Energy demand forecasting of the greenhouses using nonlinear models based on model optimized prediction method

Energy demand forecasting is able to improve the energy efficiency and energy savings of the agricultural greenhouses. A model optimized prediction (MOP) methodology is proposed to predict the energy demand of greenhouses with a better performance of accuracy and cost time. The physical model of greenhouses energy demand is built up based on the energy and mass balance. According to the sensitivity analysis of the Sobol׳ method, the uncertain parameters of greenhouse energy model are sort by the first-order and total order indices. The uncertain parameters greatly affecting the model prediction can be collected from indistinct internal parameters for calibration to save computation time. Adaptive particle swarm optimization and genetic algorithms (APSO-GA) is utilized to calibrate the uncertain parameters of energy model by using the measured data in an experimental greenhouse with surface water source heat pumps system. To speed up the convergence, adaptive operator adjusts the proportion of particles for PSO and GA and changes the weight of the adjust factor during the optimization process. Compared with GA, PSO and conventional PSO-GA, APSO-GA can improve the optimization performance with more accurate of 3.2% and save the optimization time of more than 15.4%. Predicted energy demand by the optimized model is in agreement with measured energy demand with a better accuracy of a 95.6% significant level, which proves that the MOP methodology is reliable to predict energy demand and peak load of greenhouses.