Greenhouse Temperature Prediction Based on Time-Series Features and LightGBM

Abstract

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