Proposing a Novel Approach Non-Intrusive Load Monitoring Based on Feature Extraction Matrix and KNN Machine Learning Model
Subject Areas : Communication Engineering
Behrooz Taheri
1
(Department of Electrical Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran)
Mostafa Sedighizadeh
2
(Faculty of Electrical Engineering, Shahid Beheshti University, Evin, Tehran, Iran)
Mohammad Reza Nasiri
3
(Department of Electrical Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran)
Alireza Sheikhi Fini
4
(Power System Operation and Planning Research Department, Niroo Research Institute, Tehran, Iran)
Keywords: Feature matrix, Instantaneous frequency, NILM, Hilbert transform, KNN, Extracting features,
Abstract :
In recent years, the interest in conducting research on non-intrusive load monitoring is increasing strongly due to the increase in electrical energy consumption. Numerous studies have underscored that the implementation of non-intrusive load monitoring methods, apart from various advantages such as load response, increasing the accuracy of load prediction, etc., will increase the level of cost savings for occupants of residential structures. Recently, with the adoption of techniques grounded in deep learning, the use of these methods has also increased in order to load disaggregation. However, the most important problem with these methods is the need for complex hardware in order to train and examine the techniques. For this reason, it is necessary to transfer the power signal sampled from the smart meter to data processing centers and be analyzed. In addition to the need for high-speed communication networks, this also endangers data security. Accordingly, in this article, a non-intrusive load monitoring method is presented based on extracting the feature matrix from the instantaneous frequency signal obtained from the power signal of household appliances. The most important feature of the presented method is to increase the accuracy of the classical KNN model. The presented method has been analyzed using EMBED open-access data, which includes the consumption dataset from three different apartments. The results show that the KNN model attains significantly enhanced accuracy when using the feature matrix data introduced in this article compared to other feature extraction methods.
_||_