材料期刊网

针对目前的板形缺陷识别方法精度不高、识别速度慢的问题,根据Elman神经网络模型可以反映系统动态特性,而且可以逼近任意非线性函数的特点,提出了一种利用改进的遗传算法优化Elman神经网络,使其泛化能力强、学习速度快、识别精度高,并建立板形缺陷模式识别模型的方法.为了验证该方法的识别能力,在隐层节点数与学习次数相同的条件下,分别与遗传算法优化的Elman网络和BP网络模型进行板形识别仿真对比分析.试验结果表明,改进遗传算法优化的Elman神经网络模型对板形缺陷识别精度高于BP网络等模型,并且具有收敛速度快的优点.

In order to overcome the shortcomings of current defects recognition of flatness shape method,such as low recognition accuracy and low recognition speed,an improved Elman neural network by improved genetic algo-rithm (GA)was proposed.Because of the Elman neural network model with the advantages of reflecting system′s dynamic characteristic and the characteristic of approximating any nonlinear function with arbitrary precision,the model based on this optimized algorithm can equip the recognition method with strong generalization ability,swift recognition speed and high recognition accuracy.To verify the method′s recognition ability,the Elman network model optimized by genetic algorithm and BP network model were compared and analyzed respectively under the same condition of the same number of hidden nodes and the same number of learning times.The experimental re-sults show that,the Elman neural network model optimized by improved genetic algorithm is more accurate than BP network model in defects recognition of flatness shape,and has faster convergence speed.