Open Access

ARTICLE

GBCTT: A Novel Multi-Factor Load Forecasting Model for Industrial and Commercial User Groups

Wanxing Sheng¹, Xiaoyu Yang¹, Dongli Jia¹, Keyan Liu¹, Zewei Chen^2,*
1 China Electric Power Research Institute, Beijing, 100000, China
2 School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, 100000, China
* Corresponding Author: Zewei Chen. Email:

Energy Engineering https://doi.org/10.32604/ee.2026.075810

Received 09 November 2025; Accepted 30 December 2025; Published online 21 January 2026

Abstract

Accurate forecasting of electricity consumption patterns is a fundamental task of power demand-side management (DSM), particularly for industrial and commercial users who significantly influence market supply-demand balance and price fluctuations. Traditional forecasting methods, including statistical models and deep learning approaches, often struggle to capture the complex multi-factor, non-linear, and spatio-temporal dependencies inherent in power load data. To address these limitations, this paper introduces GBCTT (GAT-TBA-CNN-TCN), a novel multi-factor load forecasting model. The model integrates a Graph Attention Network (GAT) to dynamically learn spatial dependencies among heterogeneous influencing factors (e.g., temperature, humidity, electricity prices), a Time series Behavior Analysis (TBA) module incorporating CNN for multi-scale temporal decomposition (trend, seasonality, residuals), and a Temporal Convolutional Network (TCN) for capturing long-range temporal dependencies. Comprehensive experiments were conducted on the public Panama Load Dataset and a proprietary dataset from China encompassing 16 distinct industrial and commercial sectors. The proposed GBCTT model was evaluated against baseline models including ARIMAX, BiLSTM, and Autoformer across prediction horizons of 1, 3, and 7 days. Results demonstrate that GBCTT consistently outperforms all baseline models across key evaluation metrics (MAE, MSE, RMSE, MAPE, MSPE), achieving state-of-the-art performance in short-term load forecasting. The ablation studies further validate the contribution of each core component to the model’s superior accuracy and robustness.

---

Keywords

Load forecasting; graph attention network; temporal convolutional network; industrial and commercial electricity consumption; spatio-temporal analysis

---