CN114971047A - Mobile crowd sensing-oriented user collaborative optimization method - Google Patents

Abstract

In the existing mobile crowd sensing, how to quickly assign sensing tasks to optimal executing users on the premise of ensuring sensing quality and reducing costs is the focus of the research. To solve this problem, this paper proposes a choice-aware user CSSA optimization method based on the sparrow search algorithm. The method firstly models the perceived user, and proposes the concept of user fitness. The basic information of the user is classified into four aspects: location, power, equipment and reputation, and the fitness is calculated in turn. Secondly, according to these fitness values, the priority of the perceived user is comprehensively considered, the users are classified according to the user priority, the intelligent optimization algorithm is used to simulate the process of the user completing the task, and finally the optimal user suitable for the perceived task is selected. Through the comparison experiments between the algorithm proposed in this paper and other optimization algorithms in the same environment, the results show that it has higher performance in solving the task assignment problem.

Description

A User Collaborative Optimization Method for Mobile Crowdsensing

technical field

The invention belongs to the field of mobile crowd intelligence perception, and in particular relates to a user collaborative optimization method based on an intelligent optimization algorithm.

Background technique

As a new perception paradigm, Mobile Crowd Sensing (MCS) combines artificial intelligence with the Internet of Things to utilize mobile users with smart devices (including mobile phones, smart cars, wearable devices, etc.) Execute the sensing task requested by the task requester. The inherent mobility of mobile users makes Mobile Crowd Sensing (MCS) a common platform that can replace or complement existing static sensing infrastructure. Through these data collection, MCS plays a pivotal role in the fields of traffic monitoring, environmental phenomenon observation and mobile medical treatment. The perception system is composed of a platform, a task publisher and a participant user. The task publisher can be a machine or a person. Participating in a perception task requires the participant's time and the device's perception, computing and communication resources. To stimulate broad user participation, some MCS systems offer rewards in return for valuable sensory reports submitted by users. Incentives can take different forms, including money, entertainment through games, or services provided by the system. These task budgets are allocated to participants completing sensing tasks based on the time spent sensing data or the quality of the sensing reports delivered. Therefore, in order to improve the data quality of the perception task and reduce the task cost, selecting the appropriate user is an extremely important prerequisite.

In the mobile crowd sensing system, users participate in tasks and use mobile sensing devices to collect data and upload it to the platform, so as to complete the data collection. The right user has a key impact on the completion of the entire task. In the task assignment process, finding the fit between the user and the task by analyzing the attributes of the user can be regarded as finding an optimal user selection scheme that satisfies several constraints in a random environment. The previous methods often focus on increasing the number of users and finding the optimal solution for a single user, while the task of crowdsensing requires multiple users to complete. Collecting the same type of data by multiple users can lead to increased costs and data redundancy.

Task allocation is one of the main research directions of MCS, which collects perception data from mobile users and obtains perception results to complete tasks. In general, there are many factors that affect the quality of tasks, including the amount of data collection, task duration, and task spatiotemporal coverage. More sensory data collected generally means higher mission quality; suitable mission spatiotemporal coverage also generally means higher mission quality. There are also many factors that affect the cost of tasks, including user recruitment costs, user mobility costs, and data transfer costs. In order to improve the quality of perceived data and reduce the task incentive budget, tasks can be allocated to users by analyzing the matching degree between users and tasks, and analyzing the user's fitness value.

In recent years, with the rapid development of SI (swarm intelligence) optimization algorithms, more and more intelligent optimization algorithms with superior performance have been proposed, such as gray wolf optimization (GWO) algorithm, whale optimization algorithm (WOA), and gravity search algorithm. (GSA) et al. At present, it has gradually become another main method to solve the task assignment problem in crowdsensing, and there are already many methods that combine SI optimization algorithm with the task assignment of crowdsensing.

SUMMARY OF THE INVENTION

There is still a lot of research space for multi-user collaboration to improve sensing performance. In this paper, combined with the excellent performance of the sparrow search algorithm in the intelligent optimization algorithm, a new framework CSSA (Crowd sensing SparrowSearch Algorithm) is proposed for sensing tasks. User selection to improve perceived data quality and distribution efficiency. As shown in Figure 1, by cooperating with multiple users and using their own sensing devices to collect data for aggregation, the quality of the collected sensing data can be further improved.

