【摘要】 利用人工神经网络能够逼近任意复杂函数的特性,可对在役油气管道的腐蚀剩余强度进行预测,但其缺点在于人工神经网络的权值和阈值的初始化分配具有随机性且只是一种局部优化算法,收敛过程中容易出现局部极小解。引入遗传算法的全局搜索特性和不依赖于梯度信息特性,对采用Levenberg-Marquardt(L-M)算法的BP神经网络的权值和阈值进行优化,并结合由敏感性分析确定的油气管道失效压力的影响因素,建立GA-BP(L-M)网络预测模型。采用Modified ASME B31G计算出的样本数据训练网络并进行预测。预测结果表明,GA-BP(L-M)网络预测模型可以相对更好地预测油气管道的失效压力,在满足工程需要的前提下,是一种更加科学、准确的预测模型。更多还原

【Abstract】 The remaining strength of the in-service corroded oil and gas pipeline was predicted based on artificial neural network’s ability to approximate complex function.But artificial neural network has drawbacks as follows:the initial distribution of weight and threshold value is a stochastic process and it was the local optimization algorithm,and that the local minimum solution tends to appear in the convergence process.Therefore,the weight and threshold value of BP neural network using L-M algorithm were optimized based on the global search ability and independence of the gradient information of genetic algorithm, and with consideration of the influencing factors of failure pressure of oil and gas pipeline determined by sensitivity analysis, the GA-BP(L-M) network model was built.The network was trained using sample of Modified ASME B31G and predictions were made.The results show that the GA-BP(L-M) network model can better predict failure pressure of oil and gas pipeline, which proves to be a more scientific and accurate model.更多还原