A knowledge model-aided evolutionary algorithm for dynamic multi-objective task allocation in heterogeneous crowdsensing

Task allocation is a critical issue in mobile crowdsensing. Due to the limited regional accessibility of human users, the spatiotemporal coverage of sensing data is often insufficient. To address this, unmanned aerial vehicles (UAVs) are introduced as complementary sensing units, forming a heterogeneous crowdsensing system. We analyze the characteristics of both UAVs and human users, and propose corresponding sensing capability metrics and task reward mechanisms. A dynamic multi-objective task allocation model is formulated to jointly maximize sensing quality and minimize residual costs. To solve this problem, a knowledge model-aided multi-objective evolutionary algorithm based on decomposition is proposed. The multi-objective task allocation problem is decomposed into scalar subproblems, from which domain-specific knowledge is extracted to construct a knowledge model that learns the probability distribution of user-task assignments. This model is then used to design a knowledge-guided change-response mechanism and evolutionary operators, aiming to enhance population convergence speed and performance. Extensive experiments on 25 static and 24 dynamic instances demonstrate the superiority of the proposed algorithm.