A metaheuristic-optimized hybrid deep learning and machine learning framework for robust market clearing price forecasting: A multi-scenario case study from Turkey
Journal of Engineering Research (Kuwait), 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.jer.2026.08.042
- Dergi Adı: Journal of Engineering Research (Kuwait)
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Arab World Research Source, Directory of Open Access Journals, Academic Search Ultimate (EBSCO)
- Anahtar Kelimeler: Hyperparameters, K-nearest neighbors, Long short-term memory, Market clearing price, Particle swarm optimization
- Atatürk Üniversitesi Adresli: Evet
Özet
In energy markets, the high volatility and nonlinear variation of the Market Clearing Price (MCP) have negatively impacted the performance of classical MCP forecasting algorithms. Predicting the MCP with high accuracy is extremely important, as it directly influences the decision-making processes of both generation side and demand side. In this study, a hybrid Long Short-Term Memory (LSTM) and K-Nearest Neighbors (KNN) method, with hyperparameters optimized using the Particle Swarm Optimization (PSO) algorithm, is used to estimate MCP with high accuracy. In the hybrid approach, LSTM is used because it learns the long-term dependencies of the time-dependent MCP on an hourly basis without losing them. To predict sudden fluctuations in MCP that LSTM cannot predict, the KNN method is used based on similar historical characteristics. In this study, the Sequence Length, Hidden Units, Dropout Rate, Initial Learning Rate, Mini-Batch Size, Number of Neighbors, and Alpha hyperparameters used in LSTM and KNN methods are optimized with the PSO algorithm. The performance of the prediction model created using the PSO-based LSTM-KNN hybrid method is evaluated for both the Expanding-Window Blocked Time-Series Cross-Validation and three different independent test data sets. The PSO-based LSTM-KNN hybrid method is compared with LSTM, Convolutional Neural Networks (CNN), LSTM-CNN, LSTM-Extreme Gradient Boosting (XGBoost), LSTM-Gaussian Process Regression (GPR) KNN, Naïve Persistence Forecasting (NPF), PSO-LSTM and PSO-KNN methods. The proposed PSO-based LSTM-KNN hybrid method is observed to yield very high results for all scenarios. In particular, it is found to produce significantly more successful prediction values compared to other models in independent test scenarios used to evaluate the model's prediction performance across different time periods.