滑动平均序列参数的一种自适应估计

An Adaptive Estimation of Moving Average Model

介绍关于滑动平均序列参数的一种自适应估计方法。这种估计方法最初由美国学者J.A.Barnes提出.其突出特点是计算过程简单而容易实现。但是,原算法的动态跟踪特性不好,本文对此问题的原因进行了分析并据此提出了改进算法。计算表明,改进算法在基本保留原算法其他优点的前提下,动态跟踪特性有了彻底改善。

An adaptive method is discussed in this paper for estimating moving average model. A general method for estimating an ARMA model is not a fresh subject in statistics. But most efficient algorithms are discussed in the background of after events, while on-line processing is required in some special applications, for example, the tentative plan of real time atomic time scale. So it is necessary to develop an adaptive method for estimating the parameters of a model.An algorithm of clock forecast based on second order moving average model was presented by J. A.Barnes in Third International Time Scale Algorithm Symposium (Sept., 1988, Turin, Italy) .Although it is still undetermined wether the second order moving average is a good model for atomic clock, the adaptive estimation used in the algorithm is very distinctive. There is not any complicated computation in the adaptive algorithm such as matrix operation and moment estimation. The main operation in the algorithm is computing the first and second autocorrelation of the residual.The whole estimating process is very simple and easy to realize. But there are some defects in the original algorithm. The main problem is that the tracing feature is not good enough for a dynamic system. The update gain attenuate to zero while time go on so that lost the capacity of servoing the parameters. This problem can be ignored in a constant system but it is not the same case in a variable system. From the point of viewof general meaning of an adaptive estimation,algorithm should adapt the changes of system itself. This problem is discussed in this paper and a new update gain is developed. With the new update gain, the dynamic feature of the algorithm is improved.