PV POWER INTERVAL PREDICTION BASED ON STL DECOMPOSITION AND TPA MECHANISM
Aiming at the problem that the PV power point prediction contains insufficient information to provide a sufficient basis for grid scheduling,a PV interval prediction method based on STL decomposition and temporal pattern attention mechanism is proposed.Firstly,the original PV power sequence is decomposed by STL to obtain subsequences:trend term,seasonal term and residual term.Then,the trend term is predicted by extreme learning machine(ELM);the seasonal term and the residual term are predicted by bi-directional gated recurrent unit(BiGRU)based on temporal pattern attention(TPA).Finally,the interval prediction is obtained by quantile regression,and the PV output interval prediction is obtained by superposing the two outputs.The proposed method is validated by the point prediction results and interval prediction results on the PV output dataset of a certain place in Hunan province.
photovoltaic power generationpower predictionneural networkquantile regressionbidirectional gated cyclic unit network
NETL
NSTL
万方数据
