Graph-Based Temperature and Humidity Prediction Model for Mushroom House Using Spatial–Temporal Fusion Strategy

Le résumé fourni par la source

Multivariate environmental prediction is essential for precisely regulating mushroom house Internet of Things system. Existing time-series prediction methods, such as long short-term memory (LSTM) and temporal convolutional network (TCN), consider the temporal features of multiple variables. However, the potential spatial relations between multiple variables cannot be effectively exploited. Especially, sudden environmental disturbances tend to increase the model’s predictive error. To address this challenge, we proposed a multi-input-multioutput (MIMO) prediction model employing a spatial–temporal fusion approach. The model combined TCN with graph sampling and aggregation network-based dynamic graph learning strategy (TCN-DGSA). It achieves the combined prediction of temperature and the humidity in the mushroom houses. First, TCN extracts temporal features from input data, which enhances the model’s ability to capture long-term temporal dependencies through dilation convolution. Additionally, a dynamic graph learning strategy was developed to learn spatial relationships of multiple variables. This strategy constructed implicit graph structures of input features without empirical knowledge. Then, the sampling and aggregation network effectively extracted the spatial pattern of the graph structure, and achieved the accurate multivariate prediction. Finally, the single-step and multihorizon prediction performance of the model was verified by ablation experiments. The TCN-DGSA model outperforms baseline models, achieving Mean Absolute Error (MAE), RMSE, and$R^{2}$of 0.21°C, 0.30°C, and 0.97 for temperature prediction, and 0.53%, 1.02%, and 0.98 for humidity. Further, after adding Gaussian, Poisson, and uniform noise to raw dataset, the model maintained similar MAE and RMSE across different output horizons. This result demonstrates that TCN-DGSA model has high stability and robustness in complex environments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.