Study on combination forecasting of gas daily load based on the generalized dynamic fuzzy neural network

A method based on elliptic basis function, which is a combination forecasting of gas daily load based on the GD-FNN, is proposed. Prior knowledge of fuzzy neural structure isn´t needed, nor does pre-training. It builds models by online adaptive learning algorithm completely. Nonlinear combination of weights is obtained through the network dynamic learning, which is based on the principle of minimum total error, and is not limited to the nonlinear weights. The recurrent neural network can make the training speedy, the learning algorithm simple, and the relative error fluctuations of the gradient regression neural network stable; besides, less information is needed in the gray forecasting. In addition, it can weaken the randomness of the data, by selecting the three single forecasting methods: GRNN, gray GRNN and gradient GRNN to predict the daily load and take its output as the input of GD-FNN. The system simulation experiments prove that the proposed method is of high efficiency.