Optimizing grid efficiency and guaranteeing stable energy supply depend on accurate estimate of the generation of energy, especially renewable energy. However, traditional forecasting techniques face challenges due to the inherent uncertainty and non-linear patterns in renewable energy production estimation, as it’s highly depends on nature. We propose encoder-based model idea for time-series task like solar energy forecasting. Our main proposition is to examine how a simple transformer-encoder-based model perform forecasting renewable energy. Our model uses the transformer architecture’s self-attention mechanism to identify intricate temporal patterns and dependencies. The model manages the non-stationary nature of solar energy data by combining residual connections with layer normalization and cyclical encoding for temporal features. We tested the suggested model on a large data set obtained from several solar power plants, showing that it performs better in terms of established error matrices. The outcomes demonstrate the model’s superiority over other proposed methods, offering a reliable and expandable solar energy forecasting solution. This work offers a promising framework for incorporating machine learning into smart grid systems and advances data-driven methods to optimize renewable energy consumption and more efficient demand and response (DR) system for the power grid.