Multi-objective mobile crowd sensing incentive mechanism method based on reverse auction

Abstract

The application discloses a multi-target mobile crowd sensing incentive mechanism method based on reverse auction, comprising the following steps: a multi-target crowd sensing system is mathematically modeled, and a social welfare mathematical model containing platform utility and user utility is established based on reverse auction; a social welfare maximization optimization target is constructed according to the constraint condition of limited budget; the geographical position and submitted data quality of a user at each time slot are acquired; user features are extracted and clustered; a dynamic incentive mechanism algorithm is designed based on deep reinforcement learning, and a suitable user is selected to participate in a task and paid corresponding remuneration. The application simultaneously optimizes task area coverage, platform utility and user utility under the condition of limited budget, and realizes social welfare maximization; and can dynamically select a user and provide suitable remuneration according to the change of user moving track and data quality with time, thereby improving the adaptability and overall efficiency of the crowd sensing system.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and more specifically to a method for a multi-objective mobile swarm intelligence sensing incentive mechanism based on reverse auctions. Background Technology

[0002] In recent years, with the rapid development of IoT technology, various mobile devices have embedded multiple sensors such as microphones, GPS, cameras, and gyroscopes. People can use these devices to perceive data about their surroundings in daily life, giving rise to crowdsourcing sensing as a new data sensing paradigm. Crowdsourcing sensing primarily outsources sensing tasks to a large number of unspecified individual users, fully utilizing their inherent mobility and the diverse sensing capabilities of mobile devices to achieve large-scale data collection at a relatively low cost. Due to its advantages of low cost and high flexibility, crowdsourcing sensing has been widely applied in many fields such as environmental monitoring, traffic data collection, and healthcare.

[0003] While crowdsourcing sensing technology can solve many problems, it currently faces limitations such as insufficient user participation and unstable data quality. Therefore, some research has proposed incentive models. For example, some have proposed user-centered incentive models that compensate users for costs while primarily aiming to find the optimal user set to maximize platform utility; others consider the correlation between tasks and locations, selecting the user with the lowest cost as the winner to minimize platform costs. These approaches primarily emphasize the importance of a single objective, neglecting the optimization value of other objectives.

[0004] Furthermore, incentive mechanisms can be categorized into offline and online types based on the level of prior information the platform possesses. Offline mechanisms typically assume the platform has all user information, while online mechanisms rely solely on past user information, lacking future data. In location-based services, traditional incentive strategies often predict user movement patterns based on a large amount of user movement data and then use a greedy algorithm to select suitable users for tasks. However, this often proves ineffective in real-world scenarios, as newly arrived tasks often lack sufficient collected trajectory data to accurately predict user movement patterns. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-objective mobile crowd sensing incentive mechanism method based on reverse auction, in order to solve the shortcomings of existing crowd sensing methods that mainly emphasize the importance of a single objective while ignoring other optimization objectives, and do not consider the dynamic changes and uncertainties of users in the incentive mechanism.

[0006] The present invention adopts the following technical solution:

[0007] The multi-objective mobile crowd sensing incentive mechanism method based on reverse auction includes the following steps:

[0008] Step 1: Perform mathematical modeling on the multi-objective swarm intelligence perception system, and establish a social welfare mathematical model that includes platform utility and user utility based on reverse auction;

[0009] Step 2: Based on the constraint of limited budget, construct a social welfare maximization optimization objective;

[0010] Step 3: Obtain the user's geographical location and the quality of the submitted data in each time slot;

[0011] Step 4: Extract user features and cluster them;

[0012] Step 5: Based on deep reinforcement learning, design a dynamic incentive mechanism algorithm to select suitable users to participate in the task and pay them corresponding rewards.

[0013] In a preferred embodiment, the multi-target swarm intelligence sensing system in step 1 above consists of a platform and N users. The buyer's budget is B, and they need to collect noise data over a specified period of time in a designated area. After receiving the task, the platform first divides the entire sensing area into a set of grids, where Z represents the set of grids, and each grid j∈Z has a weight w. j This indicates the importance of the grid; the entire perception process T is divided into several stages, and at the beginning of each perception stage, the user submits a quote b to the platform for their current location and the task to be performed. i The platform selects suitable users to join the winning set W based on the given information, and pays the corresponding reward p after the winning users complete the task. i , using c i This represents the true cost for user i to perform the task; this value is private information known only to the user. User utility is defined as the difference between the payment and the true cost. The formula for calculating user i's utility is as follows:

[0014]

