Skip to content
Conference Open access

Comparative Analysis of GRU, LSTM, and CNN 1D for Microclimate Prediction in IoT-Enabled Smart Greenhouse

Sep 2026 · 2026 Ninth International Conference on Vocational Education and Electrical Engineering (ICVEE) · pp. 53-58 · 0 citations · 23 references

Abstract

Smart greenhouses require reliable climate forecasts to optimize growing conditions and reduce resource waste. While Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) are widely used for time-series prediction, their comparative effectiveness on greenhouse microclimate data from arid regions remains underexplored. This paper presents a systematic comparison of GRU, LSTM, and a 1D Convolutional Neural Network (CNN 1D) using a public dataset from an IoT-equipped greenhouse in Tikrit, Iraq. All models were evaluated across three forecasting horizons (1, 3, and 6 hours) and three architectural configurations (simple, medium, deep). For 1-hour predictions, GRU achieved the best performance with RMSE of 0.94 °C, MAE of 0.38 °C, and R2 of 0.981, slightly outperforming LSTM. Wilcoxon signed-rank tests confirmed statistical significance (p < 0.001) for medium and deep configurations. GRU also requires 24% fewer parameters than LSTM. However, all models failed to capture long-term dependencies for 3-hour and 6-hour predictions, exhibiting RMSE above 8 °C and negative R2 values, highlighting the limitations of the current approach. CNN 1D offered the fastest training but lowest accuracy. These findings indicate that GRU’s simpler architecture is effective for short-term greenhouse microclimate prediction, making it suitable for resource-constrained edge deployments.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.