In a particular time scale (0–6 h)

prediction errors of photovoltaic (PV) power stations are spatio-temporally dependent on surrounding power stations. From the point of view of power stations themselves

they have a conditional-temporal dependency. Therefore

considering these characteristics

prediction accuracy of the PV output can be improved effectively. A short-term PV output probability prediction method is developed based on appearance similarity updating (ASU) and a long short-term memory network (LSTM). Results from Jilin and Inner Mongolia regions show that accuracy of the prediction method considering temporal and spatial dependence of the prediction error is higher than that of a prediction method considering only a single property

and annual root mean square error can be reduced by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$$ 3%. Moreover

conditional dependency and temporal correlation of the updated prediction error are better than those of conventional interval prediction.