关键词: 工业取用水, 数据挖掘, 小波变换模极大值, 数据重构

Abstract: Improving the quality of industrial water monitoring data is an important part of the national water resources monitoring capacity building, and the singular value problem has become a key shortcoming affecting the quality of monitoring data. Based on the analysis of the main types of singular values of industrial water use monitoring data, the industrial water use monitoring data from the national water resources management system database is selected as the study sample, and the timefrequency variation characteristic of industrial water monitoring data is extracted by wavelet transform modulus maxima model and its residual sequence is fixed by Fourier function. Thus the singular value of the monitoring data is mined using the relative error control method. Moreover, the least squares support vector machine model with chaotic particle swarm optimization is used to reconstruct the singular value data. Results show that the wavelet transform modulus maxima model can better extract the timefrequency variation characteristics of the industrial water use monitoring data sequence, but at the same time it is easy to cause the information loss of the monitoring data. However, the method of residual correction using wavelet transform by using the Fourier function can further improve the characteristic extraction effect of the water monitoring data sequence, and the relative error control can realize the efficient mining of the singular value of the monitoring data based on the wavelet transform modulus maximum characteristic sequence. For the singular value reconstruction and filling, the least squares support vector machine model of chaotic particle swarm optimization has better applicability, and its reconstruction accuracy is better than traditional statistical methods such as polynomial curve fitting and ordinary least squares support vector machine model. The singular value mining and reconstruction strategy of the above-mentioned industrial water use monitoring data provides important technical and methodological support for the advancement of national water resources monitoring capacity building.

Key words: Industrial Water, Data Mining, Wavelet Transform Modulus Maxima, Data Reconstruction