UrbanEV Veri Setinde Elektrikli Araç Şarj Talebi Tahmini: Zaman Serisi, Makine Öğrenmesi ve Derin Öğrenme Modellerinin Sistematik Değerlendirmesi

Öz

Elektrikli araç (EV) şarj talebi tahmini, akıllı şebeke operasyonları ve altyapı planlaması için kritik öneme sahiptir. Literatürde makine öğrenmesi ve derin öğrenme yöntemlerinin üstünlüğü varsayılmaktadır; ancak bu çalışmalar sıklıkla naive baseline modellerini göz ardı etmektedir. Bu çalışmamızda, 11 farklı tahmin yönteminin (4 naive/zaman serisi, 5 makine öğrenmesi, 2 derin öğrenme) UrbanEV veri seti üzerinde kapsamlı karşılaştırmasını sunmaktadır. Sonuçlar, Persistence ve Exponential Smoothing modellerinin en düşük RMSE (0.1109) değerlerini elde ettiğini göstermektedir. LSTM (%64.74 MAPE) ve Random Forest (%91.83 MAPE) gibi karmaşık modeller beklenenin altında kalmıştır. GCN (0.1986 RMSE) LSTM'den iyi performans göstermiş ancak 316 saniyelik eğitim süresi ile maliyet-etkin değildir. Veri karakteristiği analizi, yüksek otokorelasyonun (72 saat) basit modelleri avantajlı kıldığını, zayıf mekânsal korelasyonun (ortalama 0.093) GCN potansiyelini sınırladığını ortaya koymaktadır. Bulgular, veri yapısına uygun model seçiminin karmaşıklıktan daha önemli olduğunu ve operasyonel uygulamalarda hesaplama verimliliğinin göz ardı edilmemesi gerektiğini göstermektedir.

Anahtar Kelimeler

• Elektrikli araç şarj talebi tahmini
• Naive zaman serisi modelleri
• Derin öğrenme (LSTM
• GCN)
• Veri karakteristiği

Systematic Evaluation of Time Series, Machine Learning, and Deep Learning Models for Electric Vehicle Charging Demand Forecasting on the UrbanEV Dataset

Öz

Electric vehicle (EV) charging demand forecasting is of critical importance for smart grid operations and infrastructure planning. Although machine learning and deep learning methods are often assumed to be superior in the literature, many studies tend to overlook naïve baseline models. In this study, we present a comprehensive comparison of 11 different forecasting methods (4 naïve/time-series models, 5 machine learning models, and 2 deep learning models) using the UrbanEV dataset. The results indicate that Persistence and Exponential Smoothing models achieve the lowest RMSE value (0.1109). In contrast, more complex models such as LSTM (64.74% MAPE) and Random Forest (91.83% MAPE) perform below expectations. While GCN achieves better performance than LSTM (0.1986 RMSE), its 316-second training time renders it cost-inefficient for practical applications. An analysis of data characteristics reveals that high temporal autocorrelation (72 hours) favors simpler models, whereas weak spatial correlation (mean 0.093) limits the potential benefits of GCN-based approaches. Overall, the findings demonstrate that model selection aligned with data structure is more critical than model complexity, and that computational efficiency should not be neglected in operational forecasting scenarios.

Anahtar Kelimeler

• Electric vehicle charging demand forecasting
• naïve time series models
• deep learning (LSTM
• GCN)
• data characteristics.

Kaynakça

1. Ainsworth, S. K., Hayase, J., & Srinivasa, S. (2022). Git Re-Basin: Merging models modulo permutation symmetries. arXiv:2209.04836.
2. Breiman, L. (1996). Bagging predictors. Machine Learning, 24(2), 123–140.
3. Department for Transport. (2023). Road Safety Open Data (STATS19): 2023 collision file and coded-data guide, Great Britain.
4. Dietterich, T. G. (2000). Ensemble methods in machine learning. In Multiple Classifier Systems (pp. 1–15). Springer. https://doi.org/10.1007/3-540-45014-9_1
5. Fanaee-T, H., & Gama, J. (2013). Event labeling combining ensemble detectors and background knowledge. Progress in Artificial Intelligence, 2(2–3), 113–127. https://doi.org/10.1007/s13748-013-0040-3
6. Frankle, J., Dziugaite, G. K., Roy, D. M., & Carbin, M. (2020). Linear mode connectivity and the lottery ticket hypothesis. International Conference on Learning Representations.
7. Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of Computer and System Sciences, 55(1), 119–139.
8. Garouani, M., Barhrhouj, A., & Teste, O. (2025). XStacking: An effective and inherently explainable framework for stacked ensemble learning. Information Fusion, 124, 103358. https://doi.org/10.1016/j.inffus.2025.103358

9. Ho, T. K. (1998). The random subspace method for constructing decision forests. IEEE Transactions on Pattern Analysis and Machine Intelligence, 20(8), 832–844.
10. Horoi, S., Orozco Camacho, A. M., Belilovsky, E., & Wolf, G. (2024). Harmony in Diversity: Merging Neural Networks with Canonical Correlation Analysis. Proceedings of the 41st International Conference on Machine Learning, PMLR 235, 18815–18832.
11. Ilharco, G., Ribeiro, M. T., Wortsman, M., Gururangan, S., Schmidt, L., Hajishirzi, H., & Farhadi, A. (2023). Editing Models with Task Arithmetic. International Conference on Learning Representations.
12. Izmailov, P., Podoprikhin, D., Garipov, T., Vetrov, D., & Wilson, A. G. (2018). Averaging weights leads to wider optima and better generalization. Proceedings of the 34th Conference on Uncertainty in Artificial Intelligence.
13. Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2017). Simple and scalable predictive uncertainty estimation using deep ensembles. Advances in Neural Information Processing Systems, 30.
14. Lu, Z., Fan, C., Wei, W., Qu, X., Chen, D., & Cheng, Y. (2024). Twin-Merging: Dynamic integration of modular expertise in model merging. Advances in Neural Information Processing Systems, 37.
15. McMahan, H. B., Moore, E., Ramage, D., Hampson, S., & y Arcas, B. A. (2017). Communication-efficient learning of deep networks from decentralized data. Proceedings of AISTATS, PMLR 54, 1273–1282.
16. West, D., Dellana, S., & Qian, J. (2005). Neural network ensemble strategies for financial decision applications. Computers & Operations Research, 32(10), 2543–2559. https://doi.org/10.1016/j.cor.2004.03.017
17. Wolpert, D. H. (1992). Stacked generalization. Neural Networks, 5(2), 241–259.
18. Wortsman, M., Ilharco, G., Gadre, S. Y., et al. (2022). Model soups: Averaging weights of multiple fine-tuned models improves accuracy without increasing inference time. Proceedings of the 39th International Conference on Machine Learning, PMLR 162.
19. Yadav, P., Tam, D., Choshen, L., Raffel, C. A., & Bansal, M. (2023). TIES-Merging: Resolving interference when merging models. Advances in Neural Information Processing Systems, 36.
20. Yang, E., Tang, A., Shen, L., Guo, G., Wang, X., Cao, X., & Zhang, J. M. (2025). Continual model merging without data: Dual projections for balancing stability and plasticity. Advances in Neural Information Processing Systems, 38.
21. Yu, L., Yu, B., Yu, H., Huang, F., & Li, Y. (2024). Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch. Proceedings of the 41st International Conference on Machine Learning, PMLR 235, 57755–57775.
22. Zhou, Z.-H. (2012). Ensemble Methods: Foundations and Algorithms. Chapman & Hall/CRC