Water resource accounting constitutes a fundamental approach for implementing sophisticated management of basin water resources. The quality of water plays a pivotal role in determining the liabilities associated with these resources. Evaluating the quality of water facilitates the computation of water resource liabilities during the accounting process. Traditional accounting methods rely on manual sorting and data analysis, which necessitate significant human effort. In order to address this issue, we leverage the remarkable feature extraction capabilities of convolutional operations to construct neural networks. Moreover, we introduce the self-attention mechanism module to propose an unsupervised deep clustering method. This method offers assistance in accounting tasks by automatically classifying the debt levels of water resources in distinct regions, thereby facilitating comprehensive water resource accounting. The methodology presented in this article underwent verification using three datasets: the United States Postal Service (USPS), Heterogeneity Human Activity Recognition (HHAR), and Association for Computing Machinery (ACM). The evaluation of Accuracy rate (ACC), Normalized Mutual Information (NMI), and Adjusted Rand Index (ARI) metrics yielded favorable results, surpassing those of K-means clustering, hierarchical clustering, and Density-based constraint extension (DCE). Specifically, the mean values of the evaluation metrics across the three datasets were 0.8474, 0.7582, and 0.7295, respectively.
Read full abstract