This paper proposes a user selection framework, which comprehensively considers the impact of multiple user attributes on task-aware data quality, as well as the collaborative cooperation between multiple users. The goal of this strategy is to improve the quality of perceptual data while ensuring the task cost, and also consider the time selected by the user. In this paper, the influencing factors are mainly attributed to the following aspects: the distance relationship between the user and the task; the types of user equipment sensors Fit to the task; how well the user's device is performing; the user's reputation.

Due to the high complexity of the problem to be solved, the concept of fitness is proposed for the above influencing factors, and the fitness value is used as a reference to measure user priority and matching degree. The data information to be obtained is divided into the user location information obtained through positioning, the remaining battery power of the device uploaded through the user device status, and the device sensor type information. Estimate, so as to complete the control of task data quality, and analyze whether the user is suitable for completing the current task. These fitness values are used as an important reference standard for intelligent optimization algorithms, as shown in Figure 2.

The mobile crowdsensing platform is composed of many sensing users, U represents the set of users who can participate in the system when the task is released, and F=(f ₁ , f ₂ , f ₃ ,..., f _n ) represents the current participation of all users on the task The fitness value, sorted by the size of this value, the user set is . U=(u ₁ , u ₂ , u ₃ ,…, u _n ) divides users into explorers and followers in proportion, and performs task perception according to different responsibilities, and finally selects suitable perception users according to the intelligent optimization algorithm.

At the same time, in order to improve the efficiency of user selection and data quality, this paper adopts the way of participant collaboration to complete the task. Analyze the type of data that needs to be sensed for the current task, and let the user be responsible for the types of information they can provide according to the user's sensor, and then the platform combines the content uploaded by multiple users to summarize and analyze to complete the collection of sensing data. The algorithm makes use of the characteristics of mutual cooperation between groups, which is conducive to the cooperation of participant user groups in crowd sensing to complete the sensing task, and at the same time, comprehensively analyzes the matching degree of the perceived user's location, equipment and other attributes with the user. In crowdsensing, the user selection strategy based on the analysis of multiple influencing factors and the cooperation between users can improve the quality of perception data and task allocation time, which greatly improves the performance of the crowdsensing platform.

Description of drawings

FIG. 1 is a schematic diagram of enhancing user collaboration.

FIG. 2 is an overall flow chart of the present invention.

FIG. 3 is a diagram showing the relationship between sensor types and coverage.

Detailed ways

Location fitness, by calculating whether the location relationship between the task and the user matches, when the user is far away from the task, the distance the user needs to move will increase, thereby increasing the task completion time and higher task cost. In view of this situation, this paper introduces the concept of location fitness, and compares the distance between the user and the task as an indicator to measure the fitness. The calculation formula is as follows: locFit _i =(d _max -d _i )/ d _max (d _i ≤d _max ) where d _i is the distance between the user and the task, and the Euclidean distance is used for calculation. When the user's original path passes through the task-aware area, as the user gets closer to the task, the positional fitness between the user and the task will continue to improve, and the user's intention to perform the task will become more and more Therefore, the user may actively increase the speed and reduce the movement time, thereby reducing the completion speed of the task and improving the overall efficiency.

Power fitness is a consideration for the remaining power of the mobile device, because in the process of completing the task, it is necessary to continuously use the sensor to sense data, and upload the sensed data to the platform in real time through the communication network. This process needs to continuously consume the power of the mobile device. Therefore, the battery fitness is also an important consideration parameter, which determines whether the mobile device has the ability to complete the sensing data collection. When the battery is low, the user is likely not willing to participate in the task. The calculation of the power fitness is standardized according to the data in [0,1] expressed as the remaining power percentage, and the formula is as follows: eltFit _i =CE _i /E _i .

