Mobile crowd sensing task allocation method based on mab and differential privacy protection

Abstract

The application discloses a kind of mobile crowd sensing task allocation methods based on MAB and differential privacy protection, its steps include:1, according to the sensing task information released by platform and the request information submitted by worker to build auction model, unknown worker's work quality is obtained in combination with multi-arm tiger machine mechanism;2, under the sensing task allocation framework based on reverse auction, in combination with differential privacy technology, the multi-arm tiger machine task allocation method based on EXP3-IX is proposed to determine the final winning worker;3, the untrusted platform lets the winning worker selected to execute subsequent tasks, and pays corresponding loss compensation to the worker, and calculates total work quality.The present application can effectively solve the information leakage problem between the worker and the untrusted platform under the premise of fully considering how to obtain unknown worker's sensing data, to protect the worker's loss privacy, so that the worker selected by the platform can have higher work quality.

Description

Technical Field

[0001] This invention relates to the field of mobile crowd sensing technology, specifically to a mobile crowd sensing task allocation method based on MAB and differential privacy protection. Background Technology

[0002] Mobile crowdsourcing sensing systems are human-centric, utilizing everyday smart devices (such as smartphones and tablets) as basic sensing units. Through conscious or unconscious collaboration with the internet, they form an interactive crowdsourcing sensing network, enabling the distribution of sensing tasks and data collection. The crowdsourcing sensing platform aggregates and analyzes the collected data to obtain aggregated results, ultimately completing large-scale, complex sensing tasks. The key to the crowdsourcing sensing network is that ordinary users are both direct producers and indirect users of sensing data; they provide sensing data to the platform while simultaneously enjoying services based on that data. This user-centric sensing model fully leverages the advantages of internet data dissemination. Compared to traditional sensing networks that require the pre-deployment of numerous sensors, crowdsourcing sensing directly utilizes existing sensing devices, offering advantages such as low data collection costs, wide coverage, and strong scalability. For workers, performing sensing tasks consumes resources, including computing power and batteries. To encourage more workers to participate, auctions become a reasonable way to incentivize participation and allocate tasks. In the framework of the cloud platform and workers, the cloud platform is considered the buyer, and the workers are considered the sellers, constructing an auction-based model. To maximize their utility, workers need to submit their real quotes. However, because the platform is honest and curious, it obtains the worker's real quote when they submit it, inferring their true quote from the final result. This leads to privacy leaks, reducing the incentive for both workers and requesters to participate. To address the privacy issue, researchers have primarily used three methods: data perturbation, cryptography, and anonymization. Data perturbation and cryptography are mainly used to encrypt user geolocation and perception data, while anonymization is primarily used to hide user identity information.

[0003] Most current mobile crowdsourcing privacy protection solutions operate under the premise of being able to obtain information about the actual work quality of workers. However, in real-world scenarios, workers often struggle to estimate their true work quality, and platforms also lack access to information about the work quality of workers participating in the system for the first time. Therefore, the system needs to consider how to allocate sensing tasks without knowing the workers' work quality, while simultaneously addressing the issue of user privacy breaches.

[0004] Purpose of the invention

[0005] To address the shortcomings of the existing technologies, this invention provides a mobile crowd-sensing task allocation method based on MAB and differential privacy protection. This method aims to effectively solve the problem of worker information leakage and protect worker privacy without knowing the worker's sensing data, thereby enabling the platform to select workers with higher work quality.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] The present invention discloses a mobile crowd-sensing task allocation method based on MAB and differential privacy protection, characterized by its application to an untrusted platform and n workers U = {u1, u2, ..., u}. i ,…,u n In a mobile swarm intelligence sensing environment composed of}, u i Let i represent the i-th worker, i = 1, 2, ..., n. The mobile swarm intelligence sensing task allocation method is performed according to the following steps:

[0008] Step 1: Build and initialize the auction model:

[0009] Step 1.1: Let the untrusted platform publish m rounds of tasks T = {t1, t2, ..., t...} j ,…,t m The total loss compensation is B, where t j Let represent the j-th round of tasks; the untrusted platform sets a privacy length β∈(0,0.1) to determine the confusion loss range of workers; let the actual work loss of n workers U be b={b1,b2,…,b i ,…,b n}, where b i Represents the i-th worker u i The actual working losses;

