稀疏多元传感器网络跟踪用电设备状态的序列解码与部署优化关键问题研究

Tracking and breaking down the energy consumptions of massive electrical appliances is the basic problem for building energy saving. Because currently the electrical appliances are massive in buildings, which are broadly distributed and change states dynamically, it is general very challenging to tracking their states. Current fidelity monitoring approaches deploy dense power meter networks (current or power meter) to collect the power consumption data, however, such solutions generally form a large scale metering network, which suffers high deployment and maintenance costs, high data collection costs, and the difficulties in system establishment. Another method which conducts high-frequency sampling at the main power entrance to disaggregate the on/off switching states of the electrical appliances by processing the transient signal and pattern recognition, which is called Non-instrusive Load Monitoring. But such approaches require expensive device and the disaggregation method is generally not scalable because ambiguities will be general when many appliances have similar on/off switching patterns. ..To solve these difficulties, this project exploits the temporal sparseness character of the ON/OFF switching events of the massive electrical appliances, providing a feasible solution to use sparse, low-cost sensor network to track the real-time states of the massive electrical appliances. We will study on the compressive observation model of using sparse multi-model sensor network to track the on/off states of the massive electrical appliances; We propose the sequential observation model and the fast sequential decoding algorithm based on a Conditional Hidden Semi-Markov Model (CHSMM) to decode the state sequence of the appliances; we study the deployment optimization problem for reducing the deployment cost of the sparse multi-mode sensor network, and the experimental system of the sparse multi-mode sensor network to track the real-time states of the massive electrical appliances. Based on these studies, we hope to develop theory and key technologies for using sparse, multi-mode sensor network to track the states of the massive electrical appliances comprehensively, accurately and in real-time.

全面准确的感知建筑内大量用电设备的实时耗电状态是建筑节能的基本问题。由于建筑中用电设备数量大、分布广、耗电状态动态变化，现有密集部署传感器网络（电流或功率）监测用电设备耗电状态的方法，面临网络规模大、部署运营成本高、数据采集困难、系统难以实施等难题，而在入户总线上进行高频采样与信号处理的无干扰负载监测方法，面临设备造价高与可扩展性不足的问题。..本项目发掘用电设备状态变化事件的时域稀疏性特点，提出采用稀疏多元传感器网络跟踪大量用电设备状态的可行方法。研究检测高维用电设备耗电状态的低维检测模型以稀疏传感网检测用电设备实时状态变化；基于条件半隐马尔科夫模型提出快速解码大量用电设备状态序列的算法；研究稀疏多元传感网的优化部署算法；并研究稀疏多元传感器网络跟踪用电设备状态的系统，以建立全面、准确、低成本、实时跟踪大量用电设备状态的理论方法和关键技术。

基于压缩采样方法研究网络高效、实时状态监测的关键问题。本项目研究了四大具体问题：（1）基于压缩采样方法监测全连通有向网络的可行性问题和监测算法；（2）基于压缩采样方法监测大规模复杂网络的延迟与链路异常的理论与方法；（3）基于压缩采样方法监测智能电网的用点设备实时状态的理论和方法。（4）基于压缩采样的思想，研究高效的室内定位的理论、方法和系统。针对问题（1），在无向网络压缩采样的状态监测的研究基础上，通过理论分析，证明在强连通有向网络中，通过O(klogn)数量级的链路质量采样，可以高概率的实时恢复边数为n有k项异常属性的网络的边的属性。相关工作发表在Infocom2014, ICC2014等国际会议上。针对问题（2）研究了在大规模传感网、交通网络中采用压缩采样方法恢复链路属性的理论和算法，相关工作发表在Wiopt2013, CCC2013等国际会议上。针对问题（3）研究了在智能电网中，根据电流可加性，基于压缩采样方法部署尽量少的智能电表恢复网络中所有用电设备的状态的问题，相关工作发表在ACM Transactions in Embedded Computing Systems, PESGM2014, SensorKDD2012, SmartGridComm2012, IJCAI2015等国际会议和期刊上。针对问题（4），基于压缩采样的思想研究了通过分析少量测距信号的特征，估算目标位置的理论和方法，主要工作发表在IEEE Journal on Selected Area of Communications, ICNSC2014, Mobihoc2014, Secon2014等国际知名期刊和会议上。我们还将感微知著的思想应用到医疗大数据分析方面，相关应用发表在Computer Industry等期刊上。并将此项目所产生的理论和算法拓展到基于传感数据的运动状态检测方法上，相关工作已经投稿到Secon2016, IJCAI2016等会议上。基于上述研究成果搭建了两套系统：1）基于压缩采样的用电设备监测系统；2）基于压缩采样无线室内定位系统。