基于特征挖掘的ARIMA-GRU短期电力负荷预测

针对短期电力负荷随机性较强、预测精度较低的问题,提出了一种基于混沌理论、变分模态分解VMD(variational modal decomposition)、整合移动平均自回归ARIMA(autoregressive integrated moving average)模型和门控循环单元GRU(gated recurr...

ARIMA-GRU Short-term Power Load Forecasting Based on Feature Mining

Aimed at the problems in short-term power load such as strong randomness and low prediction accuracy,a combined prediction method based on the chaos theory,variational modal decomposition(VMD),autoregressive integrated moving average(ARIMA)model and gated recurrent unit(GRU)neural network is proposed.First,the phase space of the power load’s historical data is reconstructed to extract the chaotic characteristics.Second,VMD is applied to decompose the load sequence of each dimension in the phase space into a set of intrinsic mode functions(IMFs)with a good stability.Third,each group of IMFs is reconstructed as a low-frequency sequence and a high-frequency sequence according to the frequency index zero-crossing rate.Finally,the ARIMA and GRU neural network models are used to perform model training and iterative prediction on the low-and high-frequency sequences,respectively,and the predicted result is obtained by combining the predicted value of each sequence.The analysis of an example shows that the proposed method has a higher prediction accuracy compared with other intelligent algorithms.