Sunspot Time Series Forecast Based On Wavelet De-noised And BP Neural Networks

It is difficult to build models for the real-world because there are various uncertainties factors. And this is a common problem not only nature but also social sciences, It is not only possess theoretical significance but also application prospects to solve the problem. So there are many scholars from home and abroad research in this. In the past people research on the different prediction models precisions. So they don't consider the influences of the time series noise. People can't get accurately predict by traditional methods. This blocks the further development of time series study. At the same time, it maybe contains noise when collect the time-series data, and the impact of procedures is different from different time-series data, this hides the inherent laws of the time-series data, so traditional forecasting methods can't get satisfied with the results of time-series data. How to effectively build a model with the time series is an urgent problem to be solved. Wavelet transform, neural network is a hot research, however, at home and abroad to combine the study of these theories with observation noise time series prediction is rare.Sunspots are non-linear, highly complex, multi-time scale changes time series, and in the observation and data collection often effected by natural environment. This made observations contain noises. And the sequence itself is a non-linear, highly complex, multi-time scale change. It is obviously not suitable for forecasting the sunspot time series with the linear model. In this paper, we will study the days'change of sunspots. We take advantage of de-noising properties of wavelet analysis and non-linear fitting capacity of neural network, modeling of time series analysis, using wavelet analysis and neural networks combined modeling time-series data. Before construct the network, we remove noise of the time-series data with wavelet de-noising method. And then take full advantage of the BP neural network nonlinear characteristics to modeling, prediction. The experimental results show that this method of prediction accuracy is higher than the ordinary method.