The Nonparametric Regression Model And Its Application In Electricty Consumption Of The Whole Society

The total electricity consumption is defined as the total power consumption of all electricity in the country field. Power consumption data is one of the indicators of macroeconomic analysis. Macroeconomic development is accompanied by electric power development, so analysing the social consumption is very important in statistical method. To solve the problem of estimation and test theoretical distribution is unknown by non-parametric statistical methods. The statistical model of the non-parametric regression models do not need to assume that the data fit the model form and to make assumptions the parameters of model, applying for a wider field. choosing a more suitable bandwidth and kernel function is important for non-parametric regression model. To explore more effective search for the optimal bandwidth is one of the starting points of this paper, choosing a bandwidth modified instant of the cross validation which sought after many literature and scholars. The main improved idea is:we use the direct insertion method to determine an initial bandwidth, and then determine an interval of optimal bandwidth, which we make fitting mean square error minimum bandwidth by golden section method (0.618or preferred method).This method improves the computational efficiency and applicability. In this paper, we utilize randomized trials produce sample points of conditional heteroskedasticity, and use improved identification bandwidth selection method to experiment, which shows a better test results. The paper also through random experimental to validate the data itself’s distribution which has effect on kernel function selection. and the effect of model by using the Gaussian kernel,the epanechnikov kernel function, and uniform kernel function. While we are fitting data of normal distribution, exponential distribution, uniform distribution. We arrival at these conclusions:Uniform kernel fitting effect is always worse than the other two kernel fitting.; The Gaussian kernel fitting effect is not always poor than epanechnikovfitting effect, while the distribution of the data have effect on the choice of kernel function. In this paper,we establish ARMA(p,q) model and NARCH model to analysis the total electricity consumption and also discussed which is the appropriate model to the social consumption. At last, we establish mathematical model to discuss the relation between them.