遗传神经网混合编码方式的研究

Study on hybrid code method of genetic neural network

对遗传神经网中符号编码方式造成的初始种群分布不均的缺点以及二进制编码方式包含的节点信息不完全的不足进行了分析,并综合二者利弊提出了一种改进的以二进制编码方式为基础的混合编码方式,同时针对这种混合编码方式设计了一套专门的遗传操作算子,克服了两种单一编码方式的缺点,有效地提高了遗传神经网的收敛速度。利用所提出的方法对传统的5.bit Parity基准问题以及沙尘暴天气预测的实际问题进行了仿真计算,比较了编码方式改变前后的收敛速度及分类准确性,仿真结果验证了这种改进的有效性。

To deal with the deficiency in symbol code and binary code in genetic neural network, a new hybrid code method is proposed on the basis of binary code. A set of special genetic operators is also presented for the hybrid code. The presented method can increase the convergence rate of genetic neural network efficiently. The traditional 5-bit parity benchmark and the problem of forecasting dust storm are used to test the new method. The test results show the validity of the improved method.