摘要单乘神经元模型是受神经科学中单神经元计算的启发而被构造出来的模型，它常用于函数逼近和时间序列预测，具有诸多优越性。例如，单乘神经元模型结构单一，计算简便，学习效率高，逼近能力强，并且表现出比特定结构的神经网络有更好的性能。86172

而且，具有递归算法结构的非线性滤波算法既可以处理附加噪声，还可以根据新的观测值自我更新模型参数。其中扩展卡尔曼算法是一种典型的非线性滤波算法。本文结合单乘神经元模型与扩展卡尔曼滤波算法进行风速时间序列的在线预测。论文主要完成以下几方面工作： 

（1）介绍时间序列特别是混沌时间序列的相关内容，对现有各种时间序列预测方法的优缺点进行了分析和总结；

（2）单乘神经元模型在时间序列预测中有很多无可比拟的优势，本文具体地介绍了单乘神经元模型的理论，同时给出了它的算法；

（3）本文推导出了适用于非线性系统的扩展卡尔曼滤波算法，给出了其算法公式和滤波流程图；

（4）将扩展卡尔曼滤波算法结合单乘神经元网络应用于时间序列的预测，建立了时间序列在线预测模型。把该模型应用于风速时间序列的预测中，得出了很好的仿真实验结果。

毕业论文关键词：单乘神经元模型；非线性滤波算法；时间序列预测。

Abstract Single multiplicative neuron model is a model which is inspired by single neuron computation in neuron science。 It is commonly used in function approximation and time series prediction and has many advantages。 For example, the single multiplicative neuron model has a simple structure and easy way to calculate a fast speed for learning and effective ability for approximation。 So it performs better than the neural network with specific structure。

Furthermore, a recursive nonlinear filtering algorithm makes the system not only deal with the additional noise, but also update the model parameters according to the new data。 The extended Kalman algorithm is a typical nonlinear filtering algorithm。 In this paper, we will conduct the time series on-line adaptive prediction on the basis of single multiplicative neuron model and extended Kalman filter algorithm。 This paper mainly completes the following work:

(1) Introduce time series, especially the chaotic time series。 Advantages and disadvantages of various time series forecasting methods are analyzed and summarized。

(2) There are many incomparable advantages in the time series prediction of single multiplicative neuron model。 In this paper, the theory of single multiplicative neuron model is specifically introduced and at the same time, the algorithm is given。

(3) The extended Kalman filter algorithm is discussed。 We specifically introduce the formula and Filtering flow chart of it。

(4) The extended Kalman filter algorithm together with the single neuron network are applied to the prediction of time series。 So we could establish a time series forecasting model。 The model is applied to the prediction of wind speed time series, and the final simulation results are very good。

Keywords: Single multiplicative neuron model; nonlinear filtering algorithm; time series prediction。

目  录

第一章 绪 论 1

1。1 时间序列预测的研究背景和国内外研究现状 1

1。2风速时间序列预测的研究意义及其研究现状 3

1。3本论文主要研究内容 5

1。4本章小结 6

第二章 单乘神经元模型 7

2。1 人工神经元模型