Application of a Classifier Based on Data Mining Techniques in Water Supply Operation

Data mining technology is applied to extract the water supply operation rules in this study. Five characteristic attributes—reservoir storage water, operation period number, water demand, runoff, and hydrological year—are chosen as the dataset, and these characteristic attributes are applied to build a mapping relation with the optimal operation mode calculated by dynamic programming (DP). A Levenberg-Marquardt (LM) neural network and a classification and regression tree (CART) are chosen as data mining algorithms to build the LM neural network classifier and CART decision tree classifier, respectively. In order to verify the classification effect of the LM and CART, the two classifiers are applied to the operation mode recognition for the Heiquan reservoir, which is located in the Qinghai Province of China. The accuracies of the two classifiers are 73.6% and 86.9% for the training sample, and their accuracies are 65.8% and 83.3%, respectively, for the test sample, which indicates that the classification result of the CART classifier is better than that of the LM neural network classifier. Thus, the CART classifier is chosen to guide the long-series water supply operation. Compared to the operation result with the other operation scheme, the result shows that the water deficit index of the CART is mostly closest to the DP scheme, which indicates that the CART classifier can guide reservoir water supply operation effectively.