Research on Time Series Prediction Model for Multi-factor Environmental Parameters in Facilities Based on LSTM-AT-DP Model

Longwei Liang¹Hui Shi²Zhaoyuan Wang¹Shengjie Wang³Changhong Li¹Ming Diao^1*

• ¹College of Agriculture, Shihezi University, Shihezi, China
• ²International PhD School, University of Almería, Almería 04120, Spainc, Almería, Spain
• ³Shihezi University College of Mechanical and Electrical Engineering, Shihezi, China

Existing facility environment prediction models often suffer from low accuracy, poor timeliness, and error accumulation in long-term predictions under multi-factor nonlinear coupling conditions. To address these limitations, this paper proposes a facility environment prediction model based on long short-term memory-attention mechanism-data preprocessing (LSTM-AT-DP) based facility environment prediction model. The model employs a Long Short-Term Memory (LSTM) network as its core architecture for deep temporal modeling and incorporates a Data Preprocessing (DP) module.Within this module, Wavelet Threshold Denoising (WTD) eliminates high-frequency noise from raw data, while a Sliding Window (SW) technique restructures the denoised time-series data into a feature matrix to ensure structured input. Additionally, an Attention Mechanism (AT) dynamically assigns feature weights, enhancing the model's ability to extract critical temporal features and thereby improving prediction accuracy. In the 24-hour long-term prediction, the coefficient of determination (R2) of this model for temperature, humidity, and radiation prediction reaches 0.9602, 0.9529, and 0.9839, respectively, which is improved by 3.89%, 5.53%, and 2.84% compared with the LSTM model, and the root mean square error (RMSE) is reduced by 0.6830, 1.8759 and 12.952, respectively.The experimental results show that the model significantly improves the prediction accuracy and suppresses the error accumulation, which can provide technical support for the precise regulation of facility environment.