A Novel Hybrid Model for Electricity Price Forecasting Based on the Integration of Bi-directional Long Short-Term Memory and Gated Recurrent Unit
Year: 2024
Type: dissertation
Abstract: The prediction of electricity prices plays a pivotal role in the wholesale electricity markets, influencing sale prices, bidding strategies, electricity dispatch, control, and the management of market. Notably, forecasting in a deregulated electricity market is challenging due to multiple factors such as high volatility, non-stationarity and multi-seasonality of electricity prices. In response to this challenge, this research proposes a novel hybrid deep learning model employing Bi-directional Long Short-Term Memory (Bi_LSTM) and Gated Recurrent Unit (GRU) for real-time electricity price forecasting. In this model, the output sequences from the Bi_LSTM layer, which captures both past and future temporal dependencies, are directly fed into the GRU layer to refine the feature extraction. This hybrid approach not only reduces overfitting risk of a single model, but also increases robustness and adaptability of model. Three studies are conducted in New York City (NYC), electricity market to evaluate the model by systematically comparing the obtained results. First, the proposed model, Bi_LSTM-GRU, outperforms several baseline models, spanning a statistical time-series method: Auto Regressive Integrated Moving Average (ARIMA), Machine Learning approaches: Linear Regression (LR), Random Forest (RF), eXtreme Gradient Boosting (XGB), and Support Vector Regression (SVR), and Deep Learning techniques: Long Short-Term Memory (LSTM), Bi-LSTM, GRU, and Convolutional Neural Network (CNN). Secondly, the possibility of hybridizing CNN and Recurrent Neural Network (RNN) architectures has been examined. The proposed model also surpasses CNN-LSTM, CNN-Bi-LSTM, and CNN-GRU. Lastly, the potential contribution of data decomposition techniques in enhancing the proposed model has been assessed. It is found out that adding Wavelet Transform (WT) or Fourrier Transform (FT) to decompose the data leads to higher error rates.
Author: Hajigholam Saryazdi, Amirhossein
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7002478998