This paper addresses a research gap in predicting residential energy consumption by proposing a data-driven approach utilizing features from non-electric data for training the machine-learning models. Specifically, forecasting models are trained to predict the aggregated household demand for one day, seven days, and one month. Comparisons are made to forecasting models trained on historical consumption data. By employing machine learning algorithms and exploring two distinct approaches - utilizing dwelling and occupants’ data with seasonal factors (non-electric data) and historical time-series consumption data - the study provides valuable insights into energy consumption prediction based on household characteristics. Results using data from 20 households in the UK indicate that while utilizing historical consumption data yields superior performance, the proposed approach remains a viable alternative in cases where historical consumption time series are unavailable, demonstrating promising results for forecasting household energy demand.

Industrial loads offer challenges for Non-intrusive Load Monitoring (NILM), such as phase imbalance associated with 3-phase lines. However, very little NILM research has been developed so far with this respect. This work presents a load recognition technique for NILM applying low complexity Fortesque Transform (FT). The FT decomposes the unbalanced 3-phase current waveform extracted from 3-phase aggregate power measurements to balance the given load. The 3-phases current waveform is transformed into an image-like representation using a compressedeuclidean distance matrix to improve the recognition ability further. The image representation is used as input to Convolutional Neural Network (CNN) classifier to learn the patterns of labeled data. Experimental evaluation of the industrial aggregated dataset shows that FT improves recognition performance by 5.8%, compared to the case without FT.

Industrial Kitchens (IKs) are spaces with very intensive electricity consumption. Nevertheless, unlike other sectors in industry and commerce, very little research has been carried out to understand the role of Renewable Energy Sources (RES) and Battery Energy Storage Systems (BESS) in improving the energy efficiency of such spaces. In this respect, this paper presents a year-long real-world assessment of employing the combination of solar PV and BESS in a small kitchen restaurant on Madeira Island. The results suggest that the PV-BESS can become profitable due to the high consumption levels observed in the restaurant, especially when energy arbitrage is also considered. For example, in this particular case, it is possible to cover on average 26% of the total demand with energy acquired at a lower price. The results also show that in the context of IKs, and in this case of a snack bar, it is essential to account for appliances with high power fluctuations when controlling the battery to avoid unintended grid feed-in.

Nonintrusive load monitoring (NILM) techniques are increasingly becoming a key instrument for identifying the power consumption of individual appliances based on a single metering point. Particularly, deep learning (DL) models are gaining interest in this regard. However, the challenges brought by the NILM datasets and the nonavailability of common experimental guidelines tend to compromise comparison, research transparency, and replicability. The limited adoption of efficient research instruments and lack of best practices guidelines contribute in huge part to this problem, where no features, encouraging standardized formats for benchmarking, and results sharing are offered. To address these issues, we first present a brief overview of recent best practices for DL and highlight how deep NILM research can benefit from these practices. Furthermore, we suggest a novel open-source toolkit leveraging these practices, i.e., Deep-NILMTK. The proposed toolkit offers a common testing bed for NILM algorithms independently of the underlying deep learning framework with a modular NILM pipeline that can easily be customized. Furthermore,Deep-NILMTKintroduces the concept ofexperiment templatingto offer predesigned experiments allowing to enhancing research efficiency. Leveraging this concept and DL best practices, we present a case study of creating an online NILM benchmark repository1considering eight of the most popular deep NILM algorithms. All sources relative to the tool are made publicly available on Github2along with the corresponding documentation.

This paper presents a Techno-Economic assessment of the value proposition of introducing battery energy storage in the Madeira Island electric grid, where only micro-production for self-consumption is currently allowed. The evaluation was conducted against two local micro-producers using one year of energy consumption and solar PV production measurements. The assessments considered three different pairs of battery capacity/inverter size, and the outputs analyzed considering self-consumption, self-sufficiency, and energy costs. The results show that despite the increase in self-consumption and self-sufficiency, the value proposition of battery energy storage is still considerably low even considering a massive decrease in the costs of storage. Furthermore, the results also suggest that given the small size of the solar PV installations, inverters with half the size of the installed PV capacity represent the best value for money.

This paper presents the initial results of ultrasonic waste monitoring during the operation of one industrial kitchen. Undifferentiated, paper and plastic waste bins were monitored, and a heuristic-based waste disposal event detector was developed and evaluated. The results show that it is possible to identify disposal events in the three waste bins. Furthermore, it is also possible to determine when paper and plastic are compressed to make additional room.

Avatars can be employed as a motivational tool, for example, allowing non-verbal communication that can be close to human communication. We describe two lab studies where we presented participants with avatars that communicated verbally via text and visually via expressions. In the first study, participants rated five different categories of captions and corresponding avatars. Results showed that the most persuasive, consistent and trustworthy verbal feedback was given in a humanized form. The second study was an exhaustive forced choice experiment where participants chose the happiest avatar from a pair displayed. Results showed participants found visual avatars more expressive and easier to understand than their verbal counterparts, and that users respond differently when presented with negative or positive emotions. This paper contributes to a better understanding of how to design feedback for expressive avatars.

Photographs and images often have regions which are in some sense flawed. Often, there may be a stain or an undesired feature covering a significant portion of the image, and an algorithm that can "fix" the image is desired. We propose a technique based on Heeger and Bergen's texture synthesis algorithm (1995). By integrating a composition step into the aforementioned algorithm one portion of the image can be substituted with a synthetic texture derived from another portion of the image. The goal of this synthetic texture is to create a plausible patch without visible seams or repetitive features. We compare this technique, which works on areas that are "stochastic" in nature, with other variations of image replacement which utilize a combination of image composition and texture synthesis.