长期预报的相空间相似法

通过对分数维D的计算，我们用时间序列重建了一个d=INT(l+D)维的相空间。而后，将统计预报的相似理论用于相空间的相点，并假设最近邻相点经演化时间T后模相等。根据这一假设，做了一些长期预报试验，其距平符号报准率一般为60%—70%，相对误差一般不大于8%。

An Analog Method of Phase Space for Long-range Forecast

By way of the calculation of fractal dimension D, a phase space with d-dimensions (d=INT (l+D)) is established using a time series. Then, the analog theory of statistical forecast is applied to the phase points of phase space, and it is hypothesized that the mode of state at the nearest phase points after evolution time T is equal to each other. A lot of trial studies based on the hypothesis show that the forecast accuracy of anomaly sign can reach 60%—70%, and the relative error is generally less than 8%