Short-Term Photovoltaic Power Prediction Based on Similar-Day Clustering and Multi-Layer Perception

Artificial intelligence plays an important role in improving the accuracy and reliability of photovoltaic power prediction. In order to better improve the accuracy of photovoltaic power prediction, present mainstream research focuses on using the optimizing algorithm to select the hyper-parameters of the prediction model to improve the prediction level of the model. This type of model uses an optimizing algorithm to achieve the improvement of prediction accuracy. However, it doesn’ t do too much work in the original data, which partly ignores the inner value of the original data. To solve that problem, an intelligent photovoltaic power prediction model based on similar-day clustering and multi-layer perceptron is established. The ability from convolutional neural network and multi-layer perception are combined. Firstly, select the satellite cloud picture and use the convolutional neural network to work at data feature clustering, which is proposed as similar-day clustering. Sequentially the original data is divided into subsets. Secondly, the pretrained multi-layer perceptron models according to different weather types are selected. The historical metrological data are used as the model input to train the multi-layer perceptron. Thus, multi-layer perceptron can learn the nonlinear mapping relationship and coupling connection between meteorological factors and photovoltaic power. Finally, a data set of a photovoltaic power plant in northern China is used for the case study. The case study shows that compared with LSTM and multi-layer perceptron, the MAPE of the proposed prediction model is improved by 19.6% and 50.7% respectively, which fully demonstrates the superiority and accuracy of the model.