Abstract: Data mining technology is employed for the feature extraction and pattern identification of about 260 thousand water use data collected in the Monitoring and Control Program of Water Use Units by the Ministry of Water Resources. Using the k-means clustering algorithm based on the Davies-Bouldin index, the dataset is categorized into three features, i.e., feature of water saving status quo as reflected by WSE, feature of expected water saving willingness as reflected by EW, and feature of monthly water saving volatility as reflected by C_v. The WSE value of most water use units concentrates in [0.7, 0.9], EW in [0.8, 1.0], and C_v in [0.1, 0.5]. Furthermore, on the basis of the features extracted above, water use pattern is classified as five groups, namely, balanced expansion pattern, balanced contraction pattern, centralized stable pattern, fluctuated contraction pattern, and fluctuated expansion pattern. Compared with inland areas, the southeast coastal areas produce less volatile monthly water consumption. Among the five water use patterns, fluctuated contraction is the dominant pattern covering most high consumption industries with excessive productivity; fluctuated expansion and balanced expansion mainly distributes in high-tech manufacturing industry and service industry; and centralized stable pattern in agricultural industry. In addition, management suggestions are put forward respecting laws and regulations and monitoring work to offer reference for a more precise and targeted water saving management.

Key words: water saving management, data mining, water use features, water use patterns, DB index