论文部分内容阅读
提出一种新的多种群并行遗传算法 (NMPGA) ,并将其作为多层前馈神经网络(MFNNs)的学习算法 ,从而形成一类新的 MFNN模型——多种群并行进化神经网络(MPENNs)。首先 ,对一给定的网络结构 ,随机产生一初始权重的集合 ,这个集合实际上对应着一组具有相同结构但不同权重的神经网络。然后 ,采用 NMPGA对 MFNNs的权重进行进化。最后 ,性能最好的网络被选作目标问题的解。在 NMPGA算法中 ,作者采用浮点数编码来克服传统二进制编码的精度不足问题 ,并设计了专门的杂交算子和变异算子来增强算法性能。实验结果表明 ,MPENNs能成功解决异或问题、三元奇偶问题及成品烟的感官质量评价问题。
A new multi-population parallel genetic algorithm (NMPGA) is proposed and used as a learning algorithm for multi-layer feedforward neural networks (MFNNs) to form a new MFNN model - MPNs (Multi-population Parallel Evolutionary Neural Network) . First, for a given network structure, a set of initial weights is randomly generated. This set actually corresponds to a group of neural networks with the same structure but different weights. Then, the weight of MFNNs is evolved using NMPGA. Finally, the best performing network was chosen as the solution to the goal problem. In the NMPGA algorithm, the author adopts floating-point encoding to overcome the problem of insufficient precision of traditional binary encoding, and special hybridization operators and mutation operators are designed to enhance the performance of the algorithm. Experimental results show that MPENNs can successfully solve the problem of XOR problems, tripartite parity problems and sensory evaluation of finished cigarettes.