移动群智感知中基于用户意愿的多任务分配模型

Multi-Task Allocation Based on User Willingness in Mobile Crowd Sensing

以群智感知平台为中心的任务分配方法中,平台收集用户实时位置、社交关系等信息并进行集中处理,将多个任务分配给用户,以实现感知时间最小化、感知成本最小化等优化目标。但传统方法没有着重考虑用户的执行意愿,与用户意愿相悖的分配方案和大量隐私信息的泄露不利于维护用户的长时参与水平。因此提出基于用户意愿的任务路径分配模型PtrNet-TA,在用户移动端通过本地历史感知数据训练一个seq2seq的指针网络模型,任务分配时将云端服务器提供的待执行感知任务集合作为输入传入本地指针网络模型从而输出符合用户意愿的任务路径分配给对应的用户。通过仿真实验表明,PtrNet-TA序列预测准确度优于分别采用LSTM模型与LSTM + Attention模型的基准方法。

In the platform-centric task allocation method, the MCS platform collects information such as real-time location, historical movement trajectory, and social relationship of the user for centralized processing, and assigns multiple tasks to users, so as to optimize targets such as minimizing perceived time and minimizing perceived cost. The conventional method does not focus on the user’s willingness to execute. The distribution scheme which is contrary to the user’s will and the disclosure of a large amount of private information are not conducive to maintaining the long-term participation level of users. Therefore, a task path allocation method PtrNet-TA based on the user’s intention is proposed, and a seq2seq pointer network model is trained on the user mobile terminal through local history sensing data, and the to-be-perceived task set provided by the cloud server is input as an input during task assignment. The local pointer network model is passed in to output a task path that matches the user’s wishes to the corresponding user. Simulation experiments show that PtrNet-TA is better than the benchmark method using LSTM model and LSTM + Attention model.