This paper presents the Sustainable Energy Forecasting Network (SEF-Net), a novel deep learning architecture for multi-step time series forecasting in sustainable energy markets. SEF-Net integrates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Transformer models to capture complex patterns and dependencies in energy price data. We evaluate SEF-Net’s performance using a comprehensive dataset from a major European energy exchange, comparing it against traditional time series models and individual deep learning approaches. Results demonstrate that SEF-Net consistently out-performs existing methods across multiple forecasting horizons, achieving a 14.6% improvement in Mean Absolute Percentage Error for 168-h forecasts. The architecture’s ability to extract relevant features, model temporal dependencies, and capture long-range interactions makes it particularly well-suited for predicting future energy demand and prices in volatile sustainable energy markets.