselected workers, the platform awards a certain number of virtual points to increase their chances of winning the next round of bidding, thereby motivating and retaining them. Successful bidders have their virtual points reset to zero. During multiple rounds of task execution, the platform introduces a reverse auction mechanism to dynamically adjust worker bids: winning workers bid upward, while unsuccessful workers bid downward, thereby promoting fair competition and increasing long-term participation.

[0087] Specifically, for workers who fail to obtain task assignments, the platform will give a certain number of virtual points as compensation. The virtual points for compensation are shown in the following formula:

[0088]

[0089] Among them, θ and β∈(0,1) are constants, f j represents the number of failures of the worker in the continuous auction. Assuming that the number of failures tends to infinity, the upper bound of the virtual integral is:

[0090]

[0091] These virtual points can be used in the next round of bidding, thus incentivizing workers to continue participating. For workers who successfully obtain a task assignment, the platform will reset their virtual points to zero to ensure fairness and incentives. During multi-round task assignments, the platform uses a reverse auction mechanism: workers who win a task in a given round will increase their bids in the next round, while workers who do not win will lower their bids.

[0092] Figure 5 The experiment involved five different types of workers, each consisting of 20 workers. The first three types of workers were trustworthy and categorized as poor, average, and excellent based on their sensor accuracy. The last two types were random and cautious malicious workers, whose submitted data often deviated from the true value. The platform used the Wasserstein distance (shown by the red dashed line) to conduct qualification checks and successfully identified and excluded the last two types of malicious workers, completing the worker screening process.

[0093] This invention proposes a trusted worker selection and task assignment method that includes truth value inference, qualification review, and worker quality assessment. It aims to effectively identify and suppress multiple types of malicious workers, while motivating high-quality trusted workers to actively participate in the perception task assignment based on the reverse auction mechanism. In order to verify the effectiveness of this method, this paper designed and carried out a total contribution value comparison experiment, and compared the performance with a variety of typical methods. The comparison methods include: a greedy algorithm based on contribution value and bid ratio, a maximum contribution value priority algorithm, and a hybrid algorithm, while in other stages, the worker selection mechanism proposed in this paper is combined. A total of 200 workers were selected in the experimental setting, covering multiple types, and the number of each type of workers was balanced. The experimental results are as follows Figure 6 The vertical axis represents the total worker contribution of each algorithm in each round of the task. Trend analysis shows that the contribution of each algorithm decreases slightly in the early stages of the task, primarily due to large fluctuations in worker bids and contributions during the initial reverse auction. As the number of auction rounds and task assignments increases, the total contribution of our method gradually stabilizes, outperforming other traditional methods.

[0094] Specifically, although the greedy algorithm failed to surpass the proposed method, its overall performance was still better than the hybrid algorithm. This phenomenon can be explained by formula (8): the worker contribution value is positively correlated with the quality of their data, so choosing workers with high cost-effectiveness is more likely to obtain high-quality perception data. However, the greedy algorithm is prone to falling into local optimality in task allocation and is difficult to globally optimize the worker combination. The hybrid algorithm is limited by its weak ability to identify malicious workers in the truth inference stage, which makes it easier to include malicious workers in subsequent selections, thereby reducing the overall contribution value. The maximum contribution value algorithm lacks a constraint mechanism for worker bids and is prone to selecting malicious workers with high contribution values ​​but high costs, ultimately resulting in poor overall benefits. In summary, the proposed method shows better comprehensive performance in ensuring data quality, controlling budget costs, and suppressing malicious behavior.

[0095] The above disclosure is only a preferred embodiment of the mobile crowd-sensing trusted worker selection and task allocation method of the present invention. Of course, this cannot be used to limit the scope of the rights of the present invention. Ordinary technicians in this field can understand that implementing all or part of the processes of the above embodiment and making equivalent changes in accordance with the claims of the present invention still fall within the scope of the invention.

Claims

1. A mobile crowd-sensing trusted worker selection and task allocation method, characterized by: The following steps are involved: The mobile crowd-intelligence perception platform senses data based on the mobile smart devices carried by workers and obtains perception data; The mobile crowd-sensing platform estimates the true value of the sensing data based on the principle of minimizing data variance; Calculating the deviation of the mean and variance of the perception data of each worker relative to the standard data based on the true value estimation, and using this difference to conduct a credible worker qualification review; Based on the set of trusted workers, the mobile crowd-sensing platform further evaluates the contribution value of each worker and quantifies their actual performance in the sensing task; In the multi-round perception task allocation process, a reverse auction mechanism is used to dynamically adjust workers' bids, comprehensively considering the individual contribution value of workers and the size of virtual points to achieve effective task allocation.

2. The mobile crowd-sensing trusted worker selection and task assignment method according to claim 1, characterized in that; The perception data has certain differences in accuracy and consistency. To quantify the data differences, the mobile crowd-sensing platform assumes that the data collected by workers follows a normal distribution.

3. The mobile crowd-sensing trusted worker selection and task allocation method according to claim 1 is characterized in that ； The specific method in which the mobile crowd intelligence perception platform estimates the true value of the perception data based on the principle of minimizing data variance is as follows: The mobile crowd intelligence sensing platform assigns a weight to each worker based on the principle of minimizing data variance; The mobile crowd-intelligence perception platform performs a weighted summation on the expected value of each worker's data to infer the true value of the current perception task.

4. The mobile crowd-sensing trusted worker selection and task assignment method according to claim 1, It is characterized by: The degree of deviation can effectively reflect the accuracy of the worker's data and serve as the basis for evaluating its credibility. If a worker's data deviation exceeds the accuracy threshold set by the platform, they will be judged as untrustworthy workers and included in the unselected worker group, and will not participate in subsequent perception tasks for the time being; otherwise, they will be regarded as trustworthy workers, included in the selected worker group, and continue to participate in the subsequent data collection and task allocation process.

5. The mobile crowd-sensing trusted worker selection and task allocation method according to claim 1, It is characterized by: The contribution value is positively correlated with the accuracy of the workers' perception data. The higher the accuracy, the greater the impact on improving the overall data quality of the platform, and the contribution value also increases accordingly.

Images