[0015] Define the user access matrix V t (W) indicates whether the winning user has visited a specific grid before the current time slot t, where each row represents a winning user and each column indicates whether the user has visited the grid: v ij ∈{0, 1}, that is, if grid j has been visited by user i, then v ij =1, otherwise v ij =0; until the current time t, even if user i visits grid j multiple times, v ijStill counted as 1; Platform utility is defined as the platform's revenue minus the reward paid to the winning user. The platform's revenue consists of regional coverage revenue and data quality revenue. Regional coverage is expressed as the ratio of the number of grids accessed by the winner to the total number of grid regions within the trajectory time range: The coverage revenue that the platform obtains from task area coverage is defined as: System parameters used to control the gradient of regional coverage revenue are described by a logarithmic function, illustrating the diminishing marginal returns of the platform. User reputation is defined as the average quality of historical data submitted by users; the user reputation matrix is ​​as follows: in, This represents the average data quality submitted by user i up to the current time t; the data quality benefit the platform gains from the user-submitted data is: For the quality of the data submitted by user i at time t, The system parameters for controlling the data quality benefit gradient; the platform utility is represented as:

[0016] The aforementioned social welfare is the sum of platform utility and user utility: Use the quote b for the task execution i Instead of the true cost c that cannot be accurately obtained i Therefore, the mathematical model for the social welfare expression is:

[0017] The formula for constructing the social welfare maximization optimization objective in step 2 above is as follows:

[0018]

[0019] Step 4 above involves using the Adaptive Resonance Method (ARM) to extract and cluster user features. This ARM integrates the ART network and a pruning mechanism. The ART network consists of an input layer, an output layer, and fully connected bottom-up and top-down links. The input layer comprises a user access matrix and a user reputation matrix. Each node in the output layer represents a category CL, and each participant is assigned to a corresponding node. The input and output layers have two fully directional connections: the bottom-up network weights are represented by W. b This indicates that the top-down network weights are represented by W. t express.

[0020] In a preferred embodiment, the specific process of step 4 above is as follows:

[0021] Step 4.1: Concatenate the user access matrix and the user reputation matrix to form a |W|×(|Z|+1) dimensional feature matrix C. t (W);

[0022] Step 4.2: Convert the feature matrix Ct (W) Each row serves as the user's feature input to the adaptive resonant network, and the matching degree between each node and the input vector is calculated: net[j, p] represents the degree of matching between the j-th CL and user p;

[0023] Step 4.3: Select node j with the highest matching degree. * ;

[0024] Step 4.4: Calculate node j * Similarity to user characteristics:

[0025] Step 4.5: Compare the similarity with the set threshold. If it is greater than the threshold, update the network weights; otherwise, traverse all other output nodes to find nodes with similarity greater than the threshold. If none of the nodes in the output layer meet the requirements, add a new classification node and assign the user to that category. If the total number of output layer nodes exceeds n, start the pruning mechanism.

[0026] In a preferred embodiment, the dynamic incentive mechanism algorithm in step 5 above includes:

[0027] Step 5.1: Initialize network Q with random network parameters ω ω (s, a);

[0028] Step 5.2: Copy the same parameter ω - ←ω Initialize the target network

[0029] Step 5.3: Initialize the experience replay pool;

[0030] Step 5.4: Generate experience data and store it in the playback pool;

[0031] Step 5.5: Randomly sample from the replay pool for training.

[0032] Define the following parameters: Obtain n user state clusters CL from step 4, with the following states: Indicates the i-th cluster; Action: a t Select a state cluster from the user state cluster CL; Reward: r t (s t a t ) is in state s t Take action a t The formula for calculating the increase in social welfare received minus the penalties for violating budget constraints is as follows: in, This is an adjustable parameter; Reward: The reward a user receives at time t is the bid b corresponding to the user selected at time t-1. iStep 5.4 above specifically includes the following:

[0033] Step 5.4.1: Use ARM to obtain the current environment state s t ;

[0034] Step 5.4.2: Based on the current network Q ω (s, a) Select action a using the ε-greedy strategy. t

[0035]

[0036] Step 5.4.3: Use the ε-greedy strategy to select the user with the highest current reward in the cluster;

[0037] Step 5.4.4: Perform action a t The environmental state becomes s t+1 Receive reward r t ;

[0038] Step 5.4.5: (s) t a t r t s t+1 Store in the playback pool.

[0039] Preferably, step 5.5 above specifically includes the following:

[0040] Step 5.5.1: Randomly sample data {(s) from the playback pool k a k r k s k )} k＝1，...，K ;

[0041] Step 5.5.2: For each sampled data point, calculate using the target network.

