Co-evolution Based Mixed-variable Multi-objective Particle Swarm Optimization for UAV Cooperative Multi-task Allocation Problem

In recent years, the multi-aircraft cooperative control system of UAV has been widely used in many application fields, such as military strike, ocean monitoring, aerial photography on land and disaster detection. In order to describe UAV cooperative multi-task allocation scenarios more accurately, this paper proposes a UAV cooperative multi-task allocation model (M-CMTAP) that considers multiple constraints and multiple optimization objectives simultaneously, which involves mixed decision variables and multiple constraints. These constraints describe the relationships in the complex decision variables in the actual military scenarios, including UAV resource constraint, UAV type constraint, mission execution sequence constraint and multi-aircraft cooperation constraint, etc. In M-CMTAP, two optimization objectives are taken into account, i.e., the total flight range of the UAVs and the completion time for all missions. The shortest flight range of the UAVs means that the UAVs consume the least flight resources during the mission, while the minimum completion time for all missions ensures that the entire military mission can be completed quickly. In order to solve this M-CMTAP model more efficiently, this paper proposes a co-evolution based mixed-variable multi-objective particle swarm optimization algorithm named C-MOPSO. C-MOPSO firstly adopts the mixed variable coding method to represent the task allocation, where some discrete variables represent the allocation relationship between the tasks and the UAVs, while some continuous variables and some other discrete variables represent the resource consumption of the UAVs during the mission. In order to generate feasible particles satisfying various constraints, a feasible solution initialization method with constraint processing is further proposed. And a structured learning based reproduction method is then employed to update the particles to improve the diversity and convergence of the population. The structure learning based reproduction method learns the historical structural information of good solutions and the corresponding structure length can help keep good convergence and diversity of the population. After getting the position vector of the good solutions, it sequentially adds new task numbers, UAV numbers and resource consumption values satisfying the constraints, until all tasks have been allocated to the UAVs and completed. At the same time, in order to further improve the search efficiency of the algorithm, this paper introduces the idea of co-evolution to design the search strategy, and different populations exchange search information through the way of co-evolution. In C-MOPSO, the co-population co-evolves with the particle swarm, and the fast non-dominant sorting method is also adopted to select the first N dominant particles from the fusion population which is composed by the co-population and the particle swarm to update the Pbest. Due to the co-evolution based population search strategy, it can balance the diversity and convergence of the population well, which can help to generate better particles, so as to obtain better solutions. In order to verify the effectiveness of C-MOPSO, four representative M-CMTAP test cases are designed based on different UAV distributions and mission distributions. The experimental results on the four representative test problems show that, compared with some state-of-the-art co-evolution based algorithms, the proposed algorithm C-MOPSO outperforms others on convergence efficiency and diversity of solution sets. © 2021, Science Press. All right reserved.