Greenhouse Temperature Prediction Based on Time-Series Features and LightGBM

A method of establishing a prediction model the greenhouse temperature based on time-series analysis and boosting tree is proposed, aiming at problem that cannot be accurately predicted owing to nonlinear changes in closed ecosystem featuring modern agricultural technology various influencing factors. This comprehensively considers environmental parameters, including humidity inside outside greenhouse, air pressure as well changes, make more accurate greenhouse. Experiments show R2 determination coefficients different models are improved mean square error absolute reduced after adding features. Among tested, LightGBM performs best, with results decreasing by 18.61% Comparing support vector machine, radial basis function neural network, back-propagation multiple linear regression features, 11.70% 29.12% lower. Furthermore, fitting degree best among models. The therefore have important application value control.