[0042] Step 5.5.3: Update the current network Q with the objective of minimizing the loss function. ω The loss function is:

[0043] Step 5.5.4: Update the target network

[0044] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following advantages:

[0045] 1. This invention optimizes task area coverage, platform utility, and user utility simultaneously under limited budget conditions to maximize social welfare; it can also dynamically select users and provide appropriate rewards based on changes in user movement trajectories and data quality over time; and it improves the learning efficiency of deep reinforcement learning algorithms by performing cluster analysis on user location features and data quality features before using deep reinforcement learning techniques.

[0046] 2. This invention extracts features from the state of reinforcement learning and uses the similarity between users' movement patterns and reputations to dynamically cluster users, which can reduce the search space of the algorithm, improve learning efficiency and reduce computational costs. Attached Figure Description

[0047] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0048] Specific embodiments of the present invention will now be described with reference to the accompanying drawings. Many details are described below to provide a comprehensive understanding of the invention; however, those skilled in the art will not need these details to implement the invention. Well-known components, methods, and processes will not be described in detail below.

[0049] Reference Figure 1 This embodiment provides a multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction, including the following steps:

[0050] Step 1: Perform mathematical modeling on the multi-objective swarm intelligence perception system, and establish a social welfare mathematical model that includes platform utility and user utility based on reverse auction;

[0051] Step 2: Based on the constraint of limited budget, construct a social welfare maximization optimization objective;

[0052] Step 3: Obtain the user's geographical location and the quality of the submitted data in each time slot;

[0053] Step 4: Extract user features and cluster them;

[0054] Step 5: Based on deep reinforcement learning, design a dynamic incentive mechanism algorithm to select suitable users to participate in the task and pay them corresponding rewards.

[0055] The multi-target swarm intelligence sensing system of this invention consists of a platform and N users. Assuming a buyer's budget is B, and they need to collect noise data over a specified period in a designated area; after receiving this task, the platform first divides the entire sensing area into a set of grids, Z representing the grid set, where each grid j∈Z has a weight w. j The importance of the grid is indicated by , and the entire perception process T is divided into several stages.

[0056] When establishing a mathematical model for the perception data collection process based on reverse auctions, the requester is the buyer, the user is the seller, and the platform is the auctioneer. At the beginning of each perception phase, the user submits a bid (b) to the platform, specifying their current location and the task they wish to perform. i The platform selects suitable users to join the winning set W based on the given information, and pays the corresponding reward p after the winning users complete the task. i , using c i This represents the actual cost of user i performing the task; this value is private information known only to the user.

[0057] User utility is defined as the difference between the payment made and its actual cost. The formula for calculating the utility of user i is as follows:

[0058]

[0059] The winning user's geographic location and the quality of the submitted data in each time slot:

[0060] Define the user access matrix V t (W) indicates whether the winning user has visited a specific grid before the current time slot t, where each row represents a winning user and each column indicates whether the user has visited the grid: v ij ∈{0, 1}, that is, if grid j has been visited by user i, then v ij =1, otherwise v ij =0; until the current time t, even if user i visits grid j multiple times, v ij Still counted as 1; Platform utility is defined as the platform's revenue minus the reward paid to the winning user. The platform's revenue consists of regional coverage revenue and data quality revenue. Regional coverage is expressed as the ratio of the number of grids accessed by the winner to the total number of grid regions within the trajectory time range: The coverage revenue that the platform obtains from task area coverage is defined as: The system parameters used to control the gradient of regional coverage revenue are described by a logarithmic function, which describes the law of diminishing marginal returns of the platform.

[0061] User reputation is defined as the average quality of historical data submitted by a user. The user reputation matrix is ​​as follows: in, This represents the average data quality submitted by user i up to the current time t; the data quality benefit the platform gains from the user-submitted data is: For the quality of the data submitted by user i at time t, The system parameters for controlling the data quality benefit gradient; the platform utility is represented as:

[0062] Social welfare is the sum of platform utility and user utility. Because the platform cannot accurately obtain the user's true cost. i Therefore, the quotation b for the task execution is used. i Instead of the actual cost c i The mathematical model for the social welfare expression is changed to:

[0063]

[0064] The platform needs to maximize social welfare within a limited budget, using the following formula:

[0065]

[0066] Step 4 above involves using the Adaptive Resonance Method (ARM) to extract and cluster user features. The ARM method, as an unsupervised learning technique, integrates the ART network and a pruning mechanism. The ART network consists of an input layer, an output layer, and fully connected bottom-up and top-down links. The input layer comprises a user access matrix and a user reputation matrix. Each node in the output layer represents a classification CL, and each participant is assigned to a corresponding node. The input and output layers have two fully directional connections: the bottom-up network weights are represented by W. b This indicates that the top-down network weights are represented by W. t express.

