• Energy Research

Recent developments related to the energy transition pose particular challenges for distribution grids. Hence, precise load forecasts become more and more important for effective grid management. Novel modeling approaches such as the Transformer architecture, in particular the Temporal Fusion Transformer (TFT), have emerged as promising methods for time series forecasting. To date, just a handful of studies apply TFTs to electricity load forecasting problems, mostly considering only single datasets and a few covariates. Therefore, we examine the potential of the TFT architecture for hourly short-term load forecasting across different time horizons (day-ahead and week-ahead) and network levels (grid and substation level). We find that the TFT architecture does not offer higher predictive performance than a state-of-the-art LSTM model for day-ahead forecasting on the entire grid. However, the results display significant improvements for the TFT when applied at the substation level with a subsequent aggregation to the upper grid-level, resulting in a prediction error of 2.43% (MAPE) for the best-performing scenario. In addition, the TFT appears to offer remarkable improvements over the LSTM approach for week-ahead forecasting (yielding a predictive error of 2.52% (MAPE) at the lowest). We outline avenues for future research using the TFT approach for load forecasting, including the exploration of various grid levels (e.g., grid, substation, and household level).

Machine learning (ML) methods can effectively analyse data, recognize patterns in them, and make high-quality predictions. Good predictions usually come along with "black-box" models that are unable to present the detected patterns in a human-readable way. Technical developments recently led to eXplainable Artificial Intelligence (XAI) techniques that aim to open such black-boxes and enable humans to gain new insights from detected patterns. We investigated the application of XAI in an area where specific insights can have a significant effect on consumer behaviour, namely electricity use. Knowing that specific feedback on individuals' electricity consumption triggers resource conservation, we created five visualizations with ML and XAI methods from electricity consumption time series for highly personalized feedback, considering existing domain-specific design knowledge. Our experimental evaluation with 152 participants showed that humans can assimilate the pattern displayed by XAI visualizations, but such visualizations should follow known visualization patterns to be well-understood by users.

Abstract In an attempt to channel sales activities, companies often focus on ‘high value targets’ that offer attractive prospective returns. In liberalized electricity markets, commercial customers with high electricity demand constitute such high value targets. The problem when acquiring new customers, however, is that the electricity consumption is not known to the sales organization in advance. This hinders the possibility to prioritize sales targets and thus increases the acquisition cost, reduces the competitiveness within the market and ultimately leads to higher cost for electricity customers. In this study, we investigate the annual electricity consumption of enterprises by means of a dataset with 1810 company addresses in a typical town in Switzerland. We use the industry branch of the enterprises together with open big data (geographic information, online-content, social media data and governmental statistical data) to explain and predict the electricity consumption of such. Our linear regression analysis shows that information on the economic branches of the enterprises, basal area of buildings, number of opening hours and social media data can explain up to 19% of variance in electricity consumption. Economic trends (e.g., in labor market and turnover statistics) reflect changes in the electricity consumption in the investigated years 2010–2014 for several economic branches. We show, that the electricity consumption can be predicted better than a random predictor, however with a high uncertainty. Nevertheless, the open data sources can be used to identify a relevant group of companies with high consumption (more than 100,000kWh per year) with good accuracy.

Heat pumps play an important role in the energy transition. They can extract renewable energy from the air or ground and increasingly replace fossil heating systems in buildings. In operation, however, heat pumps often consume more electricity than necessary due to incorrect settings and installation deficiencies. Although many setting and installation issues are easy to fix, problems often go unnoticed, and the saving potential from quick fixes remains unclear. In a study with 297 Swiss households (41 treatment, 256 control) running for four years, we investigated an energy efficiency campaign in which the treatment group received a professional heat pump inspection and user training. We found considerable heterogeneity with respect to the savings achieved. We derived two criteria based on smart meter data that enable utilities to identify relevant households and thus boost the impact of such efficiency campaigns: For example, pre-selecting half of the households based on available information results in average savings of 1,805 kWh (15.2%) per year and household in the high-potential group compared to no savings in the low-potential group. Thus, heat pump inspections among pre-selected households can lead to large, cost-effective electricity savings, and we show that common smart meter data makes such pre-selection feasible. Energy Policy, 169 ISSN:0301-4215

