An Improved iTransformer with RevIN and SSA for Greenhouse Soil Temperature Prediction

Agronomy, Journal Year: 2025, Volume and Issue: 15(1), P. 223 - 223

Published: Jan. 17, 2025

In contemporary agricultural practices, greenhouses serve as a critical component of infrastructure, where soil temperature plays vital role in enhancing pest management and regulating crop growth. However, achieving precise greenhouse environmental control continues to pose significant challenge. this context, the present study proposes ReSSA-iTransformer, an advanced predictive model engineered accurately forecast temperatures within across diverse temporal scales, encompassing both long-term short-term horizons. This capitalizes on iTransformer time-series forecasting framework integrates Singular Spectrum Analysis (SSA) decompose variables, thereby augmenting extraction pivotal features, such temperature. Furthermore, mitigate prevalent distribution shift issues inherent data, Reversible Instance Normalization (RevIN) is incorporated architecture. ReSSA-iTransformer adept at executing multi-step forecasts for extended immediate future intervals, offering comprehensive capabilities. Empirical evaluations substantiate that surpasses conventional models, including LSTM, Informer, Autoformer, all assessed metrics. Specifically, it attained R2 coefficients 98.51%, 97.03%, 97.26%, 94.83%, alongside MAE values 0.271, 0.501, 0.648, 1.633 predictions 3 h, 6 24 48 h respectively. These results highlight model’s superior accuracy robustness. Ultimately, not only provides dependable but also delivers actionable insights, facilitating enhanced practices.

Language: Английский

Application of an improved LSTM model based on FECA and CEEMDAN VMD decomposition in water quality prediction

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 14, 2025

To address the limitations of existing water quality prediction models in handling non-stationary data and capturing multi-scale features, this study proposes a hybrid model integrating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational (VMD), Long Short-Term Memory Network (LSTM), Frequency-Enhanced Channel Attention (FECA). The aims to improve accuracy robustness for complex dynamics, which is critical environmental protection sustainable resource management. First, CEEMDAN Sample Entropy (SE) were used decompose raw into interpretable components filter noise. Then, VMD-enhanced LSTM architecture embedded FECA was developed adaptively prioritize frequency-specific thereby improving model's ability handle nonlinear patterns. Results show that successful predicting all six indicators: NH₃-N (ammonia nitrogen), DO (dissolved oxygen), pH, TN (total TP phosphorus), CODMn (chemical oxygen demand, permanganate method). achieved Nash-Sutcliffe Efficiency (NSE) values ranging from 0.88 0.99. Using dissolved (DO) as an example, reduced Mean Absolute Percentage Error (MAPE) by 0.12% increased coefficient determination (R2) 0.20% compared baseline methods. This work provides robust framework real-time monitoring supports decision making pollution control ecosystem

Language: Английский