Equipment fitness, depending on the types of data that the task needs to perceive, the requirements for sensing user equipment are also different. Therefore, it is necessary to consider the availability of sensors required on the sensing equipment of candidate participants, which will significantly affect the reliability of the data. and quality of sensing results. In order to quantify the matching degree, device fitness is used to represent the matching degree of hardware capabilities between users and tasks, and tasks are assigned to the best users in turn, which can not only reduce the cost of tasks, but also improve data quality, device information and perception. The degree of content matching is shown in Figure 3. Among them, SameSen _i =UserSen _i ∩TaskSen is the sensor overlap set, which represents the overlapped part of the user equipment sensor type set and the task-aware content set. Some of them are indispensable sensors, and other non-critical types are the better the degree of coincidence, so the degree of sensor coincidence is: equFit _i =η ₁ *(Sig _i /Sig _max )+η ₂ *(SameSen _i / TaskSen), each sensing user does not need to have all the required sensing sensor types, and the collection of various types of sensing data can be completed through the cooperation among multiple users.

Reputation fitness, so the user's reputation is evaluated based on their previous participation in the perception task. Evaluation criteria include their commitment to completing the tasks assigned to them and their ability to successfully complete those tasks. The following formulas were used to evaluate the reputation parameters of each participant respectively: CT _i =Spt _i /Ct _i , SCT _i =Sct _i /Ft _i . Among them, Spt _i represents the set of tasks that the user participated in, Ct _i represents the set of tasks selected by the user, Sct _i represents a set of tasks successfully completed by the user, and Ft _i represents the set of tasks completed by the user. Since these two parameters vary in a relative manner, their collective mean is used to give a participant's overall reputation. Therefore, the calculation formula of the individual reputation fitness of each user is as follows: repFit _i =sqrt(CT _i *SCT _i ).

When the sensing task is released, the sensing users who have the willingness to participate in the task report to the platform. The set of participants is U=(u ₁ , u ₂ , u ₃ ,…, u _n ). The platform first uses the formula F _i according to user attributes and historical data. =locFit _i +eltFit _i +equFit _i +repFit _i to calculate the fitness value F _i of each user. Initialize the positions of all users in the optimization algorithm. In order to improve the global search ability of the algorithm and avoid the reduction of population diversity in the later iteration, the method of chaotic mapping is used instead of random number generation to make the positions more random and regular. The formula is as follows : u _i+1 =u _i *μ*(1-u _i ), μ∈[3.5,4], _u i∈(0,1), the positions of n participants in the d dimension are recorded by a matrix, At the same time, the corresponding fitness values are calculated respectively.

,d，表示当前所在的维度，U_i,j ^t表示第i名感知用户在迭代t时的第j维的位置，iter_max是迭代次数最多的常数。α∈（0,1）是一个随机数，Pn（Pn∈[0,1]）和CT（CT∈[0.5,1.0]）分别代表着子区域的任务完成度和子区域的完成阈值。Q是一个服从正态分布的随机数，I是一个1*d维的矩阵，其中每一个元素都为1。当Pn<CT时，意味着当前的区域任务完成人数未达到阈值，当前搜索者可以继续在子区域内完成任务；当 Pn≥CT 时，意味着当前区域参与者数量满足任务完成的要求，所有探索者需要到其他子区域完成感知任务。Users are _sorted by fitness value Fi, and users are divided into two different identities, explorer and follower, by setting value Pr. Record the identity of the current user through the array ID[n], 0 for followers and 1 for explorers. Among them, the explorer has a high task fitness, so he is responsible for the main perception data collection, as the main completer of the perception task; while the follower corresponds to the explorer's identity, due to the low fitness, follow the explorer to collect data, Responsible for supplementing data and refining perception data types. For example, when the value of Pr is set to 0.2, the ratio of explorers and followers is 2:8. Among them, the position of the explorer is updated as follows: when Pn<CT, U _i,j ^t+1 = U _i,j ^t *exp(-i/α*iter _max ); when Pn≥CT, U _i,j ^t+1 =U _i,j ^t +Q·L. Among them, t represents the current iteration number, j=1,2,

,d, represents the current dimension, U _i,j ^t represents the position of the i-th perception user in the j-th dimension during iteration t, and iter _max is the constant with the most iterations. α∈(0,1) is a random number, and Pn (Pn∈[0,1]) and CT (CT∈[0.5,1.0]) represent the task completion degree of the sub-region and the completion threshold of the sub-region, respectively. Q is a random number that obeys a normal distribution, and I is a 1*d-dimensional matrix, where each element is 1. When Pn<CT, it means that the number of completed tasks in the current region has not reached the threshold, and the current searcher can continue to complete tasks in the sub-region; when Pn≥CT, it means that the number of participants in the current region meets the requirements for task completion, all Explorers need to go to other sub-areas to complete perception tasks.

