燃气管网及负荷风险联动模型数据
收藏资源简介：
在燃气管网运维中，往往需要通过多套系统进行监控，不同系统之间数据是隔离的，为了明确设备间的影响先后关系，及风险的衍生关系，以风险关联度模型和时空模型算法为核心，利用机器学习算法、多通道卷积神经网络、STAR模型，深度融合设备特征，进行风险联想，识别设备拓扑结构，帮助研究者理解时间和空间两个维度上的动态变化规律，揭示不同时间点或不同地点之间的相互作用和依赖关系。建立了一套完整的设备拓扑关系图，实现多层次全景化重点场所安全防控。通过获取管网设备信息，包括设备编号、设备类型、地理坐标、上报时间，近15小时出口压力、近15小时进口压力数据，基于格兰杰因果关系矩阵，计算设备在时间及地域上的关系。确定影响度相关性较高的相关设备及影响度，即如果一个设备发生异常，从时间和空间维度确定最可能影响到的其他设备。 运用小波变换，采用设备近15小时出口压力、入口压力的小时时序数据，关系矩阵、是否异常数据，输出影响设备ID；通过变换函数峰值位置时间序列相对延迟时间，计算时间预估值，输出预计影响时间，使隐患的问题可以被提前发现及处理。
In the operation and maintenance of gas pipeline networks, monitoring is often conducted via multiple independent systems with isolated data between each other. To clarify the sequential impact relationships among devices and the derivation pathways of risks, this work takes the risk correlation model and spatiotemporal model algorithm as the core framework, leveraging machine learning algorithms, multi-channel convolutional neural networks, and the STAR model to deeply fuse device features, conduct risk association analysis, and identify device topology. It helps researchers understand the dynamic change rules in both temporal and spatial dimensions, and reveals the interactions and dependencies between different time points or geographic locations. A complete set of device topology diagrams has been established to achieve multi-level, panoramic safety prevention and control for key locations. By collecting pipe network device information including device ID, device type, geographic coordinates, report time, 15-hour historical outlet pressure, and 15-hour historical inlet pressure data, the temporal and geographic relationships between devices are calculated based on the Granger causality matrix. Highly correlated devices with significant impact degrees and their corresponding impact levels are determined: when one device malfunctions, the other devices most likely to be affected can be identified from the temporal and spatial dimensions. Wavelet transform is applied, using the hourly time-series data of the device's outlet and inlet pressures over the past 15 hours, the correlation matrix, and abnormality status data to output the IDs of affected devices. By calculating the relative delay time of the time series corresponding to the peak positions of the transformation function, the estimated impact time is output, enabling hidden dangers to be detected and resolved in advance.
