取水监测数据的奇异值识别与修正策略

提高水资源监测数据的真实性与完备性是国家水资源监控能力建设的重要内容.本文基于当前国家水资源监控数据的实际统计状况,提出采用小波变换模极大值的方法实现对取用水监测数据的降噪和奇异值的挖掘,并将辨识出的奇异值进行剔除处理后的监测数据序列作为粒子群-最小二乘支持向量机模型的训练样本,进而根据拟合函数对奇异值进行修正的策略.通过对重点取用水户的取用水监测数据进行实证研究结果发现,利用小波变换模极大值可较大限度地保留取用水监测数据的原始信息,并实现对其中变动幅度偏大数据的分离,可有效降噪并观测取用水监测数据的内在变化规律;同时借助相对误差可进一步挖掘监测数据中存在的奇异值,且辨识效果要好于传统统计方法;而粒子群-最小二乘支持向量机模型对取用水监测数据的样本拟合要比普通最小二乘支持向量机、曲线拟合等方法更为有效,运用该方法修正的取用水监测数据奇异值更加符合实际取用水需求的特点.

A strategy of singular value identification and correction for water monitoring data

The improvement of water resources monitoring data quality is an important content of the national water resources monitoring capacity building project. Hence, according to the actual statistical situation of national water resources monitoring data, the method of wavelet transform modulus maxima was applied to the noise reduction of water monitoring data and its singular value identification, and then the singular value was removed so that a new time-series monitoring data sequence could be corrected. This data sequence was used as the training samples of the least squares support vector machine model optimized by article swarm optimization (PSO-LSSVM), and singular value would be corrected by the fitting function of PSO-LSSVM model. All of above methods were tested through an empirical case of water monitoring data. Results showed that the original information of water monitoring data could be kept as much as possible using the method of wavelet transform modulus maxima, because this method improved the separation of high frequency and low frequency information, so it could reduce noise effectively and observe the inherent changes in water monitoring data. Meanwhile, the singular values were excavated in water resources monitoring data based on the method of wavelet transform modulus maxima, and also its application effect was better than traditional statistical method. The sample fitting accuracy of PSO-LSSVM model was higher than LSSVM and curve fitting, so the singular value was reconstructed by PSO-LSSVM model, and these reconstructed data were consistent with the objective law of actual water demand.