TITLE:
Performance Study of the RNN-LSTM-XGBoost Algorithm for Photovoltaic Power Forecasting
AUTHORS:
Zhengyu Yang
KEYWORDS:
Photovoltaic Power Generation, Power Forecasting, LSTM, RNN-LSTM, XGBoost, Performance Testing
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.11,
November
6,
2026
ABSTRACT: Among the forces now driving the expansion of world electricity supply, solar photovoltaic (PV) generation occupies a leading position. Figures assembled by the UK energy think tank Ember for its Global Electricity Review 2026 indicate that worldwide solar output amounted to 2778 TWh during 2025, some 30% above the level of the preceding year and equivalent to about 8.8% of all electricity produced globally. Since PV output is closely tied to weather conditions and consequently varies a great deal, the continued growth of installed capacity makes the accuracy of power forecasting, and its adaptability to different forecast horizons, increasingly important. Against this background, the present study examines how the RNN-LSTM-XGBoost hybrid model behaves in practice at several forecasting horizons. The algorithmic principle of the model is discussed first, experiments on measured PV data follow, and the forecasts obtained are then set against the measured values at a number of time horizons, so that the forecasting capability of the model can be judged comprehensively and the manner in which its accuracy changes with the forecast horizon can be examined.