Research on time series prediction model for multi-factor environmental parameters in facilities based on LSTM-AT-DP...

Area of Science:

• Agricultural Engineering
• Environmental Monitoring
• Artificial Intelligence

Background:

• Existing facility environment prediction models lack accuracy and timeliness, hindering precise environmental regulation in agricultural settings.
• Challenges include multifactor nonlinear coupling and error accumulation in long-term predictions.

Purpose of the Study:

• To develop a novel facility environment prediction model to overcome limitations of existing methods.
• To enhance the accuracy and timeliness of environmental predictions for agricultural facilities.

Main Methods:

• Proposed a Long Short-Term Memory network with Attention (LSTM-AT) and Data Preprocessing (DP) model.
• Data Preprocessing involved Wavelet Threshold Denoising and Sliding Window techniques.
• An Attention Mechanism dynamically weighted features for improved temporal modeling.

Main Results:

• Achieved high determination coefficients (R²) for 24-hour predictions: 0.9602 (temperature), 0.9529 (humidity), and 0.9839 (radiation).
• Demonstrated significant improvements over baseline LSTM models, with notable gains in humidity prediction.
• Effectively suppressed error accumulation in long-term forecasts.

Conclusions:

• The LSTM-AT-DP model significantly enhances prediction accuracy and reliability for facility environments.
• The attention mechanism is crucial for identifying and weighting critical temporal features.
• This provides robust technical support for precise agricultural facility environmental regulation.