As for the follower, the follower explorer moves around to assist in completing the task. When the current explorer's fitness value is found to be lower than his own, a competition relationship occurs, and the identities of the two will change. The current user ID[i]'s The value will also be inverted, when i>n/2, the position U _i,j ^t+1 =Q*exp((U _worst ^t -U _i,j ^t )/i ² ); when i≤n/2 , U _i,j ^t+1 = U _p ^t+1 +|U _i,j ^t - U _p ^t+1 |*A ⁺ *L. Among them, U _p is the best position occupied by the explorer, and U _worst represents the current global worst position. A represents a 1*d dimensional matrix where each element is randomly assigned a value of 1 or -1, and A ⁺ =A ^T (AA ^T ) ^-1 . When the fitness value is appropriate, move near the current explorer's location; when i>n/2, it means that the follower has a very low fitness value in the current perceived user queue and cannot complete the current task, and then searches for Perception tasks in other regions.

In addition, some of the task participants are selected as monitors. They are randomly generated in the crowd and are responsible for preventing users from concentrating on the optimal sensing position and falling into a local optimum. 10%~20% of the sensing users are selected as monitors. When f When _i ≠f _g , U _i,j ^t+1 = U _p ^t+1 +β*(U _worst ^t - U _best ^t ); when f _i =f _g , U _i,j ^t+1 = U _p ^t+1 +β*(U _i,j ^t - U _best ^t ). Among them, U _best , and U _worst represent the current global optimal position and the worst position, respectively. β is used as a step size control parameter, which is a normal distribution random number that obeys the mean value of 0 and the variance of 1. f _i is the adaptation of the current sparrow individual. degree value, f _g is the current global best fitness value. In order to prevent falling into the local optimum, if the user is in the optimum position, it escapes to a random position between the optimum position and the worst position, otherwise it escapes to a random position between itself and the random position.

Based on the above user selection model, under the premise of considering data quality, task cost and allocation time, the user selection problem in this paper can be defined as: F=max∑ _i=1 ^k [(equFit _i +eltFit _i )*repFit _i ], Satisfy min∑ _i=1 ^k (d _i ), ∑ _i=1 ^k M _i =I

Among them, I is an all-one matrix of 1*j, which is used to control the sum of all user sensor types to meet the types required by the task, where j is the number of types required by the task, ∑ _i=1 ^k (d _i ) is all users The sum of the distance to the task, by minimizing the total distance, can reduce the perceived user's moving distance to complete the task, thereby reducing the cost. The purpose of optimizing the algorithm is to find the maximum value of the objective function, that is, to find the user that best meets the task requirements, thereby improving the data quality.

The above-mentioned implementation method is a preferred embodiment of the present invention, but the implementation of the present invention is not limited by the above-mentioned method, and any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention , all should be equivalent replacement modes, and all are included in the protection scope of the present invention.

Claims (3)

1. A user collaborative optimization method for mobile crowd-sensing, characterized in that it comprises a user fitness processing module, an iterative update module, and a perceived user selection module. 2. The mobile crowd-sensing user collaborative optimization method based on crowd intelligence optimization according to claim 1, characterized in that, by the degree of matching between the user and the task, position, power, equipment, reputation fitness are calculated respectively, Quantify and sort the user's priority to form the user's priority sequence for tasks. 3 . The mobile crowd intelligence sensing user collaborative optimization method based on crowd intelligence optimization according to claim 1 , characterized in that, through user cooperation and cooperation, the sparrow search algorithm is used to form the user selection optimal solution for the task. 4 .

Images