基于IGMM-Copula的入库径流过程预报误差随机模拟模型

为揭示入库径流过程预报误差的统计特征及其变化规律,进而为水电站水库优化调度提供更为准确的输入,基于AIC与BIC准则选取最优高斯混合数,同时引入K-means++算法确定高斯混合模型(Gaussian Mixture Model,GMM)的初始参数值,对GMM-Copula模型中的GMM部分进行了改进,建立了基于IGMM-Copula的入库径流过程预报误差随机模拟模型,该模型不仅在单一预见时刻径流预报误差的量化估计上更具优势,而且能通过建立误差的多维联合分布函数实现对误差序列的随机模拟。以锦屏一级水电站水库为例,应用IGMM-Copula模型对预见时刻为6 h、12 h、18 h、24 h的径流预报误差进行随机模拟。结果表明,IGMM-Copula所得拟合曲线的图形效果及适用性检验结果均优于GMM-Copula模型,且其模拟预报误差的统计参数更贴近于实测预报误差,验证了其合理性与可行性,为入库径流过程预报误差的估计与模拟提供了一种更为精确有效的方法。

Stochastic simulation model of forecast errors in the process of reservoir runoff based on IGMM-Copula

In order to reveal the statistical characteristics and changing laws of forecast errors in the process of inflow runoff,so as to provide more accurate input conditions for the optimal operation of hydropower stations and reservoirs. The optimal Gaussian mixture number is selected based on the AIC and BIC criteria, and the K-means++ algorithm in data mining is introduced to determine the initial parameter values of the Gaussian mixture model. The GMM part of the GMM-Copula model is improved, based on which the stochastic simulation model of the forecast error in the reservoir runoff process is established. The model not only has more advantageous in the quantitative estimation of the runoff forecast error at a single foreseeable moment,but also can realize the stochastic simulation of the error sequence by establishing a multidimensional distribution function of the error. The results applied to the reservoir of Jinping I Hydropower Station show that the model in this paper fits the runoff forecast errors at 6h, 12h, 18h, 24h foreseeing moments, and the graphical effect and applicability test results of the fitted curve are better than those of GMM -Copula model,and the statistical parameters of the simulation error of the IGMM-Copula model are closer to the actual error, which verifies the rationality and feasibility of the model, and provides a more accurate and effective method for the estimation and simulation of the forecast error of the reservoir runoff process method.