改进的Miller矩阵编码及其在进化人工神经网络中的应用

Miller等人提出的矩阵编码方法要求神经元的数目预先固定，且基因数目是神经元数目的平方，另外，它是一种二进制二进制编码，使用矩阵表示的网络结构却无法编码权值，从而限制了它的使用，提出了改进的Miller矩阵编码方法，该方法采用实数编码，可大大减小编码的冗余，能用于神经元数目在一定范围内可变的场合，所设计的进化神经网络经实际应用取得了很好的结果，表明该方法是有效的。

Modified Miller-Matrix Encoding Method and lts Application in Evolutionary Artificial Neural Networks

The encoding method proposed by Miller et al. lacks scalability and requires that the number of neurons be fixed in advance. It is a binary-code and uses an adjacency matrix to represent the topology of an ANN but cannot encode weights. All of these restrict its use, so this paper presented a modified Millermatrix encoding scheme, which adopts real-code and can largely reduce the redundance of encoding. This encoding method is used in the proposed evolutionary ANN. The results show that it is effective.