| Crowd Sensing,as an important component of Internet of Things(IoT),has been widely concerned by scholars.With the rapid development of Internet of Vehicles(IoV)technology,more and more vehicles have participated in the crowd sensing tasks.An excellent vehicle participant recruitment strategy is the urgent requirement of the management platforms to maximize system crowd sensing profits while saving costs.In existing research on participant recruitment,recruitment strategies are mainly based on the location information of participants or the rough estimation of the movement of participants between regions.However,in the scenarios of Internet of Vehicles,the mobility of vehicles is highly random and dynamic,so it is difficult for traditional strategies to infer the potential recruitment value of participants immediately and effectively.In addition,the distribution of perception tasks in the Internet of vehicles usually follows the spatial structure of the traffic network,which has high accuracy requirements for vehicle mobility trajectory.Therefore,in order to improve the final benefit of the crowd sensing system,vehicle predicted trajectories are taken into consideration in the formulation of recruitment scheme.In recent years,the application of deep learning has made remarkable achievements in the field of vehicle trajectory data mining.In this thesis,a neural network model is built to extract the feature information contained in the vehicle trajectory data in both temporal and spatial aspects,so as to realize the mobility prediction of the vehicle at a specific moment in the future.Due to the complexity of trajectory data,the existing mobility prediction models either have the problem of gradient vanishing or lose part of the input trajectory information,so it is difficult to meet the needs of vehicle participant recruitment.In this thesis,the trajectory features are divided into local features within subsegments and global features between subsegments by segmented processing,and the hierarchical feature extraction is carried out to reduce the computational complexity of the model.On the other hand,an attention mechanism is introduced into the model to enable it to automatically recognize the importance of each subsegment of the trajectory,thus further improving the accuracy of the mobility prediction results.Then,the performance superiority and universality of the proposed model is verified by comparison experiments on real traffic trajectory data sets.After the predicted trajectories are calculated based on the mobility prediction model,this thesis continues to design the vehicle participant recruitment strategy for the Internet of Vehicles scenarios.The basis for the formulation of vehicle participant recruitment mainly includes participant recruitment cost,vehicle predicted trajectory and crowd sensing task information.In order to meet the requirements of real-time performance and accuracy of the system,this thesis formulates the recruitment strategy based on heuristic algorithm,and proposes an adaptive genetic algorithm combined with the idea of simulated annealing.Based on the trajectory data set and the crowd sensing task simulation data,the participant recruitment experiment is carried out in this thesis.Finally,it is verified that the algorithm proposed in this thesis can obtain high-quality recruitment schemes,and has good application value and development prospect. |
