Sam West

Unsupervised feature selection using swarm intelligence and consensus clustering for automatic fault detection and diagnosis in Heating Ventilation and Air Conditioning systems

Graphical abstractDisplay Omitted HighlightsOur algorithm aims to improve the feature quality in general fault diagnosis system.The algorithm filters out redundant features using consensus evolutionary clustering.The algorithm was tested on the ASHRAE-1312-RP experimental fault data.Sensitivity & specificity were >95%, with considerably less false positives up to as low as 1.6%. Various sensory and control signals in a Heating Ventilation and Air Conditioning (HVAC) system are closely interrelated which give rise to severe redundancies between original signals. These redundancies may cripple the generalization capability of an automatic fault detection and diagnosis (AFDD) algorithm. This paper proposes an unsupervised feature selection approach and its application to AFDD in a HVAC system. Using Ensemble Rapid Centroid Estimation (ERCE), the important features are automatically selected from original measurements based on the relative entropy between the low- and high-frequency features. The materials used is the experimental HVAC fault data from the ASHRAE-1312-RP datasets containing a total of 49 days of various types of faults and corresponding severity. The features selected using ERCE (Median normalized mutual information (NMI)=0.019) achieved the least redundancies compared to those selected using manual selection (Median NMI=0.0199) Complete Linkage (Median NMI=0.1305), Evidence Accumulation K-means (Median NMI=0.04) and Weighted Evidence Accumulation K-means (Median NMI=0.048). The effectiveness of the feature selection method is further investigated using two well-established time-sequence classification algorithms: (a) Nonlinear Auto-Regressive Neural Network with eXogenous inputs and distributed time delays (NARX-TDNN); and (b) Hidden Markov Models (HMM); where weighted average sensitivity and specificity of: (a) higher than 99% and 96% for NARX-TDNN; and (b) higher than 98% and 86% for HMM is observed. The proposed feature selection algorithm could potentially be applied to other model-based systems to improve the fault detection performance.

Literature review of power disaggregation

There are mainly two classes of approaches in power disaggregation, including Intrusive Load Monitoring (ILM) and Nonintrusive Load Monitoring (NILM). This paper presents the literature review on the NILM approaches. NILM is a process for detecting changes in the voltage and current going through a house, deducing what appliances are used in the house, as well as their individual energy consumption, with a single set of sensors. Different strategies and approaches for NILM systems have been developed over the past thirty years. This paper reviews the current state of the algorithms and systems of NILM. The paper points out that NILM can be utilised presently on available commercial devices and provides meaningful feedback. Our vision on the future of NILM is also summarized.

The Virtual Power Station

This paper introduces our virtual power station concept, where individual small-scale renewable energy sites are aggregated together to form a ldquovirtualrdquo power station that appears as a single dispatchable quantity to the wider electricity system. As such a quantity has greater benefit to the wider system than the individual responses of many uncoordinated energy sites, the virtual power station can improve the payback period for renewable energy systems. The concept relies on sophisticated prediction and aggregation mechanisms, to firstly anticipate the power available from a renewable energy system, and then aggregate many small systems into one quantity with reliable output.

Power load event detection and classification based on edge symbol analysis and support vector machine

Energy signature analysis of power appliance is the core of nonintrusive load monitoring (NILM) where the detailed data of the appliances used in houses are obtained by analyzing changes in the voltage and current. This paper focuses on developing an automatic power load event detection and appliance classification based on machine learning. In power load event detection, the paper presents a new transient detection algorithm. By turn-on and turn-off transient waveforms analysis, it can accurately detect the edge point when a device is switched on or switched off. The proposed load classification technique can identify different power appliances with improved recognition accuracy and computational speed. The load classification method is composed of two processes including frequency feature analysis and support vector machine. The experimental results indicated that the incorporation of the new edge detection and turn-on and turn-off transient signature analysis into NILM revealed more information than traditional NILM methods. The load classification method has achieved more than ninety percent recognition rate.

An online load identification algorithm for non-intrusive load monitoring in homes

Non-intrusive load monitoring (NILM) systems, employed at the utility-customer interface point, provide real-time power usage data to the grid and present real-time per-appliance price data to consumers. Such information will allow consumers to participate in the electricity market, resulting in energy conservation, demand reduction and other benefits. For these reasons, NILM has become an active area of research. In the current paper, a new algorithm is proposed in which both state-switching event identification and load recognition are included. Furthermore, a statistical variable, i.e. cross correlation coefficient, and a statistical method, i.e. crossed index weight determination method, are employed. The key components of the new algorithm, including basic concepts of signal signatures, structure and methodology of the algorithm, are presented. This algorithm is verified by the experiments to identify hybrid home appliances in the laboratory. The experimental results show that the introduction of cross correlation coefficients reveals more information, and that this new algorithm offers minimal computational burden with similar performance to other NILM algorithms reported as well.

Automatic Feature Selection Using Multiobjective Cluster Optimization for Fault Detection in a Heating Ventilation and Air Conditioning System

The performance of Automatic Fault Detection and Diagnostics (AFDD) algorithms to identify faults in complex building Heating Ventilation and Air-Conditioning (HVAC) systems depend on the appropriateness of features. This paper proposes a knowledge-discovery approach for discovering characteristic features using Multi-Objective Clustering Rapid Centroid Estimation (MOC-RCE). The proposed method has been tested on experimental fault data from the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) research project 1312-RP Winter 2008 dataset. An experiment involving 100 clustering trials shows that using the proposed method, on average 15 characteristic features have been selected from the original 320 features. Sensitivity, specificity, accuracy, precision, and F-score values of greater than 95% are achieved with the provided features.

Real-Time HVAC Sensor Monitoring and Automatic Fault Detection System

Real-time monitoring of heating, ventilation, and air conditioning (HVAC) systems is crucial to maintaining optimal performance such as providing thermal comfort and acceptable indoor air quality, guaranteeing energy saving, and assuring system reliability. In a realistic situation, HVAC systems can degrade in performance or even fail due to a variety of operational problems, such as stuck open or closed air dampers and water valves, supply or exhaust air fan faults, hot or chilled water pump faults, and inefficiencies in the way HVAC systems or pieces of equipemnt are controlled. This paper presents automatic fault detection techniques, as well as a key sensor sets selection approach that can help to maintain the performance of HVAC systems, and optimise fault detection results. One key step to make sure the approach succeeds is the sensor feature selection process. This paper implements the ensemble rapid centroid estimation (ERCE) as the data-driven sensor and feature selection algorithm, which is the core method to assure the automatic fault detection can function correctly. Instead of choosing sensors manually, ERCE method can automatically select representative features that are unique and relevant to the faults in a HVAC system. The methodology presented is implemented in real-world commercial buildings with experimental results showing that different types of faults are detected successfully.

Power signature-based non-intrusive load disaggregation

For more efficient energy consumption, it is crucial to have accurate information on how power is being consumed through a single power measurement. It benefits both market participants such as retailers, network businesses and also power consumers. This means that the metering device must not only be able to distinguish between different loads on a common circuit, but also decipher their respective power consumption. This paper gives our investigation on power load signature and power decomposition. In particular, the paper focuses on the development of power decomposition algorithm based on support vector machine (SVM), i.e., estimating the power proportion of constant power (CP) loads to constant impedance (CI) loads. The power decomposition algorithm in this paper is working on steady-state power load signal.