In the light of global sustainability efforts, heat pumps offer environmental benefits, but their complexity and potential misconfigurations often lead to homeowner dissatisfaction due to inaccurate heating and lower-than-expected efficiency. Among the most important and complex settings is the heating curve and yet there are no easy-to-use methods to optimize it after its initial set-up. This study aims to develop ready-to-use guidelines for optimizing the heating curve with energy-efficient adjustments that improve room comfort and prevent suboptimal user changes, all without requiring additional sensors like room thermostats. Based on interpretable linear models, estimated on 3995 air-to-water heat pumps, located in Central Europe, we select the least energy-intensive heating curve shift for each outdoor temperature, needed to meet room thermal comfort. We find that the standard parallel shift of the heating curve is only the optimal approach when the average outdoor temperature is between 2 ⁰C and 5 ⁰C. Outside this range, the heating curve should be moved at its starting or the endpoint. Simulation shows that by translating user input to the room controller with our proposed changes, 84.42 % of the heating curves can be improved, reducing the share of misconfigured heating curves from 24.01 % to 7.08 %. This leads to an average reduction in yearly energy consumption of 4.02 % and an increase in the seasonal coefficient of performance by 2.59 % on average. By introducing ready-to-use heating curve improvement guidelines, we aim to increase efficiency and confidence in heat pump technology, ensuring its adoption to meet carbon emission targets. Applied Energy, 389 ISSN:0306-2619 ISSN:1872-9118

Data is provided to reconstruct the key figures (Figure 5, Figure 7, Figure 8 and Figure 9) of the paper “Optimization of heating curves for heat pumps in operation: outdoor temperature ranges for energy-efficient heating curve shifts”. Where Figures 7, 8 and 9 display simulated data of the study and were not directly measured. This data does not allow for a direct evaluation of the manufacturer or the heat pumps themselves, as the measurements and simulations were not designed as a controlled experiment. Potential measurement errors and uncertainties in the simulations have not been systematically validated.

Tailored energy efficiency campaigns that make use of household-specific information can trigger substantial energy savings in the residential sector. The information required for such campaigns, however, is often missing. We show that utility companies can extract that information from smart meter data using machine learning. We derive 133 features from smart meter and weather data and use the Random Forest classifier that allows us to recognize 19 household classes related to 11 household characteristics (e.g., electric heating, size of dwelling) with an accuracy of up to 95% (69% on average). The results indicate that even datasets with an hourly or daily resolution are sufficient to impute key household characteristics with decent accuracy and that data from different yearly seasons does not considerably influence the classification performance. Furthermore, we demonstrate that a small training data set consisting of only 200 households already reaches a good performance. Our work may serve as benchmark for upcoming, similar research on smart meter data and provide guidance for practitioners for estimating the efforts of implementing such analytics solutions. Electronic Markets, 28(4), 453–473

Abstract Background Users create serious amounts of Volunteered Geographic Information (VGI) in Online platforms like OpenStreetMap or in real estate portals. Harvesting such data with the help of business analytics and machine learning methods yield promising opportunities for firms to create additional business value through mining their internal and external data sources. Energy retailers can benefit from these achievements in particular, because they need to establish richer customer relations, but their customer insights are currently limited. Extending this knowledge, these established companies can develop customer-specific offerings and promote them effectively. Methods This paper gives an overview to VGI data sources and presents first results from a comprehensive review of these crowd-sourced data pools. Besides that, the value of two exemplary VGI data sources (OpenStreetMap and real estate portals) for predictive analytics in energy retail is investigated by using them in a household classification algorithm that recognizes specific household characteristics (e.g., living alone, having large dwellings or electric heating). Results The empirical study with data from 3,905 household electricity customers located in Switzerland shows that VGI data can support the recognition of the 13 considered household classes significantly, and that such details can be retrieved based on VGI data alone. Conclusion The results demonstrate that the classification of customers in relevant classes is possible based on data that is present to the companies and that VGI data can help to improve the quality of predictive algorithms in the energy sector.