In order to improve integrity management evaluation level of pipeline, the fuzzy surfacelet neural network optimized by improved dragonfly algorithm. Firstly, the domestic and foreign related research progresses of pipeline integrity management are summarized. Secondly, the pipeline integrity management evaluation index system is constructed according to management and technology characteristics of pipeline integrity management. Thirdly, the fuzzy surfacelet neural network with five layers is established by combining Surfacelet transfer, wavelet neural network and fuzzy theory. The improved dragonfly algorithm is established by improved population initialization strategy and inertia weight updating strategy. Finally, simulation analysis of pipeline integrity management is carried out based on fuzzy B-spline wavelet neural network optimized by improved particle swarm algorithm (BWNN-IPSA), fuzzy surfacelet neural network optimized by traditional dragonfly algorithm (FSNN-TDA) and fuzzy surface neural network optimized by improved dragonfly algorithm (FSNN-IDA), simulation results show that the fuzzy Surfacelet neural network optimized by improved dragonfly algorithm can achieve convergence after 500 times, it has less convergence times than other evaluation models. The mean square error of the proposed evaluation model ranges from 0.79 to 1.02%, it has less error than other evaluation models. Therefore the proposed fuzzy surface neural network optimized by improved dragonfly algorithm has higher computing precision and efficiency. The proposed evaluation model of pipeline integrity management can improve intelligent level of pipeline management, and ensure the safety and reliability of pipeline system.

为提高管道完整性管理评价水平，采用改进蜻蜓算法优化模糊曲面神经网络。首先总结了国内外管道完整性管理的相关研究进展。其次，根据管道完整性管理的管理和技术特点，构建了管道完整性管理评价指标体系。第三，结合Surfacelet转移、小波神经网络和模糊理论，建立了五层模糊Surfacelet神经网络。通过改进种群初始化策略和惯性权重更新策略建立了改进蜻蜓算法。最后，基于改进粒子群算法优化的模糊B样条小波神经网络(BWNN-IPSA)、传统蜻蜓算法优化的模糊曲面神经网络(FSNN-TDA)和优化的模糊曲面神经网络对管道完整性管理进行仿真分析通过改进蜻蜓算法（FSNN-IDA）仿真结果表明，改进蜻蜓算法优化的模糊Surfacelet神经网络在500次后即可收敛，收敛次数少于其他评价模型。所提出的评价模型的均方误差在0.79%～1.02%之间，误差小于其他评价模型。因此所提出的由改进蜻蜓算法优化的模糊表面神经网络具有更高的计算精度和效率。