Forecasting Agricultural Trade Based on TCN-LightGBM Models: A Data-Driven Decision

Research on World Agricultural Economy, Journal Year: 2025, Volume and Issue: unknown, P. 207 - 221

Published: Jan. 16, 2025

Facing the increasing complexity and dynamic fluctuations of global agricultural trade market, accurate forecasting plays a key role in supporting policy formulation, stabilising market optimising resource allocation. In order to increase precision stability predictions, this research suggests hybrid model built on temporal convolution network (TCN) lightweight gradient boosting tree (LightGBM). The TCN module effectively captures long-term dependence characteristics time series data through dilated convolution, which improves model’s ability identify seasonal periodic trends. LightGBM module, other hand, makes use decision trees excels at efficiently handling nonlinear relationships avoiding overfitting. Experimental results show that TCN-LightGBM outperforms traditional models terms mean square error (MSE), absolute (MAE) prediction accuracy. Specifically, compared with ARIMA, LSTM, alone or alone, achieves accuracy 91.3% test data, MSE MAE 0.021 0.115 respectively, significantly improving stability. addition, parameter sensitivity analysis shows maintains highly consistent trend under different configurations, verifies robustness its practical application value. This study provides data-driven support tool high strong stability, providing new solution for complex tasks.

Language: Английский

---

A Novel Hybrid Model Combining LMD, MSCA, and SCINet for Electricity Forecasting

Electronics, Journal Year: 2025, Volume and Issue: 14(8), P. 1567 - 1567

Published: April 12, 2025

To address the challenges of modeling complex nonlinear features and multi-scale temporal patterns in electricity forecasting, this paper proposes a SCINet model optimized with Local Mean Decomposition (LMD) Multi-Scale Channel Attention (MSCA). The first applies LMD to decompose original consumption sequence into multiple intrinsic mode functions, effectively capturing trends fluctuations at different frequencies. Building on framework, MSCA module is introduced enhance model’s ability focus critical through feature extraction inter-channel correlation modeling, enabling it better capture variations dependencies across time scales. Experimental results show that single-step, short-term prediction, proposed LMD-MSCA-SCINet achieves prediction MAE = 0.1539, MSE 0.0901 R2 0.9003, which are 63.4% 72.1% lower than Informer (MAE 0.4207, 0.3235), respectively, further reduce by 37.3% 63.4%,respectively, compared basic 0.245, 0.246). These verify superiority practical value method handling power forecasting tasks.

Language: Английский

---

Time-delay reservoir computing based on VCSEL for short-term load forecasting

Optics Communications, Journal Year: 2025, Volume and Issue: unknown, P. 131838 - 131838

Published: April 1, 2025

Language: Английский

---

Short-Term Power Load Forecasting Based on CEEMDAN-WT-VMD Joint Denoising and BiTCN-BiGRU-Attention

Electronics, Journal Year: 2025, Volume and Issue: 14(9), P. 1871 - 1871

Published: May 4, 2025

Short-term power load forecasting is crucial for safe grid operation. To address the insufficiency of traditional decomposition methods in suppressing high-frequency noise within multi-source noisy time series, this study proposes a hybrid model integrating CEEMDAN-WT-VMD joint denoising with BiTCN-BiGRU-Attention architecture. The methodology comprises three stages: (1) CEEMDAN raw data to mitigate mode mixing and extract stationary IMF components; (2) wavelet threshold filter interference while preserving reconstructing low-frequency signals; (3) secondary feature using Variational Mode Decomposition (VMD) enhance stability. A architecture combines Bidirectional Temporal Convolutional Network (BiTCN) long-term dependency capture, Gated Recurrent Unit (BiGRU) dynamic extraction, an attention mechanism key pattern emphasis. final value generated by progressively accumulating predictions decomposed components. Empirical analysis based on from region Australia demonstrates that, through horizontal vertical comparative experiments, proposed method significant improvements both accuracy stability compared other frontier models.

Language: Английский

---

Integrated Energy System Load Forecasting with Spatially Transferable Loads

Energies, Journal Year: 2024, Volume and Issue: 17(19), P. 4843 - 4843

Published: Sept. 27, 2024

In the era of dual carbon, rapid development various types microgrid parks featuring multi-heterogeneous energy coupling presents new challenges in accurately modeling spatial and temporal load characteristics due to increasingly complex source–load diversified interaction patterns. This study proposes a short-term forecasting method for an interconnected park-level integrated system using data center as case study. By leveraging spatially transferable heterogeneous correlation among electricity–cooling–heat loads, optimal feature set is selected effectively characterize loads Spearman analysis. fed into multi-task learning (MTL) combined with convolutional neural network (CNN) long- memory (LSTM) model generate prediction results. The simulation results demonstrate efficacy our proposed approach characterizing across different parks, enhancing track “spikes” achieving superior accuracy.

Language: Английский

---

A singular spectrum analysis-enhanced BiTCN-selfattention model for runoff prediction

Earth Science Informatics, Journal Year: 2024, Volume and Issue: 18(1)

Published: Dec. 12, 2024

Language: Английский

---