[0010] Step 1.2, any i-th worker u i After obtaining the privacy length β from the untrusted platform, the i-th worker u i The unit privacy length is determined to be Δβ;

[0011] Step 1.3: Define the i-th worker u i confusion loss set in, Represents the i-th worker u i The k-th confusion loss, and Among them, s k Let represent the parameter of the k-th confusion loss, and L represents the number of confusion losses;

[0012] Step 1.4: The untrusted platform uses equation (1) to construct the i-th worker u. i Obfuscation function:

[0013]

[0014] In equation (1), It is the kth confusion loss Mapped to the i-th worker u i Actual working losses b i The probability of; It measures the confusion loss of the kth digit. Mapped to the i-th worker u i Actual working losses b i The function of the distance between them, and △q is the global sensitivity, and Among them, b max b represents the maximum confusion loss in the confusion loss set of all workers. min Represents the minimum confusion loss in the confusion loss set of all workers; It is the i-th worker u i The y-th confusion loss; y∈[1,L];

[0015] Step 1.5, the i-th worker u i Download the obfuscation function from the untrusted platform and derive the obfuscation loss set based on the obfuscation function. probability distribution in, Represents the i-th worker u i The kth confusion loss The probability of being selected, according to the probability distribution From the confusion loss set Randomly select a confusion loss As the i-th worker u i The obfuscation and loss are then uploaded to the untrusted platform; rand∈[1,L];

[0016] Step 2: The untrusted platform uses a% of the total loss compensation B during the exploration phase to obtain the i-th worker u. i The quality of work; where a represents a parameter, and a∈[0,1];

[0017] Step 2.1: Initialize j = 1; Initialize i = 1;

[0018] Step 2.2: Define the total work quality of the (j-1)th round as... and initialize

[0019] Define the loss in the (j-1)th round as and initialize

[0020] Step 2.3, the i-th worker u i Execute the j-th round of task t j And obtain the i-th worker u i Execute the j-th round of task t j work quality Therefore, equations (2) and (3) are used to calculate the task t in the j-th round. j Overall work quality and the jth round of tasks t j loss

[0021]

[0022]

[0023] Step 2.4: Assign i+1 to i, and j+1 to j. Determine if j>n is true. If true, proceed to step 2.5; otherwise, return to step 2.3.

[0024] Step 2.5: Define the random probability p of the j-th round. j When p j When p < τ, execute steps 2.6 and 2.9. j When ≥τ, execute steps 2.7-2.9; where τ represents the threshold; and τ∈[0,1];

[0025] Step 2.6: Assign j%n to i, and select the i-th worker u. i Execute the j-th round of task t j ; where % represents the remainder;

[0026] Step 2.7: Calculate the task t of the j-th round according to formula (4). j Select the i-th worker u i probability

[0027]

[0028] In equation (4), Represents the i-th worker u i Execute the j-th round of task t j The weights after; when j = n+1, let

[0029] Step 2.8: Calculate the i-th worker u according to formula (5). i Execute task t in round j+1j+1 The weight after

[0030]

[0031] In equation (5), The parameter that controls the change in weight;

[0032] Step 2.9: Calculate using equations (2) and (3) and And judge If the condition is true, assign j+1 to j and execute step 3; otherwise, return to step 2.5 and execute sequentially.

[0033] Step 3: The untrusted platform uses the remaining loss compensation (1-a%)×B during the utilization phase to obtain the i-th worker u. i Work quality;

[0034] Step 3.1, the untrusted platform is based on probability. Select the i-th worker u i To perform subsequent tasks, an objective function with the goal of maximizing the total work quality Q is established using equation (6), and the constraint equations of the objective function are established using equation (7):

[0035]

[0036]

[0037] In equations (6) and (7), k represents the total number of tasks completed during the exploration phase. Represents the j-th round of task t j Selected workers Represents the j-th round of task t j workers Execute the j-th round of task t j The quality of work In the j-th round of task t j workers From the confusion loss set B * win A randomized confusion loss is selected from the data.

[0038] Step 3.2: The untrusted platform compensation completes the j-th round of task t. j workers loss

[0039] Step 3.3: After assigning j+1 to j, determine... Check if j≥m is true. If true, the task allocation is complete and step 4 is executed. Otherwise, return to step 3.1 and execute sequentially.

[0040] Step 4: Calculate the total work quality Q using equation (8):

[0041]

[0042] In equation (8), This represents the total work quality of the k rounds of tasks completed during the exploration phase.

