Research On Modeling And Control Strategies Of A Microclimate System In Greenhouses

Researching on modeling and control strategies of a microclimate system in greenhouses is important to reach the aim of high-output, high-quality, high-benefit at harvet, and becomes one of the research hotspots in the field of greenhouses. A greenhouse microclmate system is a complex system with characteristics of long delay, strong coupling, gradually time-varying and nonlinear. Therefore, it is difficult to establish an accurate mechanism model and carry out effective control. To solve above problems, in this paper, we use the Extreme Learning Machine (ELM) to estalish a prediction model, and put forward control strategies of the conventional Proportion Integration Differentiation (PID) combining with ELM.The main research contributions in this work are shown as follows:(1)In order to establish an accurate mechanism model on microclimate in a greenhouse, the ELM has been employed to estalish the prediction model, with the discussion of the influence on the performance of the model by using different numbers of hidden nodes and different activation functions. According to the experiment results, compared with the Back Propagation (BP), Elman, and Support Vector Regression (SVR) approaches, the ELM has faster training speed and better simulation precision.(2)In order to improve instability of ELM performance, we use kernel-based ELM (KELM) to refine the prediction model. The learning parameters of KELM were optimized by using the Genetic Algorithm (GA). Experiment results show that, compared with the grid algorithm, the GA has less optimization time and better performance; compared with other methods, the KELM has the least training time, highest decision factors, and the most stable performance.(3)In order to achieve satisfactory results by using conventional PID control strategies in a greenhouse, a control strategy using conventional PID strategies combining with online sequential ELM (OSELM) has been proposed. The control strategy takes advantage of OSELM’s self-learning ability to adjust the parameters of the PID control factors. Experiment results have depicted that, compared with the conventional PIDs and the Radial Basis Function (RBF) neural network PIDs, the proposed control strategy has better traceability, immunity and robustness.