Proactively selection of input variables based on information gain factors for deep learning models in short-term solar irradiance forecasting

As the proportion of solar power generation increases, accurate solar irradiance forecast used to connect solar power to the grid has become crucial. Multi-parameter prediction is one of the most commonly-used methods for solar irradiance forecast. Effective additional variables can improve the accuracy of the model, while invalid additional variables can lead to over fitting or under fitting of the model. To address this issue, this paper proposes the information gain factor as the basis for proactively selecting input variables. Firstly, the experiment combines 10 kinds of additional variables with GHI and inputs them into five models: auto regressive model (AR), gradient boosting decision tree (GBDT), convolutional neural network (CNN), long short-term memory (LSTM) and convolutional long short-term memory (ConvLSTM). Then, the impact of additional variables on prediction accuracy is analyzed and used as a basis for verifying the feasibility of the proposed method. Finally, the Pearson correlations and information gain factors between these variables and GHI are calculated separately. The results indicate that the information gain factor is more suitable as a basis for selecting input variables than the Pearson coefficient.

Automated summary

This paper proposes an approach to proactively select input variables based on the information gain factor for improving the accuracy of solar irradiance forecast. The feasibility of this method is validated through experiments, and the suitability of the information gain factor is compared with Pearson correlations.