A hybrid temporal convolutional network and Prophet model for power load forecasting

Abstract Accurate and effective power system load forecasting is an important prerequisite for the safe stable operation of grid normal production society. In recent years, convolutional neural networks (CNNs) have been widely used in time series prediction due to their parallel computing other characteristics, but it difficult CNNs capture relationship sequence context meanwhile, easily leads information leakage. To avoid drawbacks CNNs, we adopt a temporal network (TCN), specially designed series. TCN combines causal convolution, dilated residual connection, fully considers correlation between historical data future data. Considering that has strong periodicity greatly influenced by seasons holidays, Prophet model decompose fit trend component, season holiday component. We use forecast respectively, then least square method fuse two models, make respective advantages improve accuracy. Experiments show proposed TCN-Prophet higher accuracy than classic ARIMA, RNN, LSTM, GRU, some ensemble can provide more decision-making references departments.