基于PSO的神经网络优化证券投资组合方法研究

The Portfolio Method Optimization of Neural Network Based on PSO

针对传统人工神经网络中的BP(back propagation)神经网络自身局限以及其迭代次数多、收敛精度不高和泛化性差等缺点,提出了一种基于粒子群(particle swarm optimizer,PSO)算法的BP神经网络优化证券投资组合方法。在BP神经网络优化方法中,采用PSO算法替代了BP神经网络的梯度下降法,得到最优解,从而对BP神经网络模型进行优化。将该方法应用于证券投资组合的优化中,实验结果证明:该优化方法优于传统的BP神经网络优化方法。

In response to the shortcomings in traditional artificial neural network,such as excessive iterations,low convergence accuracy and poor generalization of BP（Back Propagation）neural network,an optimized portfolio approach of BP neural network based on particle swarm algorithm is proposed.According to the BP neural network optimization method,the gradient descent method is replaced by the PSO algorithm,which obtains the optimal solution so as to optimize the BP neural network model.Through the application of the proposed method into the optimization of the portfolio,this study shows that the method is obviously superior to the traditional BP neural network optimization.