Watershed Flood Forecasting Based on Cluster Analysis and BP Neural Network

• Published in: Computer Supported Cooperative Work and Social Computing Vol. 917; pp. 498 - 506
• Main Authors: Wang, Wangsong, Tang, Yan
• Format: Book Chapter
• Language: English
• Published: Singapore Springer 2019; Springer Singapore
• Series: Communications in Computer and Information Science
• Subjects: BP neural network; Flood prediction; Fuzzy C-means clustering; Principal component analysis; Pseudo-F statistic
• ISBN: 9789811330438; 9811330433
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-981-13-3044-5_37

This paper makes an improvement on the traditional flood forecasting method. There are two shortcomings in the conventional prediction process. First, it is not rigorous for clustering by selecting fixed features. Many factors affect the flood, so this leads the loss of useful information and inaccurate clustering results. Second, number of clusters is also fixed, the optimal clustering under multiple clusters is not studied. For the above problems, we propose a new approach by increasing the impact of flood properties before the fuzzy C-means clustering and improving the PCA analysis through leveraging the KMO test and Bartlett test to determine the advantages and disadvantages of the data set to extract, we utilize the gravel map to determine the number of extraction of the main components. When using our improved approach for flood prediction, it can effectively filter out useless and redundant information, and reduce the dimension of high-dimensional data. Then based on the processed data, we use pseudo-F statistics to discover optimal quantity of clusters, and use BP neural network to classify outputs online. After the improvement, we find that the system can achieve higher accuracy for the classification of floods and overcome the shortcomings of the traditional flood forecasting method.