The Research On Low Flow Of Time Evolution And Forecast In Xiangjiang River

This article has outlined domestic and foreign to the waterless research primary coverage, and take the runoff material in Xiangtan hydrologic station of Xiangjiang River as the basis, has carried on the characteristic analysis to the low flow, and has carried on the trend analysis using Mann-Kendall non-parameter order related inspection method and the piecewise linearity return method to the low flow of Xiangjiang River. Then has made the frequency analysis to the low flow, finally using support vector machine, the projection pursuit regression , the BP neural network to build model, to forecast the 7 daily average current runoff of Xiangtan station. The primary content as follows:1. This article provides the present domestic and foreign low flow frequency analysis and low flow forecast method study situation, carries on system's analysis summary, and proposed the goal and significance of this paper research.2. Use the glide average method to calculate the smallest 7 daily average runoff, take the1960-2005 year runoff data of Xiangjiang River as representative, carries on the characteristic analysis of the low flow tendency to the Xiangjiang River, and carries on the trend analysis using Mann-Kendall non-parameter order related inspection method and the piecewise linearity return method to the low flow of Xiangjiang River, the result show that the three indicators are of rise trend.3. Studies the low flow probability distribution characteristic of the Xiangjiang River, through probability plot correlation coefficient to select the appropriate theory distribution linear to the Xiangjiang River. According to the results, we can see that the correlation coefficients of P-III distribution, Gumbel distribution, the Log P-III distribution are all above 98%, it is more desirable, and suitable for the low flow frequency calculate of Xiangjiang, result show that the P-III distribution is ideal and convenient.4. Finally, take the Xiangjiang River as the study object, using the 45 year smallest 7 daily average material of Xiangjiang River to establishment low flow prediction model. First, using the previous smallest 40 years year 7 daily average runoff material to training the model, then use the forecast model to forecast the left runoff, then to confirmation the runoff by the actual amount. Through the low flow frequency analysis and the forecast research of the Xiangjiang River, has made certain contribution to Hunan Province's waterless resources investigation and the research. This text selects the projection pursuit regression, the support vector machine, the BP neural network to build model, the result show that the impact of the support vector machine is the best , and the accuracy is the highest.