论文部分内容阅读
针对非线性或者非高斯系统辨识问题,该文提出一种混合最小熵准则和分布估计算法的参数辨识方法.该方法将最小误差熵引入系统参数辨识的准则函数,解决了传统辨识准则大多针对高斯系统,或者对先验知识的依赖无法得到满足,或者不具有适应性的问题;在分布估计算法的迭代过程中,加入随机个体作为新种群的一部分,增加了种群多样性,避免早熟收敛.通过对标准测试函数的寻优以及对benchmark 经典非线性系统无噪声和不同噪声情形下的辨识,并与经典算法和已发表较新算法进行对比,结果表明了该算法的优越性.最后,基于现场运行历史数据,将该文算法应用于火电厂协调系统传递函数的参数辨识,显示了该文算法对于热工对象建模的适用性和有效性.“,”A novel hybrid approach combining a minimum entropy criterion and estimation of distribution algorithm (EDA) is proposed to solve the identification problem for nonlinear non-Gaussian systems. A minimum error entropy (MEE) criterion is introduced for system identification since most conventional identification criteria have those shortages:1) providing the system is Gaussian; 2) needing the priori knowledge, which cannot be satisfied in many cases; 3) lacking adaptability. Random individuals are added into offspring generations of EDA to maintain population diversity. Simulations have been designed on some benchmark functions and a nonlinear system to validate the performance of the proposed MEE-EDA. Results show it outperforms some other methods. Its application to the identification for a coordinated control system (CCS) in a power plant demonstrates that the proposed MEE-EDA method provides an efficient and practical way to tackle the identification problem for thermal dynamic processes.