Implementing Interval Prediction of PV Generation Using Bootstrap-Based Extreme Learning Machines

Power system operation facing renewable generation uncertainty has been investigated intensively, which requires the stochastic intervals of the uncertain variables for robustness. The interval prediction can be resolved based on statistical inference theory and artificial neuron networks. This paper provides the implementation of bootstrap-based extreme learning machines (BELMs) for PV generation interval prediction. First, the methodology of BELMs is revisited. Second, the implementation process, including data preprocessing, data selection, model fitting, hyperparameter tuning, and forecast and interval prediction, are explicitly presented. The proposed implementation is tested based on historical numerical weather and PV generation data. The results show that the application is effective in terms of reliability and sharpness.