Solar plus wind country-wide electrical power forecasts across successive years by optimized data matching

Forecasting the combined power generated from intermittent solar plus wind capacity hourly on a country-wide basis from underlying environmental and market variables poses challenges. An optimized data matching algorithm demonstrates its capabilities to do this with meaningful accuracy. For data recorded for the year 2016 matched with hourly records from year 2015, it achieves prediction accuracy of RMSE = 895 MW and R² = 0.9967. It does so by exploiting information from underlying variables, some of which are quite well correlated with the quantity of renewable power generation (e.g. wind speed and solar irradiance) and others that are very poorly correlated (e.g. quantity of precipitation, snow depth, cloud cover), without involving correlation calculations. The algorithm is also configured to provide ongoing short-term forecast, 1 h (t + 1) to 6 h (t + 6) ahead, throughout 2016 using historical data for different durations. For t + 1 forecasts across that year using a single tuned solution, it achieves forecasting accuracy of RMSE ~ 2100 MW; R² ~ 0.97. Comparisons with the forecasts of an autoregressive integrated moving average (ARIMA) model for January 2016 indicate that TOB is outperformed in forecasting accuracy for the t + 1 to t + 11 h ahead but that TOB outperforms ARIMA in forecast accuracy for t + 13 to t + 30 h ahead. This suggests that the two models have the potential to complement each other in day-ahead forecasting. Potential remains to further improve TOB’s forecasting accuracy by retuning the algorithm’s optimization on a regular basis, e.g. monthly or quarterly, enlarging the historic dataset used for matching and applying adjustments for capacity growth. The promising results justify multi-year studies as data becomes available for subsequent years.