进化回归神经网络对时序数据和关联数据的建模能力研究

重点研究进化回归神经网络对时序数据和关联数据的建模能力。针对两个标准问题，采用不同形式的建模数据，比较了前向网络和回归神经网络的建模及预测效果，进一步将进化算法用于不同结构回归神经网络的训练并比较了它们的建模能力。仿真结果表明回归神经网络对时序关联数据有很好的建模和预测能力，相比于前向网络，无需过程时序特点的先验知识，可以采用最简单的建模数据形式。而进化算法相比于常规的梯度下降算法，用于训练不同的回归网络结构通用性好，且训练过程不受局部极小问题的困扰，适当规模的训练过程可以获得性能良好的神经网络模型。

Modeling Capability of Evolving Recurrent Neural Network for Time Series Data and Correlative Data Series

The modeling capability of evolving recurrent neural network for Time Series Data and Correlative Data Series is discussed and compared to that of FNN. Data format for modeling, capabilities of various RNN structures and evolutionary operations for RNN training are also investigated. The simulation results demonstrate high modeling capabilities of RNN with simplest training data format for both of the cases. Furthermore, evolving algorithms, which can escape from local minima and get more suitable RNN models for the proposed problems, are more feasible for training various RNN structures.