递阶进化算法的小波网络在设备状态预测中的应用

为改善小波网络的学习和逼近性能 ,基于递阶结构染色体提出了递阶进化算法用以实现小波网络的设计和训练。该算法采用包含控制级基因和参数级基因的递阶结构的染色体 ,分别对网络结构和网络参数进行编码 ,并根据编码特点将遗传算法与进化规划结合进行进化操作 ,可实现同时对网络结构与网络参数进行进化设计和学习训练。该算法不仅克服了梯度下降算法中的局部极小和网络训练不收敛问题 ,也使网络结构更优 ,从而提高了网络训练效率和网络的工作性能。就函数逼近问题和水轮机组的状态预测问题进行了事例研究 ,验证了所提出的算法的优越性和可行性

Wavelet network based on hierarchical evolving algorithm for machine condition forecasting

A hierarchical evolving algorithm was proposed to improve the performance of wavelet networks. The algorithm used hierarchical chromosomes to encode the structure and parameters of the wavelet network. The genetic algorithm and evolutionary programming were used to simultaneously design and train the network through evolution. The algorithm overcame the local minimum problem in training and optimized the convergence and the network structure. Experimental results for function approximation and condition forecast for hydroturbines shows the superior performance and potential of the proposed hierarchical evolving algorithm based wavelet networks.