Runoff Analysis Of Three Gorges And Its Forecast

Water resources system and hydro-environment is a nonlinear energy-system with vast, complicated, dynamic characteristics, in which the variance of runoff is taken as its main factor, for its changes dominants the whole system's changing trend , and may greatly affect the resources, environment and regional economy. Due to the impacts caused by many factors such as climate, geography, characteristic of the basins, the changing trend of flow in Three Gorges Region takes on a great deal of uncertainty featuring nonlinear, multi-timescale and chaos. In addition, there still exist many limits that the hydrology process is not clear and the datum which affect the hydrology process are inadequate. Therefore, mathematics physics method has difficulty in clearly describing each hydrology process till now. In the past, researchers mainly studied the evolvement characteristic of the runoff using the traditional method or based on a single hydrology station and then make analysis and forecast, however, they're hard to obtain satisfactory results and also facing many difficulties. To get rid of this problem, new theories and methods are needed to be continuously introduced to the researches on water resource, besides, various methods can be properly combined to research on the whole basin systematically.We sum up the advantages the preceding results, on this basis, modern new theories, such as wavelet transformation, chaos and support vector machine, and their combination with traditional methods are introduced to make a profound study on the long-term variant characteristic of runoff of Three Gorges, such as period, trend, relativity and chaos etc. Based on the analysis above, a new mufti-time-scale hybrid model for runoff series prediction is proposed with the support vector machine and wavelet, and then the runoff interval forecast models and its solving methods are studied by using clustering algorithm and chaotic time series method. The results provide the future annual runoff process for the water resources policy makers, researchers and the public. Generally speaking, we conclude the following primary findings:Through wavelet transform, we made an analysis on the characteristic of period and its changing trend using the natural annual runoff time series in the Three Gorges basin in mufti-time scales. The results demonstrated that the annual runoff time series has a period of about 30 years' in long term variety, a middle-term period of 12～18 years and a short term period of 3~8 years, and has a more and more obvious decreasing tendency on the whole. In addition, the flood and dry periods of annual runoff differ with different scales. Seen from the first main period of each station, it's found that Cuntan station enters flood season after 2006, Wanxian station, Yichang Station enter dry season after 2003 and 2004, respectively.Considering the influence of noise and the disadvantages of traditional noise eliminating technologies erenow, the wavelet theory is applied, which not only de-noises well, but also retains the effective composition in the original signal. Based on the correlation analysis of the natural annual runoff time series and the processed series in the Three Gorges basin, the characteristic of self correlation which change with time are probed. In the long-term change, natural runoff time serial itself have no self correlation in the Three Gorges basin; but after wavelet de-noising, runoff time serial of there have certain characteristic of self correlation, which turns smaller as the time increases.Considering that the variety of the hydropower and water resource system in time domain has a multi-layer structure and characteristic of locality, we make time series decomposition using velvet transforming method in different time scales and analyses their mutual correlation with mutual correlation analysis method and verifies its correctness. The result demonstrated that the mutual correlations of annual runoff time serials in any two stations between Cuntan station, Wanxian station and Yichang station were totally different in different time scales, which manifested a correlation, or negative, or independent, with positive as its main trend with the changing of the dimension. With the increase of the dimension, the mutual correlation strengthens continuously, but weakens or it appears a great negative correlation as the time increases.Based on the reconstruction of the phase space of runoff for the Three Gorges basin, it is found that the maximum Lyapunov exponent of each station's runoff serial are larger than zero, which means that the runoff of each station has chaos characteristic, and it has been further verified by saturated correlation dimension method and Cao method. In addition, it is also found that the saturation correlation dimension turns larger from upper stream to downstream, which is caused because of the joining of branches, the section raining along the river and the regulation of reservoir.In building a hybrid water resources model, there is a trend to combine certain and uncertain analysis methods probably in recent times. We proposed a new hybrid time serial wavelet prediction model ,which combines the Daubechies's wavelet in the multi time scales theory, and the Mallat's algorithm with support vector machine with the traditional support vector machine(SVM) modeling method. Compared to the traditional methods based on support vector machine and chaos model, the hybrid model has a higher prediction precision and can better reflect the changing characteristic of runoff, as well as having advantages of clear concept, integrated structure, easy operated, thus make it a effective approach to study the evolving principle of the runoff time series. The results have shown that the new model is meaningful and feasible.In order to avoid the error caused by several uncertain factors such as embedding dimension, time delay and similar states extracted method, we proposed a chaotic time series method for long-term runoff interval forecasting based on the chaotic characteristic of runoff. First, the phase space of runoff time series is reconstructed and similar states of current phase point are searched by the clustering algorithm, based on which the interval values of the future are determined. Finally, we verified its feasibility and effectiveness by applying it to monthly runoff forecasting.