Service price forecasting of urban charging infrastructure by using deep stacked CNN-BiGRU network

In order to improve the accuracy of short-term time series price forecasting, a hybrid model convolutional neural network (CNN) coupled with bi-directional gated recurrent unit (Bi-GRU) and attention mechanism is proposed. The architecture first uses CNN extract valid information from data construct feature vectors, then delivers vectors into Bi-GRU for training, finally self-attentive fully exploit input predict load values. Relative single use existing datasets, we deploy complex urban environment evaluate trained model. conclusions show that proposed has outstanding advantages in terms both predictive metrics performance.