Improved multi-objective evolutionary algorithm for optimization control in greenhouse environment

A greenhouse environment control system plays a decisive role in greenhouse production processes and is a complex system to control. This paper provides an overview of a greenhouse control system and control technologies. We investigated the issue of a greenhouse climate control system based on temperature and humidity, and formulated a greenhouse climate dynamic model. The control strategy was presented for the dynamic model made use of conventional Proportional Integral and Derivative (PID) control algorithms in which it combined with an modified multi-objective evolutionary algorithm (MNSEA-II) based on NSGA-II. In MNSEA-II, mixed mutation strategy and local search strategy were utilized to tune two PID controller parameters, and the integrated time square error (ITSE) was considered as one of performance criteria. The mixed mutation strategy based on game theory could utilize adaptively the advantages of a different mutation operator to maintain the globe search capacity of population for a diversity of Pareto solutions, and the local search strategy could speed the convergence of algorithms to achieve more precise solutions. The mixed mutation strategy and the local search strategy could obtain an equilibrium between the diversity and precision of Pareto solutions. An evolutionary optimization process was employed to approximate the set of Pareto solutions, which was used to tune PID controller parameters to achieve good control performance. The tuning scheme has been tested for greenhouse climate control by minimizing ITSE and control increment or rate in a simulation system. Simulation results showed the effectiveness and usability of the proposed method for step responses. The obtained gains were applied in PID controllers and could achieve good control performance such as small overshoot, fast settling time, and less rise time and steady state error. The proposed optimization method offers an effective way to implement simple but robust solutions providing a good reference tracking performance in a closed loop, and the non-dominated Pareto optimal solutions have better distribution and faster convergence at the same time.