摘要
结合遗传算法与倒传递神经网络进行工业股票指数预测 ,使用 5个输入变量 :周成交额增减幅、周振荡幅度、周涨跌幅、5日EMA波动、DIF波动值 ,并将下周涨跌幅设为输出目标进行训练 ,以取得较理想的预测结果。对于传统上选择适合的神经网络拓扑结构效率较低的问题 ,本文对于遗传算法的引入大大提高了搜索到最优结构的速度。
This essay integrates genetic algorithm and back-propagation neural network to forecast industrial stock index. We use 5 weekly variables as inputs which are trading volume increase, maximum index value changing rate, closing index value changing rate, EMA and DIF value changes, and set the closing index change rate of next week as the output target to train the neural network in order to get a good result. Traditionally, the selection of an optimal neural network topology is not efficient, this essay uses genetic algorithm to assist selecting neural networks and improves the efficiency.
出处
《湖南大学学报（社会科学版）》
2004年第6期59-64,共6页
Journal of Hunan University(Social Sciences)
基金
国家自然科学基金资助项目 (70 2 730 33)
关键词
神经网络
遗传算法
股指预测
neural network
genetic algorithm
stock index forecast
