论文部分内容阅读
一般的神经网络的结构是固定的,在实际应用中容易造成冗余连接和高计算成本。该文采用了协同量子差分进化算法(cooperative quantum differential evolution algo-rithm,CQGADE)以同时优化神经网络的结构和参数,即采用量子遗传算法(quantum genetic algorithm,QGA)来优化神经网络的结构和隐层节点数,采用差分算法来优化神经网络的权值。训练后的神经网络的连接开关能有效删除冗余连接,算法的量子概率幅编码和协同机制可以提高神经网络的学习效率、逼近精度和泛化能力。仿真实验结果表明:用训练后的神经网络预测太阳黑子和蒸汽透平流量具有更好的预测精度和鲁棒性。
The structure of a general neural network is fixed, in practical applications easily lead to redundant connections and high computational costs. In this paper, the cooperative quantum differential evolution algo-rithm (CQGADE) is used to optimize the structure and parameters of the neural network at the same time. The quantum genetic algorithm (QGA) is used to optimize the structure and hidden Layers of nodes, the use of differential algorithm to optimize the weights of neural networks. The trained neural network connection switch can effectively remove redundant connections. The algorithm of quantum probability amplitude coding and coordination can improve the learning efficiency, approximation accuracy and generalization ability of neural network. The simulation results show that the forecasting accuracy and robustness of the predicted sunspots and steam turbines with the trained neural network are better.