PHOTOVOLTAIC POWER FORECASTING METHOD BASED ON SKIP CONNECTION STRATEGY COMBINED WITH TCN-BiLSTM

Aimed at the problems of short-term nonlinearity of data and insufficient long-term periodicity capture in the process of photovoltaic power prediction, a photovoltaic power prediction method based on skip connection strategy and TCN-BiLSTM model is proposed. Similar samples were found through the grey correlation degree to form the prediction data set of the day. Temporal convolutional network (TCN) was used to extract local features to maintain the time series of features. The bi-directional long short-term memory (BiLSTM) network using the skip connection strategy was used to learn the long-term and short-term sequence features and combine the attention mechanism to adaptively focus on more important historical states. The experimental results based on actual power station data show that this method can effectively predict photovoltaic power generation, and has smaller prediction error and higher robustness than the general model.