The Research Of Wei River Runoff Prediction System Based On Machine Learning

Under the influence of human activity and rapid climate change, runoff transformations show the characteristics of randomness, fuzziness and gray. In the changing environment, the hydrological series will no longer satisfy the consistency and representation requirements.Further more, traditional runoff prediction models are based on specify watershed and specific predictors, and their abilit:y to promote application is poor. Under the backgrounds that environmental changes have become the norm in the water cycle and developing runoff forecasting models that can be response to a changing environment is a trend in the future, using the commonly utlized three kinds of supervised, parameteric machine learning algorithms,namely support vector machine, gradient boosting decision tree and deep neural networks, build the year and month runoff model for Xian Yang station which is located in Wei River. According to hydrology forecast specification, calculated the predicted percent of pass for each model in training and testing set. Finally, based on the general integration of knowledge visualization platform, built the runoff forecast simulation system for Wei River and a runoff prediction example was simulated in the system. Here the conclusions of this paper:(1) All the three models, SVM, GBDT and DNN can well predict the change trend of the measured runoff series and it has important guiding significance for qualitative runoff prediction.(2) For runoff predictions of year, the percent of pass of GBDT is 95% in training set and 66% in testing set. Compared with other two models, GBDT model has highest precision.(3) For monthly runoff prediction, the percent of pass of prediction of GBDT is highest(42%),DNN's percent of pass is lower (25%), while the SVM has lowest percent of pass on training set. On the testing set, SVM has the highest percent (40%), GBDT has lower percent (30)and DNN has lowest percent (0). Comprehensive analysis to know that GBDT is the best model.(4) From the examples of simulation of runoff prediction, we can know that the forecasting system of Wei River has the characteristics of dynamic adapt to different changes and demand,and the runoff prediction models built on this system has good popularization and application ability.