Machine Learning Weather Analogs for Renewable Forecasting

Abstract

The Analog Ensemble (AnEn) is a technique to generate forecast ensembles using a deterministic model and associated observations. It has been widely used for weather forecasting and renewable energy production forecasting. Ensemble members are chosen based on a similarity metric that calculates the weighted multi-variate Euclidean distance. However, identifying optimal weights is a challenging and computationally expensive process, which often involves performing a constrained exhaustive search. As a result, only a few predictors were selected and optimized in previous AnEn studies. A new Machine Learning similarity metric is proposed address this shortcoming. First, a Deep Learning network is trained to generate latent features using all the temporal multi-variate input predictors. The analogs for a target forecast are selected by computing the Euclidean distance of the latent variables between the target and all previous forecasts. This process can be performed in parallel for each location. The proposed method does not require prior predictor selection and weight optimization. It is tested using wind speed and solar irradiance forecasts in Pennsylvania for 2017 to 2019. Results show that the proposed method is capable of handling a large number of predictors and it outperforms the original similarity metric in RMSE, bias, and CRPS. Since the transformation network is trained empirically, the proposed method has been found to be more flexible for searching through a longer historical record.