Solar Forecast Reconciliation and Effects of Improved Base Forecasts

Forecasting of solar PV generation plays an important role in power system operations. Forecasts are required on various geographical and temporal scales, which can be modeled as hierarchies. In a geographical hierarchy, the overall forecast for the region can either be obtained by directly forecasting the regional time series or by aggregating the individual forecasts generated for the sub-regions. This leads to a problem known as aggregate inconsistency as the two sets of forecasts are most likely different due to modeling uncertainties. Hence, practice is not optimal. Statistically optimal aggregation known as reconciliation, has been proven to provide aggregate consistent forecasts. Reconciliation helps system operators to have a superior foresight in a region-wise level, which eventually results in efficient system planning. The focus of this paper is on improving reconciliation accuracy. In addition, the effects of more accurate disaggregated and aggregated forecasts on the final reconciled predictions have been analyzed. A total of 318 simulated PV plants in California have been used to build a geographical hierarchy. More accurate NWP based and aggregated level forecasts are obtained with model output statistics and artificial neural network models. Significant improvements are observed in reconciled forecasts without using any exogenous information.