The generated electricity of a solar plant exhibits complex nonlinear dynamics and uncertainties due to fluctuations in solar radiation and temperature. As a result, conventional algorithms struggle to accurately model this complexity, while machine learning algorithms can deliver the necessary forecasting performance. We tackled the issue of intelligent day-ahead solar power forecasting by utilizing a modified fuzzy neural network equipped with the multi-dimensional convolution block and fuzzy attention mechanism, which features a multi-dimensional convolution block and an advanced fuzzy attention mechanism. Unlike standard one-dimensional convolution blocks, our multi-dimensional convolution block is specifically designed to effectively compress and analyze time series data. This innovative design captures temporal dependencies at multiple resolutions, enabling a thorough representation of both short-term and long-term patterns in the hourly electricity output of a solar plant. The findings from our comparative simulation studies of the intelligent day-ahead solar power forecasting system based on the modified fuzzy neural network equipped with the multi-dimensional convolution block and fuzzy attention mechanism demonstrate its advantages and competitive performance relative to both the modified fuzzy neural network equipped with the fuzzy attention mechanism and the Informer model.