Application of the evolutionary algorithms for task allocation in uncertain environments with stochastic tuning

In the process of the Unmanned Aerial Vehicle (UAV) swarms conducting search-attack missions, target recognition is sometimes inaccurate, due to observation errors. This may lead to misjudgment of the targets' types, resulting in a decrease in the swarm’s efficiency. In response to this problem, the target recognition error is first modelled as a recognition matrix P. Secondly, a tuning matrix Q is introduced to reduce detrimental effects induced by recognition errors. In target recognition uncertain environments, we develop a task allocation model with the tuning matrix Q as the decision variable with an aim at maximizing the reward of task assignment. Afterwards, a problem-specific evolutionary algorithm termed the Stochastic-Tuning-based Evolutionary Algorithm (ST-EA) is designed to optimize the above formulated constrained optimization model. Numerical experiments are performed on six test instances with different P values and different numbers of UAVs. Experimental results demonstrate the efficiency of the proposed ST-EA and the superiority when it is compared with ST-EA-NoQ, a variant of ST-EA without using the tuning matrix Q. Furthermore, the stability and convergence of the proposed algorithm is further demonstrated. Finally, by comparing with the Stochastic-Tuning-based Differential Evolution Algorithm (ST-DE), we show the advantages of ST-EA.