[0043] The present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the mobile crowd sensing task allocation method, and the processor is configured to execute the program stored in the memory.

[0044] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the mobile swarm intelligence sensing task allocation method.

[0045] Compared with existing solutions, the advantages of this invention are as follows:

[0046] 1. This invention applies differential privacy technology within an auction-based mobile crowd intelligence perception framework. It obtains the work quality of workers through a multi-armed slot machine mechanism and probabilistically obfuscates the bidding methods of workers through an exponential mechanism, thereby achieving privacy protection, ensuring the security of workers' bidding information, and attracting more workers to participate.

[0047] 2. This invention uses differential privacy to protect the privacy information of workers on untrusted platforms. Compared with using encryption technology to achieve security, it reduces the computation and communication overhead of encryption algorithms and improves the efficiency of task allocation.

[0048] 3. In the task allocation process, this invention selects winning workers based on the EXP3 MAB strategy to participate in subsequent rounds of tasks, enabling the platform to achieve higher overall work quality within a limited budget. Attached Figure Description

[0049] Figure 1 This is a framework diagram of the present invention;

[0050] Figure 2 This is a flowchart of the main steps in this invention. Detailed Implementation

[0051] In this embodiment, a mobile crowd sensing task allocation method based on MAB and differential privacy protection is described, such as... Figure 1The diagram illustrates an application to an untrusted platform and n workers, U = {u1, u2, ..., u...}. i ,…,u n In a mobile swarm intelligence sensing environment composed of}, u i This represents the i-th worker, where i = 1, 2, ..., n, such as... Figure 2 As shown, the mobile crowd sensing task allocation method is performed according to the following steps:

[0052] Step 1: Construct and initialize the auction model: Based on the information submitted by the platform and workers, treat the untrustworthy platform as the buyer and the workers as the seller, thereby constructing an auction model;

[0053] Step 1.1: Have the untrusted platform publish m rounds of tasks T = {t1, t2, ..., t} j ,…,t m The total loss compensation is B, where t j Let represent the j-th round of tasks; the untrusted platform sets a privacy length β∈(0,0.1) to determine the range of worker confusion loss. Here, β = 0.02 is set. If the value of β is set too large, it will increase the platform's compensation; let the actual work loss of n workers U be b = {b1,b2,…,b}. i ,…,b n}, where b i Represents the i-th worker u i The actual work loss; the tasks here are of the same type, because the work quality of a worker performing the same type of task within a certain period of time will not deviate too much, which is used to ensure the determination of the work quality of unknown workers in subsequent steps.

[0054] Step 1.2, any i-th worker u i After obtaining the privacy length β from the untrusted platform, the i-th worker u i The unit privacy length is determined to be Δβ;

[0055] Step 1.3: Define the i-th worker u i confusion loss set in, Represents the i-th worker u i The k-th confusion loss, and Among them, s k Let represent the parameter of the k-th confusion loss, and L represents the number of obfuscation losses. Generally, L takes a value in [5,10], which ensures the randomness of obfuscation and reduces computational cost.

[0056] Step 1.4: Using formula (1) to construct the i-th worker u on an untrusted platform iObfuscation function:

[0057]

[0058] In equation (1), It is the kth confusion loss Mapped to the i-th worker u i Actual working losses b i The probability of; It measures the confusion loss of the kth digit. Mapped to the i-th worker u i Actual working losses b i The function of the distance between them, and △q is the global sensitivity, and Among them, b max b represents the maximum confusion loss in the confusion loss set of all workers. min Represents the minimum confusion loss in the confusion loss set of all workers; It is the i-th worker u i The y-th confusion loss; y∈[1,L];

[0059] Step 1.5, the i-th worker u i Download the obfuscation function from the untrusted platform and derive the obfuscation loss set based on the obfuscation function. probability distribution in, Represents the i-th worker u i The kth confusion loss The probability of being selected, according to the probability distribution From the confusion loss set Randomly select a confusion loss As the i-th worker u i The obfuscation and loss are then uploaded to untrusted platforms; rand∈[1,L];

[0060] Step 2: The untrusted platform uses a% of the total loss compensation B during the exploration phase to obtain the i-th worker u. i The quality of work; where a represents a parameter, and a∈[0,1]. Here, a=0.3 is set. If the value of a is set too large, it will lead to excessive overhead in the exploration phase, making it difficult to achieve a higher overall quality of work.

