新激活函数下前馈型神经网络及其在天气预报中的应用

FORWARD NEURAL NETWORK WITH NEW ACTIVATION FUNCTION AND THE APPLICATION ON THE WEATHER FORECAST

摘要: 本文为提高基于BP算法的人工神经元网络的学习速度，提出新组合激活函数并取得了显着效果的基础上，又应用于天气预报中，并与基于BP算法的神经网络（标准和带动量项）以及多自适应线性单元的神经元网络进行仿真比较，在预报准确率和学习速度方面获得了令人比较满意的结果．本模拟程序在Turbo-Pascal/6.0环境下编制，在IBMPC386和486机器上调试通过并运行．

Abstract: In order to improve the learning rate of neural networks based on backpropagation, this paperprovides new combination activation function and its remarkable results. Then this new activation function isapplied to weather forecast and the results are compared with those of neural networks based on Backpropagation (Standard and Momentum Term)and Madaline neural networks. Simulation results shw that the accuracy and learning rate of weather forecasting both are satisfactory.This program is written in TURBO-PASCAL 6.0 and run on IBM PC/386 and 486.