Automated classification of appliances using elliptical fourier descriptors

Abstract

Abstract-Non-intrusive load monitoring methods aim to disaggregate the total power consumption of a household into individual appliances by analysing changes in the voltage and current measured at the grid connection point of the household. The goal is to identify the active appliances, based on their unique fingerprint. This information can be communicated to the electricity provider and the end-user enabling the potential of smart grids. An informative characteristic to attain the appliance classification is the voltage-current trajectory. In this paper, this trajectory is represented as a binary image from which the contours are extracted. From these contours, the elliptic Fourier descriptors are calculated and used as input for several classification algorithms outputting the appliance name. Benchmarking this method on the PLAID dataset shows that the descriptors can yield a prediction accuracy up to 79%, comparable to the state-of-the-art, based on only a very compact representation (12 numbers).

VI-Based Appliance Classification Using Aggregated Power Consumption Data

“…The object recognition results in a vector of size 4 · e. This vector can be used as input for classification algorithms. As our previous work [11] shows that the random forest obtains the best performance, this is the only classifier that will be used to classify the descriptors. A random forest (RF) is an ensemble technique that classifies the data using a collection of decision trees.…”

“…This section briefly discusses the methods presented in [11] and [12]. Both methods for appliance classification cast the problem as an image recognition problem.…”

“…In [11], the image classification problem is rephrased as an object recognition problem. The contours of an object are identified from the image, characterized by elliptic Fourier descriptors and then classified with a label.…”

“…In our previous work, the problem of classifying appliances based on their VI trajectories is addressed as an image recognition problem. A first work is based on detecting contours [11]. It represents the trajectory as a pixelated image and describes a classical method for image recognition that: (1) finds the contours, (2) calculates the elliptic Fourier descriptors of the contours, and (3) trains machine learning methods using these elliptical Fourier descriptors.…”

“…Section II explains first how the current and voltage signal of a single appliance can be extracted from the aggregated data, second it restates how the current and voltage signal are transformed into a pixelated or weighted pixelated image, and at last, it briefly discusses the image recognition methods. For a complete discussion, we refer the reader to the original papers [11], [12]. The evaluation setup is described in Section III.…”