Authors:
S. Rubin Bose, J. Angelin Jeba, N. Christy Evangeline, R. Regin, J. Rahila, Madhan Raj Gopi Akila

Addresses:
Department of Electronics and Communication Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Electronics and Communication Engineering, S.A. Engineering College, Chennai, Tamil Nadu, India. Department of Electronics and Instrumentation Engineering, Madras Institute of Technology, Chennai, Tamil Nadu, India. School of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Electrical and Electronics Engineering, Dhaanish Ahmed College of Engineering, Chennai, Tamil Nadu, India.

Abstract:

The increasing integration of photovoltaic (PV) systems into electrical grids presents significant challenges to grid stability due to the inherent volatility and intermittency of solar power generation. Accurate short-term power forecasting is therefore critical for effective grid management, optimal resource scheduling, and maintaining a reliable balance between energy supply and demand. This paper presents a deep learning methodology for forecasting plant-level solar power output at a 15-minute temporal resolution. A Long Short-Term Memory (LSTM) network, a specialised type of recurrent neural network, is developed to predict DC power output from a solar power plant. The model leverages a unified dataset comprising historical power generation data from 22 inverters and co-located meteorological sensor readings, including solar irradiation and ambient and module temperatures. A rigorous data pretreatment pipeline uses advanced time-series feature engineering techniques, such as lag features, rolling-window statistics, and sine-cosine transforms for cyclical temporal features. A recursive forecasting technique trains a stacked LSTM architecture to predict power output. The model's performance is assessed using regression measures such as MAE, RMSE, and R^2$ on an unseen test dataset. The model's high accuracy in capturing the complex, non-linear dynamics of solar power generation shows that LSTM-based approaches can improve the operational efficiency and reliability of modern power grids with high renewable energy penetration.

Keywords: Photovoltaic Power Forecasting; Long Short-Term Memory (LSTM); Time-Series Analysis (TSA); Renewable Energy Integration; Grid Stability; Machine Learning (ML).

Received on: 02/04/2025, Revised on: 21/06/2025, Accepted on: 28/07/2025, Published on: 03/01/2026

DOI: 10.69888/FTSCL.2026.000600

FMDB Transactions on Sustainable Computer Letters, 2026 Vol. 4 No. 1, Pages: 55–64

• Views : 89
• Downloads : 12

Download PDF