<![CDATA[Greenhouse microclimate conditions can change rapidly, making reactive control less effective in maintaining stable environmental conditions. Short-term forecasting can help anticipate these changes and support timely control decisions. This study presents a secondary analysis of previously reported experimental results from an Internet of Things (IoT)-based melon greenhouse to examine the relationship between forecasting accuracy, forecasting horizon, and operational control performance. The analysis compares persistence, standard Long Short-Term Memory (LSTM), and modified LSTM forecasting approaches and examines environmental and resource-use indicators under predictive control. At a 10-minute forecasting horizon, the modified LSTM achieved a temperature root mean square error (RMSE) of 0.50 °C and a relative humidity RMSE of 2.07%, representing reductions of 21.88% and 20.69%, respectively, compared with the standard LSTM. Temperature RMSE increased to 1.12 °C at a 30-minute horizon, indicating reduced prediction accuracy over the longer horizon. The reported predictive-control configuration was associated with a 45.61% reduction in daytime periods exceeding the 35 °C operational threshold and a 44.74% reduction in daily fan switching events. Reductions in water, electrical energy, nutrient stock, and labour consumption were also reported. These operational outcomes reflect the overall control configuration and cannot be attributed exclusively to the forecasting model. The analysis is limited by the unavailability of raw time-series data for independent validation and statistical significance testing. The findings highlight the importance of evaluating forecasting accuracy and horizon alongside operational control requirements in greenhouse management.]]>
Copyrights © 2026
