Low-Frequency Non-Intrusive Load Monitoring of Electric Vehicles in Houses with Solar Generation: Generalisability and Transferability

Abstract

Electrification of transportation is gaining traction as a viable alternative to vehicles that use fossil-fuelled internal combustion engines, which are responsible for a major part of carbon dioxide emissions. This global turn towards electrification of transportation is leading to an exponential energy and power demand, especially during late-afternoon and early-evening hours, that can lead to great challenges that electricity grids need to face. Therefore, accurate estimation of Electric Vehicle (EV) charging loads and time of use is of utmost importance for different participants in the electricity markets. In this paper, a scalable methodology for detecting, from smart meter data, household EV charging events and their load consumption with robust evaluation, is proposed. This is achieved via a classifier based on Random Decision Forests (RF) with load reconstruction via novel post-processing and a regression approach based on sequence-to-subsequence Deep Neural Network (DNN) with conditional Generative Adversarial Network (GAN). Emphasis is placed on the generalisability of the approaches over similar houses and cross-domain transferability to different geographical regions and different EV charging profiles, as this is a requirement of any real-case scenario. Lastly, the effectiveness of different performance and generalisation loss metrics is discussed. Both the RF classifier with load reconstruction and the DNN, based on the sequence-to-subsequence model, can accurately estimate the energy consumption of EV charging events in unseen houses at scale solely from household aggregate smart meter measurements at 1–15 min resolutions.

Explainability-Informed Feature Selection and Performance Prediction for Nonintrusive Load Monitoring

“…HMM approaches are inefficient when the number of disaggregated appliances increases and they have high computational complexity [ 1 ], which is not the case for UGSP-based methods [ 1 , 11 ]. Popular supervised NILM approaches include decision tree (DT) [ 12 ], boosting-based ensemble algorithms [ 13 ], and deep learning (DL)-based approaches (see [ 2 ] for a review), all of which require well-labelled datasets for training the models. Supervised ensemble and DL approaches are capable of obtaining good generalisability and transferability to houses never seen before, but larger training datasets are required and are often complex [ 1 , 2 ].…”

A Non-Intrusive Identification Approach for Residential Photovoltaic Systems Using Transient Features and TCN with Attention Mechanisms

“…Reference [30] uses three intelligent algorithms, namely RF, KNN, and an artificial neural network (ANN), for the classification and power decomposition of electric vehicles and PV systems. Reference [31] utilizes a sequence-to-subsequence Deep Neural Network (DNN) with a conditional Generative Adversarial Network (GAN) approach for the low-frequency non-intrusive monitoring of electric vehicles in dwellings containing photovoltaic devices; reference [32] proposes a novel non-intrusive load monitoring method based on a ResNet-seq2seq network to decompose the power of household appliances and with distributed energy resources in residential areas. All of these methods utilize steady-state power as a feature for identification based on PV grid disconnection behavior and steady-state power characteristics.…”

Toward Transparent Load Disaggregation—A Framework for Quantitative Evaluation of Explainability Using Explainable AI

“…For AI-based NILM, the majority of work has focused on addressing technical robustness in the form of accuracy, reliability and reproducibility across different datasets [2], [6], [7] and data transparency through the use of public, peerreviewed and well-documented datasets [8], [9], with limited research in the area of privacy protection [10]- [12] and technical explainability [13]- [15]. The majority of deep learningbased NILM approaches are designed as "black-box" systems due to their inherent algorithmic complexity and absence of explainability.…”