Abstract: The outputs of hydropower, wind power and photovoltaic in a hydro-wind-photovoltaic complementary energy system (HWPCES) are integrated into the power grid. The traditional method forecasts these three outputs separately and then sums them up as the system’s total power capacity, but such a method suffers from error accumulation and lacks consideration of spatiotemporal complementarity. To improve the forecasting accuracy, first we consider spatiotemporal correlation and complementarity, and select certain predictors from the teleconnection factors and power factors. Then, a point forecasting model and an interval forecasting model are constructed based on the Long Short-term Memory network and the Lower Upper Bound Estimation method. Finally, joint forecasting of the power outputs is achieved. This study selects the Ertan HWPCES as the case study. The results show that in the verification period of total power forecasting, its Nash-Sutcliffe efficiency coefficient reaches 0.908 by the joint forecasting method, or an increase of 0.016 compared to the accumulation method. The interval forecasting method achieves a reduction of 0.352 in the coverage width-based criterion. Our new method is useful for complementary operation of hydropower, wind power, and photovoltaic.

Key words: hydro-wind-photovoltaic complementary energy system, joint forecast of power output, point forecast, interval forecast, Long Short-term Memory, Lower Upper Bound Estimation