[0061] Step 2.1: Initialize j = 1; Initialize i = 1;

[0062] Step 2.2: Define the total work quality of the (j-1)th round as... and initialize

[0063] Define the loss in the (j-1)th round as and initialize

[0064] Step 2.3, the i-th worker u i Execute the j-th round of task t j And obtain the i-th worker u i Execute the j-th round of task t j work quality Therefore, equations (2) and (3) are used to calculate the task t in the j-th round. j Overall work quality and the jth round of tasks t j loss

[0065]

[0066]

[0067] Step 2.4: Assign i+1 to i, and j+1 to j. Determine if j>n is true. If true, proceed to step 2.5; otherwise, return to step 2.3.

[0068] Step 2.5: Define the random probability p of the j-th round. j When p j When p < τ, execute steps 2.6 and 2.9. j When the threshold is greater than or equal to τ, proceed with steps 2.7-2.9; where τ represents the threshold value; and τ∈[0,1], and here τ=0.2 is set.

[0069] Step 2.6: Assign j%n to i, and select the i-th worker u. i Execute the j-th round of task t j ; where % represents the remainder;

[0070] Step 2.7: Calculate the task t of the j-th round according to formula (4). j Select the i-th worker u i probability

[0071]

[0072] In equation (4), Represents the i-th worker u i Execute the j-th round of task t j The weights after; when j = n+1, let

[0073] Step 2.8: Calculate the i-th worker u according to formula (5). iExecute task t in round j+1 j+1 The weight after

[0074]

[0075] In equation (5), This represents the parameter that controls the change in weights; it is set here.

[0076] Step 2.9: Calculate using equations (2) and (3) and And judge If the condition is true, assign j+1 to j and execute step 3; otherwise, return to step 2.5 and execute sequentially.

[0077] Step 3: The untrusted platform uses the remaining loss compensation (1-a%)×B during the utilization phase to obtain the i-th worker u. i Work quality;

[0078] Step 3.1, Untrusted platforms are determined based on probability. Select the i-th worker u i To perform subsequent tasks, we use equation (6) to establish an objective function that maximizes the total work quality Q, and use equation (7) to establish the constraint equations for the objective function:

[0079]

[0080]

[0081] In equations (6) and (7), k represents the total number of tasks completed during the exploration phase. Represents the j-th round of task t j Selected workers Represents the j-th round of task t j workers Execute the j-th round of task t j The quality of work In the j-th round of task t j workers From the confusion loss set B * win A randomized confusion loss is selected from the data.

[0082] Step 3.2: Compensation for the untrusted platform completes the j-th round of task t. j workers loss The platform's compensation here will always outweigh the losses incurred by the worker due to confusion, thus ensuring the worker's personal rationality.

[0083] Step 3.3: After assigning j+1 to j, determine... Check if j≥m is true. If true, the task allocation is complete and step 4 is executed. Otherwise, return to step 3.1 and execute sequentially.

[0084] Step 4: Calculate the total work quality Q using equation (8):

[0085]

[0086] In equation (8), This represents the total work quality of the k rounds of tasks completed during the exploration phase.

[0087] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0088] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

[0089] In summary, the mobile crowd-based sensing task allocation method proposed in this invention obtains the work quality of unknown workers based on the multi-armed slot machine mechanism, protects the privacy information of workers by combining the exponential mechanism in differential privacy, and selects workers with the MAB strategy of EXP3 to maximize work quality to perform relevant tasks during the sensing task allocation process, thus ensuring the completion quality of the sensing tasks.

