【摘要】 为了保障输气站管道系统稳定运行，针对管道应力场难以实时反演的问题，提出了一种融合BP神经网络代理模型与改进粒子群优化算法的应力场反演方法。首先，根据有限元模型使用BP神经网络建立了载荷-应变代理模型，代理模型与有限元计算的平均相对误差为0.010 4%,大幅减少了计算时间。之后，提出了改进粒子群优化算法，通过引入传感器筛选策略降低了反演载荷和应变误差，减小了传感器受环境影响产生的误差。实验表明：该方法在3次加载实验中效果良好，反演载荷平均误差为6.64%,大于200με传感器平均误差为6.15%,应变预测平均误差为4.04%,为管道系统数字孪生模型的实时监测与安全评估提供了高效技术方案。更多还原

【Abstract】 To ensure stable operation of gas transmission station pipeline systems and address the difficulties in real-time inversion of pipeline stress fields, this paper proposes a stress field inversion method that integrates a BP neural network surrogate model with an improved particle swarm optimization algorithm. First, a load-strain surrogate model is built using a BP neural network based on finite element modeling, achieving a mean relative error of 0.010 4% between the surrogate model and finite element calculations while markedly reducing computational time. Then, an improved particle swarm optimization algorithm is developed by incorporating a sensor screening strategy to mitigate errors in inverted loads and strains, effectively reducing the error caused by environmental impacts on sensors. Experimental results demonstrate the method achieves superb performances in three loading tests: the mean error of inverted loads reaches 6.64%, sensors with strains exceeding 200 με exhibit a mean error of 6.15%, and the average strain prediction error remains at 4.04%. This paper may provide some insights into real-time monitoring and safety assessment of digital twin models in pipeline systems.更多还原