subsequently, a scheme of modeling unknown nonlinear systems is presented . in it a feedforward structure of wavelet neural networks is adopted, and with a hybrid error performance index, a novel on-line dynamic gradient algorithm for training neural networks is put forward . as a result, the real system state and the derivative of the state are accurately identified at the same time, thus a more accurate network model is obtained 随后考虑了对未知非线性动态系统进行建模的方案,其中采用了前向的小波神经网络结构,通过选取一种混合的误差性能指标,提出一种新的在线动态梯度算法训练神经网络,使得最终实现对未知系统状态及其导数(系统函数)同时精确辨识,从而得到一个较为精确的网络模型。
this paper discuss a modeling and predicting means for nonlinear systems proceeding from nonlinear systems modeling and predicting theory, whch is based on drnn model . this means overcomes the fact that ar model is used only in linear systems, at the same time it connects itself with approximation theory symbolic statistics and conjugate gradient algorithm, and formulate a system of large watercrafts motion modeling and predicting which is based on drnn model, and simulate it 本论文从非线性系统建模与预报的理论及应用观点出发,系统地阐述了一类适用于非线性系统的建模预报方法??基于drnn模型的建模预报方法,克服了ar模型仅局限于线性的情况,同时结合逼近论、数理统计等知识,运用共轭梯度算法,提出并建立了基于对角回归神经网络的大型舰船运动建模预报系统,并进行了仿真。
firstly the stochastic gradient algorithm based on minimum mutual information ( mmi ) is researched, and this algorithm is simple and stable, but its convergence speed is slow . secondly the natural gradient algorithm based on riemann space is researched . finally easi algorithm, iterative inversion algorithm and some 首先研究了基于最小化互信息的随机梯度算法,该算法简单稳定但收敛较慢,然后研究了基于黎曼空间的自然梯度算法,最后介绍了easi算法、迭代求逆算法以及其余一些算法。
firstly the stochastic gradient algorithm based on minimum mutual information ( mmi ) is researched, and this algorithm is simple and stable, but its convergence speed is slow . secondly the natural gradient algorithm based on riemann space is researched . finally easi algorithm, iterative inversion algorithm and some 首先研究了基于最小化互信息的随机梯度算法,该算法简单稳定但收敛较慢,然后研究了基于黎曼空间的自然梯度算法,最后介绍了easi算法、迭代求逆算法以及其余一些算法。