Claims

1. A mobile crowd sensing task allocation method based on MAB and differential privacy protection, characterized by applying... Given an untrusted platform and n workers U = {u1, u2, ..., u...} i ,…,u n In a mobile swarm intelligence sensing environment composed of}, u i Let i represent the i-th worker, i = 1, 2, ..., n. The mobile swarm intelligence sensing task allocation method is performed according to the following steps: Step 1: Build and initialize the auction model: Step 1.1: Let the untrusted platform publish m rounds of tasks T = {t1, t2, ..., t...} j ,…,t m The total loss compensation is B, where t j Let represent the j-th round of tasks; the untrusted platform sets a privacy length β∈(0,0.1) to determine the confusion loss range of workers; let the actual work loss of n workers U be b={b1,b2,…,b i ,…,b n }, where b i Represents the i-th worker u i The actual working losses; Step 1.2, any i-th worker u i After obtaining the privacy length β from the untrusted platform, the i-th worker u i The unit privacy length is determined to be Δβ; Step 1.3: Define the i-th worker u i The confusion loss set in, Represents the i-th worker u i The kth confusion loss, and Among them, s k Let represent the parameter of the k-th confusion loss, and L represents the number of confusion losses; Step 1.4: The untrusted platform uses equation (1) to construct the i-th worker u. i Obfuscation function: In equation (1), It is the kth confusion loss Mapped to the i-th worker u i Actual working losses b i The probability of; It measures the confusion loss of the kth digit. Mapped to the i-th worker u i Actual working losses b i The function of the distance between them, and △q is the global sensitivity, and Among them, b max b represents the maximum confusion loss in the confusion loss set of all workers. min Represents the minimum confusion loss in the confusion loss set of all workers; It is the i-th worker u i The y-th confusion loss; y∈[1,L]; Step 1.5, the i-th worker u i Download the obfuscation function from the untrusted platform and derive the obfuscation loss set based on the obfuscation function. probability distribution in, Represents the i-th worker u i The kth confusion loss The probability of being selected, according to the probability distribution From the confusion loss set Randomly select a confusion loss As the i-th worker u i The obfuscation and loss are then uploaded to the untrusted platform; rand∈[1,L]; Step 2: The untrusted platform uses a% of the total loss compensation B during the exploration phase to obtain the i-th worker u. i The quality of work; where a represents a parameter, and a∈[0,1]; Step 2.1: Initialize j = 1; Initialize i = 1; Step 2.2: Define the total work quality of the (j-1)th round as... and initialize Define the loss in the (j-1)th round as and initialize Step 2.3, the i-th worker u i Execute the j-th round of task t j And obtain the i-th worker u i Execute the j-th round of task t j work quality Therefore, equations (2) and (3) are used to calculate the task t in the j-th round. j Overall work quality and the jth round of tasks t j loss Step 2.4: Assign i+1 to i, and j+1 to j. Determine if j>n is true. If true, proceed to step 2.5; otherwise, return to step 2.

3. Step 2.5: Define the random probability p of the j-th round. j When p j When p < τ, execute steps 2.6 and 2.

9. j When ≥τ, execute steps 2.7-2.9; where τ represents the threshold; and τ∈[0,1]; Step 2.6: Assign j%n to i, and select the i-th worker u. i Execute the j-th round of task t j ; where % represents the remainder; Step 2.7: Calculate the task t of the j-th round according to formula (4). j Select the i-th worker u i probability In equation (4), Represents the i-th worker u i Execute the j-th round of task t j The weights after; when j = n+1, let Step 2.8: Calculate the i-th worker u according to formula (5). i Execute task t in round j+1 j+1 The weight after In equation (5), The parameter that controls the change in weight; Step 2.9: Calculate using equations (2) and (3) and And judge If the condition is true, assign j+1 to j and execute step 3; otherwise, return to step 2.5 and execute sequentially. Step 3: The untrusted platform uses the remaining loss compensation (1-a%)×B during the utilization phase to obtain the i-th worker u. i Work quality; Step 3.1, the untrusted platform is based on probability. Select the i-th worker u i To perform subsequent tasks, and use equation (6) to establish an objective function aimed at maximizing the total work quality Q, and use equation (7) to establish the constraint equations of the objective function: In equations (6) and (7), k represents the total number of tasks completed during the exploration phase. Represents the j-th round of task t j Selected workers Represents the j-th round of task t j workers Execute the j-th round of task t j The quality of work In the j-th round of task t j workers From the confusion loss set B * win A randomized confusion loss is selected from the data. Step 3.2: The untrusted platform compensation completes the j-th round of task t. j workers loss Step 3.3: After assigning j+1 to j, determine... Check if j≥m is true. If true, the task allocation is complete and step 4 is executed. Otherwise, return to step 3.1 and execute sequentially. Step 4: Calculate the total work quality Q using equation (8): In equation (8), This represents the total work quality of the k rounds of tasks completed during the exploration phase.

2. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the mobile crowd sensing task allocation method based on MAB and differential privacy protection as described in claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by the processor, performs the steps of the mobile swarm intelligence sensing task allocation method based on MAB and differential privacy protection as described in claim 1.

Images