In a Wireless Sensor Networks (WSN) based fluid pipeline leak monitoring system, numerous sensors are deployed along the pipeline networks. A great amount of measurements are continuously transmitted from the sensor nodes to their corresponding sink nodes. The energy consumed on data transmission dominates the power depletion of a WSN system. To reduce the amount of data transmission and prolong the lifetime of WSN, in this paper, a Combined Dual-Prediction based Data Fusion (CDPDF) method is proposed. Transmissions are only triggered if the measurement is substantially different from the predicted value. Furthermore, unlike existing methods which establish the predictor by merely considering the measurements from a single sensor, the proposed CDPDF learns and updates the predictor by integrating measurements from multiple neighboring sensors, hence the spatial cross-correlation is taken into account and the prediction accuracy is significantly improved. In this paper, an Enhanced Leak Detection and Isolation (EnLDI) method is also proposed in which several important parameters, such as the friction factor and the pressure wave propagation speed, can be online updated, resulting in improvement of the leak localization accuracy. Experimental case studies are conducted. By employing the proposed CDPDF and EnLDI methods in pipeline networks monitoring, the accuracy of leak isolation is significantly increased with reduced data transmission demands. Note to Practitioners—This work delivers a hybrid scheme that combines machine learning based data fusion and transmission, with model-based leak detection and isolation. The work is motivated by the problem of high energy consumption on data transmission and poor leak diagnosis accuracy in WSN based pipeline networks monitoring system. To reduce the energy consumed during frequent transmissions among sensor nodes, in this paper, a machine learning based data fusion method is proposed which can eliminate most of the redundant transmissions. Among the investigated schemes, the Extreme Learning Machine (ELM) based predictor can not only achieve satisfactory prediction accuracy but also has low computational cost, hence it can be easily implemented in most of the embedded micro-controller systems in practice. At the base station of a WSN, in the leak diagnosis phase, traditional model-based methods employ the fixed model parameters which should be adjustable in different pressure and flow conditions etc. In this paper, an online model parameter estimation procedure is introduced and incorporated in the scheme designed to estimate the leak size and location, thus, the leak localization accuracy is significantly improved. Moreover, the algorithmic procedures, mathematical expressions, evaluation process and results are also provided for practical implementation.

中文翻译：

---

基于双预测的数据融合与增强型WSN管道监测系统泄漏检测和隔离方法相结合


在基于无线传感器网络（WSN）的流体管道泄漏监测系统中，沿管道网络部署了大量传感器。大量的测量数据不断地从传感器节点传输到其相应的汇聚节点。数据传输所消耗的能量在无线传感器网络系统的功耗中占主导地位。为了减少数据传输量并延长WSN的寿命，本文提出了一种基于组合双预测的数据融合（CDPDF）方法。仅当测量值与预测值显着不同时才会触发传输。此外，与仅考虑单个传感器的测量值建立预测器的现有方法不同，所提出的 CDPDF 通过集成多个相邻传感器的测量值来学习和更新预测器，因此考虑了空间互相关性，预测精度为显着改善。本文还提出了一种增强型泄漏检测和隔离（EnLDI）方法，该方法可以在线更新摩擦系数和压力波传播速度等几个重要参数，从而提高泄漏定位精度。进行了实验案例研究。通过在管网监测中采用所提出的 CDPDF 和 EnLDI 方法，泄漏隔离的准确性显着提高，同时减少了数据传输需求。从业者须知——这项工作提供了一种混合方案，将基于机器学习的数据融合和传输与基于模型的泄漏检测和隔离相结合。 针对基于WSN的管网监测系统数据传输能耗高、泄漏诊断精度差的问题。为了减少传感器节点之间频繁传输时消耗的能量，本文提出了一种基于机器学习的数据融合方法，可以消除大部分冗余传输。在所研究的方案中，基于极限学习机（ELM）的预测器不仅可以达到令人满意的预测精度，而且计算成本较低，因此可以在实践中轻松地在大多数嵌入式微控制器系统中实现。在无线传感器网络的基站，在泄漏诊断阶段，传统的基于模型的方法采用固定的模型参数，这些参数应该在不同的压力和流量等条件下进行调整。本文介绍并结合了在线模型参数估计过程该方案设计用于估计泄漏尺寸和位置，从而显着提高了泄漏定位精度。此外，还给出了算法流程、数学表达式、评估过程和结果，以供实际实施。