Analysis of Machine Learning Models for Anomaly Detection Using PMU data

Abstract

Accurate short-term electricity price forecasting (STEPF) is critical for efficient energy market operations, guiding investment strategies, resource allocation, and consumer behavior. This study introduces a hybrid deep learning approach specifically designed to improve STEPF accuracy by leveraging historical Hourly Ontario Energy Price (HOEP) data from 2017 to 2019. The model integrates advanced techniques, including data preprocessing and denoising through a Stacked Denoising Autoencoder (SDAE), along with enhanced temporal modeling via Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (GRU) networks. By capturing the complex dynamics inherent in electricity pricing data, the proposed hybrid model significantly enhances forecasting accuracy. Trained on data from 2017 and 2018, with 2019 used for testing, the model achieves a strong correlation coefficient (R = 99.86%) and substantially lowers forecasting errors. Comparative evaluations against established forecasting methods highlight the model's superior performance. This work demonstrates the practical value of deep learning techniques in the energy sector, particularly in responding to the volatility of demand and supply in real-time electricity markets.

Authors

Arash Moradzadeh, Mostafa Mouhammadpourfard, Yang Weng, Suhas Pol, SM Muyeen

Keywords

Energy management, Short-term electricity price forecasting, Hybrid deep learning models, Temporal data analysis,Electricity market forecasting

---

Publication Type

Conference

---

Digital Object Identifier

https://doi.org/10.1109/TPEC63981.2025.10906930

---

Full Citation

A. Moradzadeh, M. Mouhammadpourfard, Y. Weng, S. Pol and S. M. Muyeen, Hybrid Deep Learning Model for Accurate Short-Term Electricity Price Forecasting, 2025 IEEE Texas Power and Energy Conference (TPEC), College Station, TX, USA, 2025, pp.

View Article