摘要
针对管道焊缝腐蚀问题构建基于改进鲸鱼优化算法(Improved Whale Optimization Algorithm, IWOA)的概率神经网络(Probabilistic Neural Network, PNN)剩余强度预测模型。首先,通过种群初始化、非线性收敛因子和惯性权重因子提高鲸鱼优化算法的寻优速度和精度;然后,利用IWOA算法优化PNN的光滑因子,构建IWOA-PNN预测模型;最后,以水压爆破试验数据为基础,使用MATLAB软件进行仿真试验,并与另外2个模型进行对比分析。结果表明:IWOA-PNN模型的ERMS为0.633 1,EAR为2.19%,R^(2)为0.954 6,均优于PNN和鲸鱼优化算法(Whale Optimization Algorithm, WOA)-PNN模型;IWOA-PNN模型与传统模型相比误差更小,能够更为准确地预测焊缝腐蚀后剩余强度,为管道的维修和更换提供参考。
Pipeline weld failure is one of the main reasons for pipeline failure. However, there is relatively little research on irregular areas of pipelines. This paper analyzes this problem and establishes an Improved Whale Optimization Algorithm(IWOA)-Probabilistic Neural Network(PNN) model to predict the residual strength of pipeline weld corrosion. Based on the probabilistic neural network, the residual strength of the pipeline weld is predicted, and the smoothing factor parameters are adjusted through the IWOA to improve the prediction performance of the model. This paper improves the whale optimization algorithm in three aspects: using chaotic mapping to initialize the population, adjusting the convergence factor to nonlinearity, and adding inertial weighting factors. Through such an improvement, the optimization accuracy and speed of the Whale Optimization Algorithm(WOA) algorithm can be improved, and the accuracy of the model prediction can be improved. Then we collect the data in the literature and adjust it to make it suitable for the residual strength study of pipeline welds. Combined with the adjusted water pressure blasting data for experimental verification, the prediction results are obtained through MATLAB software simulation and compared with the prediction results of the two models of PNN and WOA-PNN. The results show that the Average Relative Error(EAR) of the IWOA-PNN model is 2.19%, the Root Mean Square Error(ERMS) is 0.633 1, and the coefficient of determination(R~2) is 0.954 6. It shows that the model built in the article has smaller prediction errors than the PNN model and WOA-PNN model, and the model fits well. Such results prove that the model constructed in this paper is better than the unimproved model. Compared with it, it is closer to the real data and has good prediction accuracy and fitting effect. Applying this model in actual engineering to analyze and predict the remaining strength of pipeline welds can reduce calculation time and improve work efficiency. On the other hand, this model can provide theoretical support for pipeline repair and replacement.
作者
骆正山
肖雨
王小完
LUO Zheng-shan;XIAO Yu;WANG Xiao-wan(School of Management,Xi'an University of Architecture and Technology,Xi'an 710055,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2023年第2期435-441,共7页
Journal of Safety and Environment
基金
国家自然科学基金项目(41877527)
陕西省社科基金项目(2018S34)。
