Optimization of above-ground environmental factors in greenhouses using a multi-objective adaptive annealing genetic algorithm

Abstract

The greenhouse environment represents a dynamic, nonlinear system characterized by hysteresis and is influenced by a myriad of interacting environmental parameters, posing a complex multi-variable optimization challenge. This study proposes a multi-objective adaptive annealing genetic algorithm to optimize above-ground environmental factors in greenhouses, addressing the challenges of variable environmental conditions and extensive heating and humidity infrastructure. Initially, after analyzing the multi-objective model of greenhouse above-ground environmental factors, including temperature, relative humidity, and CO2 concentration, a comprehensive multi-objective, multi-constraint model was developed to encapsulate these factors in greenhouse environments. Subsequently, the model optimization incorporated multi-parameter coding of decision variables, a fitness function, and an annealing dynamic penalty factor. Validation conducted at Yangling Agricultural Demonstration Park revealed that the application of multi-objective adaptive annealing genetic algorithms (schemes 1 and 2) significantly outperformed the single-objective genetic algorithm (scheme 3) and the traditional genetic algorithm (scheme 4). Specifically, the improvements included a reduction in average temperature rise by 2.64 °C and 5.29 °C for schemes 1 and 2, respectively, equating to 20 % and 34 % decreases. Additionally, average humidification reductions of 2.39 % and 3.9 % were observed, alongside decreases in the total lengths of heating and humidification pipes by up to 2.99 km and 0.443 km, respectively, with a maximum reduction of 14 % in heating pipes. The integration of an annealing dynamic penalty factor enhanced the adaptive climbing ability of schemes 1 and 2, improving static stability and robustness. Furthermore, the number of iterations required to achieve convergence was reduced by approximately 170–240 times compared to schemes 3 and 4. This reduction in iterations also resulted in a significant decrease in running time by 5–13 min, corresponding to time savings of 31 %–56 %, thereby achieving further optimization.