Monthly Run Off Forecasting Researcl Based On Wavelet Transform And LSSVM

Water is the material ensure human survival, the size of runoff has very close ties to development and utilization of water, the results of Runoff predictions has important guiding significance for the development of the region’s drought, flood, reservoir optimization scheduling strategy research.Today’s,the water resources increasingly scarce make us more and more understand that the problem of water is very seroious,However, because of its complicated by various factors influenced, research on runoff forecasting has become a hot and difficult research scholars.In this paper,we use Manas River Hongshanzui hydrological station measured runoff data 55 a to make the study. First of all,we used the statistical knowledge to analysis runoff interannual variability and changes in the inner diameter and found that the Manas River runoff interannual variability relatively stable.However,the distribution was uneven during the year and high degree of concentration in time and space, making the mountain breaking floods frequently in summer, in other seasons and plains drought phenomena occur easily, so we use monthly runoff data to predict the runoff.Using RBF kernel function’s LSSVM model to predict monthly runoff time series, due to penalty factor C and kernel parameter determination and combine seriously affect the performance of the model. The parameters optimization use the fruit fly optimization algorithm,and compared it with the particle swarm algorithm, verified that the fruit fly optimization algorithm was more good. In order to change the search range of process of the fruit fly optimization algorithm, proposed variable step search fruit fly optimization algorithm, variable step fruit fly optimization algorithm could improve the optimization of speed and has good predictive results.The combined effects of a variety of different factors make the runoff time series with high non-stationary characteristics, wavelet analysis method is a good time-frequency analysis technique, The non-stationary runoff time series is decomposed into relatively stable coefficient sequence prediction, could weaken the interaction between each frequency component, and fully explore the usefully information contained the raw data.Therefore, coupled the wavelet transform and variable step fruit fly optimization algorithm optimize LSSVM, Constituting the WT-VSFOA-LSSVM model to predict the monthly runoff.The simulation results show that the addition wavelet transform model to predict runoff absolute error is relatively flat and the prediction accuracy is significantly improved,the results has reached hydrological forecasting Grade A standard and able to provide more reliable data to support operational, scientific management and rational allocation of safety reservoir.