Wind power prediction in new stations based on knowledge of existing Stations: A cluster based multi source domain adaptation approach

Abstract

Historical wind power production figures are not available when a new wind farm goes into power production. It is thus difficult to forecast power productions of such wind farms that is required for demand management. Wind power is a function of weather variables and it is likely that weather patterns of the new station is similar to some existing operational wind farms. It will thus be interesting to investigate how the forecast/prediction models of the existing wind farms can be adapted to generate a prediction model for new stations. On this regard, we explore a particular branch of machine learning called Multi Source Domain Adaptation (MSDA). MSDA approaches identify a weighing mechanism to fuse the predictions from the source models (i.e. existing stations) to produce a prediction for the target (i.e. new station). The weights are computed based on similarity of data distributions between source and target. Conventional MSDA approaches utilise an instance based weighting scheme and we identified that fails to capture the data distribution of wind data sets appropriately. We thus propose a novel cluster based MSDA approach that captures wind data distribution in terms of natural groups that exist within data and compute distribution similarity (and source weight) in terms of cluster distributions. Experimental results demonstrate that cluster based MSDA approach can reduce regression error by 20.63% over instance based MSDA approach.