Electrical Load Forecasting using a Novel BI-GRU Encoder Decoder Model

Abstract

Precisely forecasting electrical load, especially through univariate time series analysis, is pivotal for effectively operating and planning power systems. This research introduces a hybrid model leveraging univariate time series and deep learning techniques. The model combines the Bidirectional Gated Recurrent Units (Bi-GRU) based encoder-decoder structure with the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture to enhance single-step short-term electrical demand forecasting. Integrating Bi-GRUs ensures adept capture of temporal dependencies, while CNNs meticulously extract spatial features. Concurrently, LSTMs provide a robust mechanism to memorize long-range dependencies. The model's competence was rigorously assessed through evaluations using the publicly available American Electric Power (AEP) dataset, which represents real-world electrical load patterns. Findings highlight that the proposed model outstrips competing models in algorithmic stability and prediction accuracy. With a Mean Absolute Percentage Error (MAPE) of 80.032, this research posits a promising avenue for utilizing deep learning in univariate time series power load prediction.