基于多变量时序数据的窨井燃气泄漏预警系统设计

Design of an Early Warning System for Gas Leakage in Manholes Based on Multivariate Time-Series Data

摘要:

针对窨井内燃气泄漏监测问题,单变量阈值法存在的误判率高、响应延迟大等问题,而且单变量的阈值判别未考虑环境因素,更易漏判误判；因此提出了基于多变量时序数据的窨井燃气泄露智能监测与预警系统；利用双向长短期记忆网络(Bi-LSTM)双向时序特征捕捉能力,通过对燃气泄漏相关环境参数的研究,设计了系统的总体架构,通过多个数据采集终端节点定时上传平台数据,可监测到区域内的窨井中燃气环境信息；预警的方法是每个节点采集到的多个传感器的变量时序数据构建数据集,训练可同时输入多变量时序数据的Bi-LSTM网络模型,预测燃气泄漏,实现平台自动预警；经实验通过部署多个终端节点采集甲烷浓度、温度、气压等环境参数,构建多维度时序数据集,训练可多变量输入的网络模型,模型在独立数据集上训练准确率达98.3%,对比单变量阈值方法预警时间提前了48.5分钟(实测均值)；经实际应用可满足工程上的应用,系统具备高环境适应性,可规模化部署于城市窨井场景,较早地发出燃气泄漏预警。

Abstract:

To address the challenges of gas leakage monitoring in underground utility tunnels, where traditional single-variable threshold-based methods suffer from high misjudgment rates, significant response delays, and a lack of environmental factor integration (leading to frequent false positives and false negatives), this study proposes an intelligent gas leakage monitoring and early-warning system based on multivariate time-series data. Leveraging the bidirectional temporal feature extraction capability of Bidirectional Long Short-Term Memory (Bi-LSTM) networks, the system architecture is designed to incorporate critical environmental parameters associated with gas leakage. Multiple data acquisition terminal nodes periodically upload methane concentration, temperature, air pressure, and other environmental data to a centralized platform, enabling real-time monitoring of gas-related conditions across regional utility tunnels. For early-warning functionality, multivariate time-series datasets are constructed from sensor data collected by each node, and a Bi-LSTM model capable of processing multivariate inputs is trained to predict gas leakage events, triggering automated platform alerts. Experimental results demonstrate that the model achieves a training accuracy of 98.3% on an independent dataset, outperforming single-variable threshold methods by providing warnings 48.5 minutes earlier on average (empirical mean). Practical implementations confirm the system’s superior environmental adaptability and scalability for deployment in urban utility tunnel scenarios, significantly reducing leakage risks through timely and accurate hazard detection.