Akademik Veri Yönetim Sistemi
Computational Economics, 2025 (SCI-Expanded)
This study addressed the challenge of accurately predicting electricity prices, a critical task for energy market stakeholders. First, ANN models compared with five deep learning models as CNN, LSTM, CNN-LSTM, EMD-CNN-LSTM, and CEEMDAN. Second, employing Wavelet Coherence Analysis helps to understand the relationships. Specifically, four models were developed, each incorporating a different industry production index while keeping other inputs constant, then these four models are employed for the ANN and the other deep learning models. This approach integrated diverse economic indicators, aiming to create more robust and comprehensive models for Türkiye. The results revealed that the CEEMDAN-CNN-LSTM model performed best, with an MSE of 0.0098, outperforming simpler models CNN, LSTM, and ANN models in predictive accuracy and stability. However, in some cases simplier architectures as ANN Model 4 (R = 0.958, MSE = 2.48E-03) outperformed complex models. Additionally, Wavelet Coherence Analysis revealed significant phase alignment between the energy industry production index and electricity prices at medium frequencies during two distinct periods. These results confirmed the ANN models’ and deep learning models’ effectiveness and highlighted the dynamic relationships between electricity prices and economic indicators. This dual-method approach offered a novel and robust tool for electricity price forecasting, surpassing previous efforts by providing deeper insights and higher accuracy. This study’s findings were significant for decision-makers in the energy sector, as they provide more reliable forecasting tools, enhancing strategic planning and market operations.