Water supply forecasting based on developed LS-SVM

The water supply forecasting is of great use both for the decision-making and management of water plan, and for the security of social basic lives. But due to the restriction on the getable data and the complexity of water flow, the water supply forecasting has indeed formed a typical nonlinear regression problem, which is still unsolved by traditional methods. To solve above problem, a novel forecasting model based on Developed LS-SVM for the water supply forecasting is presented, where two algorithms are integrated with each other 1) Least Squares Support Vector Machines (LS-SVM) is the basic algorithm with special adaptability and advantage in nonlinear higher-dimensional regression problems with small samples; 2) Principal Component Analysis(PCA) is a well-known tool to extract linear attributes. In the model, first the attributes of water supply forecasting are reconstructed by PCA in order to overcome the degradation of the latent noise and redundancy in LS-SVM inputs. Second, the mined attributes with better information are fed to LS-SVM for water supply regression. In this way, the accuracy of LS-SVM is fatherly enhanced by combining with PCA. And then the performance of the water supply forecasting is improved. Simulation result shows that the proposed method may increase the accuracy of water supply forecasting.