基于BP神经网络的非线性函数逼近仿真研究

Study on the Simulation of Nonlinear Function Approximation Based on BP Network

神经网络的建模方法的主要特点是模型易于实现(辨识模型即为神经网络本身)和对非线性映射的逼近性能良好,为非线性系统黑箱辨识问题提供了一种十分有效的工具。本文在研究BP神经网络MATLAB实现的基础上,提出了一个用于非线性函数逼近的3层BP网络。训练结果表明:预测值与期望值具有较好的一致性,说明BP网络具有良好的预测能力。

The neural network modeling approach has many advantages such as easy to realize （because the identification model is the neural network）, good approximation performance for nonlinear mapping, etc. This modeling approach can also provide a very effective tool for black box identification problem in nonlinear system. This article set up a 3-layers BP network for nonlinear function approximation based on MATLAB implementation. The training results showed that the predicted value and expected value had a good consistency. The simulation results also illustrated that the BP network had well ability for prediction and generalization.