Multistep ahead prediction of temperature and humidity in solar greenhouse based on FAM-LSTM model

Abstract

Solar greenhouses offer favorable climatic environments for the production of off-season crops in northern China. Greenhouse temperature and humidity are critical environmental factors influencing crop growth; therefore, it is crucial to predict the temperature and humidity. In this research, we constructed a feed-forward attention mechanism- long-short term memory (FAM-LSTM) model for the multistep prediction of temperature and humidity in a solar greenhouse. The FAM-LSTM model considered both internal and external environmental factors influencing crop growth, where the factors include temperature, humidity, light, soil temperature, and soil moisture. Then, we performed corresponding experiments employing this model in predicting both temperature and humidity at prediction horizons of 12, 24, 36, and 48 h. The FAM-LSTM model stands out due to its high accuracy compared with different models. The temperature prediction error of the model at four prediction horizons is primarily within [−0.5, 0.9], [−2.1, 1.7], [−0.9, 2.7], and [−1.8, 2.9] ℃, respectively. The humidity prediction errors of the model at four prediction horizons are mainly within [−1.9, 1.8], [−4.4, 3.6], [−6.9, 2.4], and [−5.8, 4.4] %, respectively. This study offers an effective temperature and humidity prediction approach for precise environmental regulation and damage warnings in solar greenhouses.