Time Series Forecasting Based On Network

A time series is a sequence of data points, typically consisting of successive measurements made over a time interval. Time series forecasting is the use of a model to predict future values based on previously observed values. Time series forecasting is applied in the economic intervention before the Second World War. During the Second World War and after the Second World War, time series forecasting is widely used in the military science, space science, industrial automation and weather forecasting. Time series forecasting includes collecting and identifying the historical materials of some social phenomenon and arranging them in series; analyzing time series, seeking the change law of social phenomenon according to the time, and obtaining the certain patterns; forecasting the future situation of the social phenomenon based on these patterns.Time series forecasting has developed for a long time. In the beginning, time series forecasting is based on the description. In the 17th century, with the development of random variable and statistical mathematics, time series forecasting steps into the era of statistical analysis. Then, utilizing the characteristics in frequency domain in the time series, frequency domain analysis and spectrum analysis of time series receive much progress. However, the main development of time series forecasting focuses on the time domain analysis. Especially, linear time series forecasting methods have made great development which autoregressive and moving average methods are on behalf of.Compared with linear time series research, nonlinear time series analysis and forecasting are still in their infancy. Recently, an active research issue is applying complex network theory into the time series analysis. From the perspective of complex network, develop a method which can map time series to complex network and so that, we can find out the structure characteristics of time series and deeply understand the structure and dynamics mechanism of complex network. Inspired by this idea, we hope to construct a new time series forecasting method which can contain physical meaning and preserve time information. In order to do that, we combine time series and network, and establish a new time series forecasting model. Inspired by some pioneering achievements including the Visibility Graph and Link Prediction, we use Visibility Graph to convert time series into network, and then based on the Link Prediction find out the relationship between each node in the network. Finally, network is converted into time series according to the visibility criterion to measure the value of forecasting datum. Different from some other common forecasting methods, observed data do not present, historical information. Instead, they are looked as independent individuals. Thus, the proposed model can use the relationship information to forecast.The experiments show that with a small amount of data the proposed method can indeed achieve good results. When there is only a small amount of data and take the one-step forward forecasting, the proposed method achieves better performance compared with ARIMA model, exponential smoothing model, and neural network prediction model. Finally, this propose model are applied in the Taiwan stock Exchange Capitalization Weighted Stock Index and also achieves good forecasting performance. Hence, these experiments demonstrate the proposed model is effective...