Temporal Convolutional Networks for Cloud-Based Renewable Energy Forecasting: A Deep Learning Approach to Power Production and Battery Health Prediction

• Published in: Conference record of the Industry Applications Conference pp. 1 - 6
• Main Authors: Tsai, Po-Yang, Lee, Tsung-Lu Michael, Wu, Chung Yi, Huang, Chien-Chung
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.06.2025
• Subjects: Batteries; Battery State of Health; Cloud-based Prediction; Convolutional neural networks; Deep learning; Forecasting; Power system stability; Predictive models; Production; Renewable Energy Forecasting; Renewable energy sources; Smoothing methods; Stability analysis; Temporal Convolutional Networks
• ISSN: 2576-702X
• DOI: 10.1109/IAS62731.2025.11061607

This paper presents a cloud-based forecasting system for renewable energy generation and battery State of Health (SOH) prediction using Temporal Convolutional Networks (TCN). As the adoption of renewable energy accelerates, precise forecasting of both power production and storage capacity becomes critical for grid stability and efficiency. Using hourly data collected from operational renewable plants and energy storage systems over one year, our methodology employs polynomial regression for data smoothing before applying a five-layer TCN architecture. The model demonstrates a superior temporal pattern recognition compared to traditional LSTM approaches, particularly for multi-horizon forecasting of intermittent resources. Containerized and deployed via Google Cloud's Vertex AI with Grafana visualization, the system enables scalable implementation across diverse renewable assets. The results show an effective capture of both short-term variability in power production and long-term battery degradation trends, which is essential to optimize renewable energy dispatch and storage operations.