基于数据科学的复杂天然气管网系统状态演化机制分析方法研究

The difficulties for analyzing complicated processes of system state evolutions, caused by local disturbances in natural gas pipeline networks, is heavily threatening the supply reliability of natural gas. The difficulties derive from the characteristics of pipeline networks including complex topology structures, multiple sources of disturbances and delayed responses due to transient behaviors. The traditional way to solve this problem is modeling the pipeline network mathematically by integrating a large number of equations which represent the physical models of the units. And, then, this system of differential equations is solved by numerical algorithms to predict the states of the pipeline network. However, this kind of method based on reductionism concept is incapable to follow the steps of development of smart transformation and reliability-based management in natural gas systems, from the perspectives of accuracy and efficiency. Considering that, the methods of system science and data science are introduced here to propose a new method for analyzing the system dynamics of gas pipeline networks and predicting the state evolutions of based on their historical behaviors. This new method mainly including: (1) the data mining method based on system complexity theory and machine learning methods, for behavior patterns extraction from high-dimension, heterogeneous and time-series data set; (2) the knowledge (association rules) mining method based on Complex Network method and Data Mining method, to find the transition rules between the extracted behavior patterns. The target of the work is to develop an effective analysis and prediction method of state evolution process for complex natural gas pipeline network systems, to overcome one of the algorithm problems in the way of system pre-warning and reliability analysis.

天然气管网系统的复杂拓扑结构和干扰源、以及输送介质强压缩性导致系统水力反馈滞后，致使其运行状态演化过程复杂、局部事件发展轨迹与风险难以预测，威胁管道系统供气可靠性。传统上通过整合单元物理模型建立系统数学模型，对其进行数值求解模拟系统状态演化轨迹。随着管网规模和复杂性不断提升，这种还原论方法在时效性与准确性上已无法满足智能化与可靠性管理要求。本项目引入系统科学与数据科学的方法，提出根据系统历史运行轨迹预测系统状态演化的复杂天然气管网系统动力学特性分析新思路。具体包括：基于系统复杂性理论与机器学习方法的高维异构、非线性、时序性数据的信息提炼方法，提取复杂天然气管网系统的行为模式；提出基于复杂网络理论与数据挖掘方法提炼高维、非线性复杂系统数据关联规则的方法，揭示复杂天然气管网系统行为模式的转化规律，形成快速、准确的天然气管网系统状态演化模拟方法，突破系统事故预警与可靠性分析中的一个算法瓶颈。

天然气管网系统运行状态的准确、高效预测是事故预警与可靠性分析的关键问题之一。天然气管网拓扑结构复杂，且系统状态演化呈现慢瞬变特性，两者相互叠加致使系统状态的时间-空间演化机制高度复杂。本项目针对复杂天然气管网的运行状态预测问题，研究管网运行状态演化规律的智能挖掘与建模方法，目标是提出一种不依赖于水力热力数学模型求解、而是通过系统历史运行轨迹快速、准确预测天然气管网系统状态演化的新方法。现已完成了研究任务和成果指标，主要创新如下：.建立了基于工艺特性与运行知识的管网系统关键动力学参数识别方法，提出了复杂管网系统海量、非线性耦合动力学参数的特征融合模型，实现了复杂管网运行状态演化关键主控因素的准确识别，形成了复杂管网系统演化动力学的相空间构建方法；提出了面向管网高维运行参数的系统状态演化轨迹数据关联提取模型，构建了管网系统状态演化轨迹的行为模式分割与识别模型，提出了基于图论的管网系统行为模式表征与语义信息提取的新方法，实现了管网行为模式的智能感知；建立了天然气管网系统行为模式之间的关联规则与因果关系分析模型，构建了管网系统状态演化规律的智能反演方法，提出了基于复杂网络理论的管网系统状态演化动力学建模方法；明确了管网系统功能转变的基本动力学特征，建立了耦合图卷积模型与递归深度学习的复杂管网运行状态预测方法。项目发展了系统动力学和数据科学的交叉融合方法，为解决复杂天然气管网系统的风险预警、可靠性评价与优化等问题提供了新的方法支持。.项目成果发表论文24篇，其中SCI论文14篇（一区Top论文10篇），EI收录论文3篇；作国内会议邀请报告1次、国际会议报告2次；获省部级科技进步一等奖1项；申请专利4项，获软件著作权2项；培养研究生5名。项目成果在西气东输管网系统、华南管网系统得到转化应用，学术交流活跃。