针对如何提高短期电力负荷预测精度的问题,提出基于核主成分分析法(Kernel principal component analysis,KPCA)和改进的回声状态网络(Echo state network,ESN)算法相结合的方法对短期电力负荷进行预测研究。通过卡尔曼滤波(Kalman filt...针对如何提高短期电力负荷预测精度的问题,提出基于核主成分分析法(Kernel principal component analysis,KPCA)和改进的回声状态网络(Echo state network,ESN)算法相结合的方法对短期电力负荷进行预测研究。通过卡尔曼滤波(Kalman filtering,KF)方法训练回声状态网络的输出权值,引入修正因子对卡尔曼滤波的协方差矩阵进行修正,从而实现回声状态网络结构参数的调整,获得理想的网络结构模型。采用Lyapunov理论验证了改进回声状态网络算法的收敛性。采用核主成分分析法对气象因素进行降维处理,获得能够体现数据信息的主元信息。通过UCI(University of California Irvine)数据集仿真对比,验证了该算法相比于ESN、SVM(Support vector machine)、BP(Back propagation)、GA(Genetic algorithm)等算法具有更高的预测精度。在考虑气象因素的前提下,对短期负荷预测进行仿真实验,实验结果显示在正常天气和存在气象突变的情况下,改进的回声状态网络算法较GA-ESN和GA-BP算法有更高的预测精度,验证了该方法的实用性。展开更多
A major issue in radar quantitative precipitation estimation is the contamination of radar echoes by non-meteorological targets such as ground clutter,chaff,clear air echoes etc.In this study,a fuzzy logic algorithm f...A major issue in radar quantitative precipitation estimation is the contamination of radar echoes by non-meteorological targets such as ground clutter,chaff,clear air echoes etc.In this study,a fuzzy logic algorithm for the identification of non-meteorological echoes is developed using optimized membership functions and weights for the dual-polarization radar located at Mount Sobaek.For selected precipitation and non-meteorological events,the characteristics of the precipitation and non-meteorological echo are derived by the probability density functions of five fuzzy parameters as functions of reflectivity values.The membership functions and weights are then determined by these density functions.Finally,the nonmeteorological echoes are identified by combining the membership functions and weights.The performance is qualitatively evaluated by long-term rain accumulation.The detection accuracy of the fuzzy logic algorithm is calculated using the probability of detection（POD）,false alarm rate（FAR）,and clutter–signal ratio（CSR）.In addition,the issues in using filtered dual-polarization data are alleviated.展开更多
文摘针对如何提高短期电力负荷预测精度的问题,提出基于核主成分分析法(Kernel principal component analysis,KPCA)和改进的回声状态网络(Echo state network,ESN)算法相结合的方法对短期电力负荷进行预测研究。通过卡尔曼滤波(Kalman filtering,KF)方法训练回声状态网络的输出权值,引入修正因子对卡尔曼滤波的协方差矩阵进行修正,从而实现回声状态网络结构参数的调整,获得理想的网络结构模型。采用Lyapunov理论验证了改进回声状态网络算法的收敛性。采用核主成分分析法对气象因素进行降维处理,获得能够体现数据信息的主元信息。通过UCI(University of California Irvine)数据集仿真对比,验证了该算法相比于ESN、SVM(Support vector machine)、BP(Back propagation)、GA(Genetic algorithm)等算法具有更高的预测精度。在考虑气象因素的前提下,对短期负荷预测进行仿真实验,实验结果显示在正常天气和存在气象突变的情况下,改进的回声状态网络算法较GA-ESN和GA-BP算法有更高的预测精度,验证了该方法的实用性。
基金supported by a grant(14AWMP-B079364-01) from Water Management Research Program funded by Ministry of Land,Infrastructure and Transport of Korean government
文摘A major issue in radar quantitative precipitation estimation is the contamination of radar echoes by non-meteorological targets such as ground clutter,chaff,clear air echoes etc.In this study,a fuzzy logic algorithm for the identification of non-meteorological echoes is developed using optimized membership functions and weights for the dual-polarization radar located at Mount Sobaek.For selected precipitation and non-meteorological events,the characteristics of the precipitation and non-meteorological echo are derived by the probability density functions of five fuzzy parameters as functions of reflectivity values.The membership functions and weights are then determined by these density functions.Finally,the nonmeteorological echoes are identified by combining the membership functions and weights.The performance is qualitatively evaluated by long-term rain accumulation.The detection accuracy of the fuzzy logic algorithm is calculated using the probability of detection（POD）,false alarm rate（FAR）,and clutter–signal ratio（CSR）.In addition,the issues in using filtered dual-polarization data are alleviated.
