The uncertainty associated with solar photo-voltaic (PV) power output is a big challenge to design, manage and implement effective demand response and management strategies. Therefore, an accurate PV power output forecast is an utmost importance to allow seamless integration and a higher level of penetration. In this research, a neural network ensemble (NNE) scheme is proposed, which is based on particle swarm optimization (PSO) trained feedforward neural network (FNN). Five different FFN structures with varying network complexities are used to achieve the diverse and accurate forecast results. These results are combined using trim aggregation after removing the upper and lower forecast error extremes. Correlated variables namely wavelet transformed historical power output of PV, solar irradiance, wind speed, temperature and humidity are applied as inputs to the multivariate NNE. Clearness index is used to classify days into clear, cloudy and partial cloudy days. Test case studies are designed to predict the solar output for these days selected from all seasons. The performance of the proposed framework is analyzed by applying training data set of different resolution, length and quality from seven solar PV sites of the University of Queensland, Australia. The forecast results demonstrate that the proposed framework improves the forecast accuracy significantly in comparison with individual and benchmark models. Index Terms-PV power output forecasting, solar irradiance, neural network ensemble (NNE), ensemble network (EN), particle swarm optimization, clearness index, cloudy day (CLD), partially cloudy day (PCD) and clear day (CD).