[0067] The specific process of the adaptive resonance method ARM is as follows:

[0068] Step 4.1: Concatenate the user access matrix and the user reputation matrix to form a |W|×(|Z|+1) dimensional feature matrix C. t (W);

[0069] Step 4.2: Convert the feature matrix C t (W) Each row serves as the user's feature input to the adaptive resonant network, and the matching degree between each node and the input vector is calculated: net[j, p] represents the degree of matching between the j-th CL and user p;

[0070] Step 4.3: Select node j with the highest matching degree. * ;

[0071] Step 4.4: Calculate node j * Similarity to user characteristics:

[0072] Step 4.5: Compare the similarity with the set threshold. If it is greater than the threshold, update the network weights; otherwise, traverse all other output nodes to find nodes with similarity greater than the threshold. If none of the nodes in the output layer meet the requirements, add a new classification node and assign the user to that category. If the total number of output layer nodes exceeds n, start the pruning mechanism.

[0073] Define the following parameters: Obtain n user state clusters CL from step 4, with the following states: Indicates the i-th cluster; Action: a t Select a state cluster from the user state cluster CL; Reward: r t (s t a t ) is in state s t Take action a t The formula for calculating the increase in social welfare received minus the penalties for violating budget constraints is as follows: in, This is an adjustable parameter; Reward: The reward a user receives at time t is the bid b corresponding to the user selected at time t-1. i .

[0074] The dynamic incentive mechanism algorithm in step 5 above includes:

[0075] Step 5.1: Initialize network Q with random network parameters ω ω (s, a);

[0076] Step 5.2: Copy the same parameter ω - ←ω Initialize the target network

[0077] Step 5.3: Initialize the experience replay pool;

[0078] Step 5.4: Generate experience data and store it in the playback pool;

[0079] Step 5.5: Randomly sample from the replay pool for training.

[0080] Step 5.4 specifically includes the following:

[0081] Step 5.4.1: Use ARM to obtain the current environment state s t ;

[0082] Step 5.4.2: Based on the current network Q ω (s, a) Select action a using the ε-greedy strategy. t :

[0083]

[0084] Step 5.4.3: Use the ε-greedy strategy to select the user with the highest current reward in the cluster;

[0085] Step 5.4.4: Perform action a t The environmental state becomes s t+1 Receive reward r t ;

[0086] Step 5.4.5: (s) t a t r t s t+1 Store in the playback pool.

[0087] Step 5.5 specifically includes the following:

[0088] Step 5.5.1: Randomly sample data {(s) from the playback pool k a k r k s k )} k＝1，...，K ;

[0089] Step 5.5.2: For each sampled data point, calculate using the target network.

[0090] Step 5.5.3: Update the current network Q with the objective of minimizing the loss function. ω The loss function is:

[0091]

[0092] Step 5.5.4: Update the target network

[0093] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction, characterized in that, Includes the following steps: Step 1: Perform mathematical modeling on the multi-objective swarm intelligence perception system, and establish a social welfare mathematical model that includes platform utility and user utility based on reverse auction; Step 2: Based on the constraint of limited budget, construct a social welfare maximization optimization objective; Step 3: Obtain the user's geographical location and the quality of the submitted data in each time slot; Step 4: User features are extracted and clustered using the Adaptive Resonance Method (ARM). This ARM method integrates the ART network and a pruning mechanism. The ART network consists of an input layer, an output layer, and fully connected links from bottom to top and top to bottom. The input layer consists of a user access matrix and a user reputation matrix. Each node in the output layer represents a classification cluster (CL), and each participant is assigned to a corresponding node. The input and output layers have two fully directional connections: the bottom-up network weights are represented by W. b This indicates that the top-down network weights are represented by W. t The specific process is as follows: Step 4.1: Concatenate the user access matrix and the user reputation matrix to form a |W|×(|Z|+1) dimensional feature matrix C. t (W), where W represents the winning set and Z represents the grid set; Step 4.2: Convert the feature matrix C t (W) Each row serves as the user's feature input to the adaptive resonant network, and the matching degree between each node and the input vector is calculated: net[h, i] represents the degree of matching between the h-th cluster CL and user i; Step 4.3: Select the node h with the highest matching degree. * ; Step 4.4: Calculate node h * Similarity to user characteristics: Step 4.5: Compare the similarity with the set threshold. If it is greater than the threshold, update the network weights; otherwise, traverse all other output nodes to find nodes with similarity greater than the threshold. If none of the nodes in the output layer meet the requirements, add a new classification node and assign the user to that category. If the total number of output layer nodes exceeds m, start the pruning mechanism. Step 5: Based on deep reinforcement learning, design a dynamic incentive mechanism algorithm to select suitable users to participate in the task and pay them corresponding rewards; the dynamic incentive mechanism algorithm includes: Step 5.1: Initialize network Q with random network parameters ω ω (s,a), where s represents the environment state and a represents the action; Step 5.2: Copy the same parameters as ω to ω - and initialize the target network. Step 5.3: Initialize the experience replay pool; Step 5.4: Generate experience data and store it in the playback pool; Step 5.5: Randomly sample from the replay pool for training.

2. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 1, characterized in that: The multi-target swarm intelligence sensing system in step 1 consists of a platform and n users. The buyer's budget is B, and they need to collect noise data in a designated area over a period of time. After receiving the task, the platform first divides the entire sensing area into a set of grids, where Z represents the set of grids, and each grid j∈Z has a weight w. j This indicates the importance of the grid. The entire perception process T is divided into several stages. At the beginning of each perception stage, the user submits their current location and a quote b for the task to be performed to the platform. i The platform selects suitable users to join the winning set W based on the given information, and pays the corresponding reward p after the winning users complete the task. i , use c i This represents the true cost for user i to perform the task; this value is private information known only to the user. User utility is defined as the difference between the payment and the true cost. The formula for calculating user i's utility is as follows:

3. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 2, characterized in that: Define the user access matrix V t (W) indicates whether the winning user has visited a specific grid before the current time slot t, where each row represents a winning user and each column indicates whether the user has visited the grid: v ij ∈{0,1}, meaning if grid j has been visited by user i, then v ij =1, otherwise v ij =0; until the current time t, even if user i visits grid j multiple times, v ij Still counted as 1; Platform utility is defined as the platform's revenue minus the reward paid to the winning user. The platform's revenue consists of regional coverage revenue and data quality revenue. Regional coverage is expressed as the ratio of the number of grids accessed by the winner to the total number of grid regions within the trajectory time range: The coverage revenue that the platform obtains from task area coverage is defined as: System parameters used to control the gradient of regional coverage revenue are described by a logarithmic function, illustrating the diminishing marginal returns of the platform. User reputation is defined as the average quality of historical data submitted by users; the user reputation matrix is ​​as follows: in, This represents the average data quality submitted by user i up to the current time t; the data quality benefit the platform gains from the user-submitted data is: For the quality of the data submitted by user i at time t, The system parameters for controlling the data quality benefit gradient; the platform utility is represented as:

4. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 3, characterized in that, Social welfare is the sum of platform utility and user utility. Use the quote b for the task execution i Instead of the true cost c that cannot be accurately obtained i Therefore, the mathematical model for the social welfare expression is:

5. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 4, characterized in that: The formula for constructing the social welfare maximization optimization objective in step 2 is as follows:

6. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 1, characterized in that, Define the following parameters: Obtain H user state clusters CL from step 4, with the following states: Indicates the h-th cluster; Action: a t Select a state cluster from the user state cluster CL; Reward: r t (s t ,a t ) is in state s t Take action a t The formula for calculating the increase in social welfare received minus the penalties for violating budget constraints is as follows: in, This is an adjustable parameter; Reward: The reward a user receives at time t is the bid b corresponding to the user selected at time t-1. i .

7. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 1, characterized in that, Step 5.4 specifically includes the following: Step 5.4.1: Use ARM to obtain the current environment state s t ; Step 5.4.2: Based on the current network Q ω (s,a) selects action a using the ε-greedy strategy. t : Step 5.4.3: Use the ε-greedy strategy to select the user with the highest current reward in the cluster; Step 5.4.4: Perform action a t The environmental state becomes s t+1 Receive reward r t ; Step 5.4.5: (s) t ,a t ,r t ,s t+1 Store in the playback pool.

8. The multi-objective mobile swarm intelligence sensing incentive mechanism method based on reverse auction as described in claim 1, characterized in that, Step 5.5 specifically includes the following: Step 5.5.1: Randomly sample data {(s) from the playback pool k ,a k ,r k ,s k )} k＝1,...,K ; Step 5.5.2: For each sampled data point, calculate using the target network. Step 5.5.3: Update the current network Q with the objective of minimizing the loss function. ω The loss function is: Step 5.5.4: Update the target network

Images