基于LM遗传神经网络的短期股价预测

摘要

随着人工智能的不断发展,BP神经网络作为其中一种重要的技术,被广泛应用在股票预测领域。 BP神经网络有很强的非线性逼近能力、自学习自适应等特性,故非常适合解决股价预测中的一些复杂问题。但其在实际的应用过程中还存在一些问题导致其不能很好地进行预测,如网络收敛速度比较慢和容易产生局部最优值等缺点。针对BP神经网络自身存在的这些不足,提出了一种改进的BP神经网络算法。就是通过LM算法改进BP神经网络里的梯度下降法并用遗传算法优化网络参数,即网络的初始权值和阈值,从而提高了网络的收敛速度和搜索全局最优值的能力。用改进后的网络对股票短期价格进行仿真测试,结果表明,改进后的BP神经网络模型有着更快的收敛速度和更高的精确性。%With the development of artificial intelligence,BP neural network,as one of the important technology,is widely used in stock prediction. The neural network,which has the capabilities of non-linear approach,self-learning and self-adaption,is very suitable for sol-ving some complex problems in the stock market. But there are many problems in practical applications result in its poor prediction,such as low convergence speed and local minimum. In order to deal with the defects,an improved BP neural network is proposed by using Lev-enberg-Marquardt ( LM) algorithm to improve the gradient descent in BP neural network and Genetic Algorithm ( GA) to optimize the network’s initial weights and thresholds. It enhances the convergence speed of the algorithm and the ability to search the global optimiza-tion. The model is simulated on short-term stock price prediction and the results indicate that the improved BP model has high conver-gence speed and accuracy.