Neural network model of phase space and its application in hydrologic medium-and-long-term prediction

The hydrologic system is an open and complicated system, and also is a dynamic non-linear multiplexed systems, the space-time change of hydrologic factors has high non-linear characteristics. On the one hand, it is an important process which is with the natural factors such as weather, climate and topography interactions and inter-dependent in hydrologic cycle. On the other hand, it is affected differently by human product activities such as the basin development degree, local cultural level and so on, thus has formed complicated evolution rule of hydrologic system. f problems of non-linear hydrologic changes on essence are studied form the linear angle or approximately, there must be its inevitably limitation. To forecast the future evolution behavior of hydrologic dynamic system, we must know the topological structure or change rule of trace of system attractor. And phase space is the most ideal and the most direct-viewing space to describe the topological structure of system attractor. To construct the dynamical forecast model of time series, we must reconstruct the phase space of hydrologic dynamical system. The method of reconstructing the phase space can use the time series to reconstruct a low-dimension space whose dynamical system attractor does not change. According to theory of chaotic phase space, we established the single-point model, multi-point model, lineal model, three-parameter D(m,τ, k) model of local similarity model. The philosophy and algorithm of four kinds of model of chaotic phase space are also introduced, then its applicable for hydrology is discussed. This paper makes a prediction study about the month series of the Baishan reservoir of the second Songhua River using the above four kinds of models. It is reasonable and superior to use this model in medium-and-long-term hydrologic prediction. The application of the model in the long term runoff prediction of Baishan reservoir is shown, the result of calculation shows that the models are highly effective and is worthy of popularization and application. The research in this paper shows that applying the theory of chaotic phase space in the medium-and-long-term prediction of runoff system uses much more information of the time series than traditional methods. It is effective to reveal the non-linear structure of the hydrologic dynamical system, and it is a new method different from traditional definite method and random method. We should point out that it is significant for raising flood prediction precision to further explore the prediction method of phase space as well as easy methods.