Prediction on Ecological Water Demand Based on Support Vector Machine

This paper introduces a model which combine support vector machine with genetic algorithm to predict the ecological water demand. With the sharply increasing conflict of supply and demand of water resources, the ecological water demand volume is becoming scarce. The prediction of ecological water demand is an important part to the water resource programming and management. Yet the scarce samples and the self limitation of the conventional forecast method make the precision low. The support vector regression machine (SVRM) is based on statistics learning theory with the rule of the structure risk minimum. It has some merits, such as dealing with the data of small sample, the high dimension, the global optimization and the excellent generalization ability. As far as the problem of the memory which the accessing kernel matrix increases with the number of samples is concerned, solving the Lagrange multipliers (the coefficient of the samples) is the difficult. The paper adopts the common optimal method---genetic algorithm (GA) to solve the sample coefficients. Compared with the traditional models of urban water demand forecast, GA-SVRM is based on the stable math theory, has the high precision forecast, better applicability, general value in the complex ecological water demand prediction.