遗传算子对径流预报神经网络模型的影响

采用均匀设计来安排遗传算子组合进行数值实验,研究了遗传算子对神经网络径流预报精度的影响。研究发现,输入模式对最终种群中个体的分布影响不明显,不同算子组合对其影响则明显得多。通过回归分析发现,采用不同算子组合优化神经网络初始权重径流预报精度差别较大,对未归一网络的优化效果较归一网络好,同时采用数据归一输入模式与遗传算法优化神经网络初始权重未产生叠加效果。

Genetic Operators Affect on ANN of Runoff Forecast

Genetic algorithms (GA) can effect the artificial neural network(ANN) runoff forecast accuracy by optimizing the ANN initial weights and biases. In this paper, the numerical experiment was designed by the uniform design to compose the difference in runoff forecast accuracy between various combinations of genetic operator. The results show that data-in model is little effect to the individual distributing in final population while the different genetic operator combination is much more effect. Regression estimation indicates the genetic operator that effect main is different at the two data-in model; there is great difference to runoff forecast accuracy among various combinations of genetic operators to optimize the ANN the initial weights and biases, and the optimizing result from ANN which no scale the data to\0,1\] is better than that from ANN scaling, there is no effect superpose to runoff forecast accuracy between scaling data-in and optimizing ANN initial weights and biases by GA.