Abstract: In response to the issue of the drastic fluctuation in PV power output in a short period of time under transitional weather, which will result in lower prediction accuracy, a PV power day-ahead prediction method is proposed based on multi-type weather identification. A new model with multi-layer division is proposed for weather identification. Firstly, the weather state index is utilized to reflect the weather characteristics. Secondly, a Gaussian mixture model is employed to extract the power fluctuation characteristics in clustering way. Finally, these two models are crossed and combined based on the concept drift algorithm, so as to distinguish the turning weather days and four kinds of smooth weather days for improving the weather type identification precision. Meanwhile, an interval prediction model based on quantile regression is proposed for power prediction. Firstly, the significant meteorological features of the five weather types are selected according to the transfer entropy respectively, taking into full consideration the specificity of weather patterns. Subsequently, to enhance the model’s generalization ability, the multilayer perceptron neural network, convolutional neural network, and bidirectional long- and short-term memory neural network are modularly integrated. Finally, the neural network quantile regression model is combined and the prediction interval is generated. The effectiveness of the proposed model in point prediction and interval prediction is verified using the data collected from a photovoltaic field located in Shanghai, China.