基于DRNN的直接多步自适应预测器设计

Design of Direct Multi-step Adaptive Predictor Based on DRNN

介绍了一种基于对角回归神经网络(DRNN)的直接多步自适应预报器的设计方法。由于其有特殊记忆能力组成单元和特殊的组织结构形式,这种预报器可以在线通过对一个动态系统的输入输出样本的学习,自动建立这个动态系统的模型,而无需事先了解该系统的参数。因此,该预报器可以根据系统过去的输入输出样本和当前可以测量到的系统输入,来获得系统未来的输出值。本文首先介绍了这种预报器模型的结构和输出输入的映射关系,然后给出了一个用于系统预测分析的预测器的具体设计过程并给出一些仿真实验结果,介绍了这种预报模型在船舶机舱智能监控系统中的具体应用。

A new direct multi-step adaptive predictor based on the diagonal recurrent neural network is presen- ted. The new predictor does not require deciding any system parameter in advance. Because of the special structure and the retentive units, the system parameters can be identified and modified on line through the learning from a time series of a real dynamic process. So the predictor could obtain the future output value of the analyzed system according to its past output value and some measurable input signals. The structure of the predictor and its map relation are described. Then a prediction module is designed for trend analysis. Some experimental results with its application in the intelligent monitoring system of marine engine room are provided.