Also referred to as Non-Intrusive Load Monitoring (NILM), the critical task is to estimate individual load or appliance consumption in Watts, not simply when an appliance was switched on or off. For the Eco-bot project, and in order to meet the reliability constraints of the pilots’ customer base, NILM algorithms should meet accuracy, sampling rate, submetering data availability and near real-time operation constraints. It is worth noting that with current data protection requirements imposed on utilities across Europe, it is now almost impossible to obtain historical data for the building to be disaggregated and a clear need is emerging for transfer learning or training-less solutions. Furthermore, NILM algorithms are typically proposed for meter measurements available at relatively high sampling rates (0.02Hz or higher), but there is a clear challenge in NILM approaches for disaggregation of smart meter measurements available hourly. Having evaluated NILM offerings in the literature in light of the above constraints, our findings are that no state-of-the-art load disaggregation algorithms could be used readily to meet reliability and scalability requirements. Therefore, the Eco-bot consortium developed novel transfer learning-based load disaggregation algorithms for different resolutions, as well as communicating effectively with the back-end in updating NILM results for all pilots daily, demonstrating innovation in this task going beyond the state-of-play in load disaggregation algorithm development. Eco-bot will validate effectiveness of transfer learning and the findings could be of high interest to industrial stakeholders such as energy retail and energy service providers.
The eco-bot project aims at increasing energy consumers’ engagement towards sustainable energy consumption. To achieve this goal, it is necessary to create a model of individual energy consumer behaviour based on the results of behavioural studies combined with the analysis of data obtained from smart metering and other sources. An important element of the eco-bot application will be a module responsible for recognizing customer behaviour that relates to energy saving and their relation to environmental protection. Designing such a module is possible thanks to the use of statistical tools and, above all, the classification model. The classification model, that will be used as a building block in the eco-bot application, belongs to the family of data mining methods. The effect of this model is customer segmentation, which will allow the provision of personalized recommendations that will help change their behaviour to more pro-ecological.
Considering the sphere of consumption and its individual elements, including rationality of consumer behaviour, it should be noted that individual elements of this system are mutually intertwined with each other and simultaneously they build a multilaterally linked system with the external environment. The actions of various determinants cause changes both in the level as well as the structure of consumption. Considering only non-economic factors related to the shaping of individual attitudes and resulting mainly from their personality traits, the adopted system of values, motives, emotions, rules of conduct can be seen in purchasing activities of individual characteristics of each consumer. Therefore, it is extremely important to know the needs of consumers, the level of electricity demand and the changes in consumer attitudes. Hence, an useful skill is to predict the behaviour of energy consumers and their assessment. Changing economic, political, social and cultural conditions have an extremely significant impact on consumers’ lifestyles. The present environment is characterized by pluralism of styles and ideologies, as well as by stressing consumer’s individualism through consumption, the importance of information and the diversity of consumer’s orientation. All these factors as well as the requirements of the modern consumer were tried to be included in the segmentation of eco-bot users. While constructing the segmentation of eco-bot customers, the authors took into account that factors influencing consumer behaviour are characterized by a fairly high dynamics of changes. What is more, the change in external conditions often contributes to the change in internal stimuli, which play an extremely important role in consumption process.
Natural Language Processing
Advancements in technology and specifically in Natural language processing (NLP) allows bots to be truly intelligent. We can now process and analyze natural language data much more accurately than we were able to in the past. The Eco-Bot project is faced with the challenge to build conversation agents that will interact with a diverse audience of electric power utility customers, residential users and energy management system users in multiple languages.
On top of managing complex requests, offering when necessary menus and options to guide the discussion while giving due priority to the user’s inquiries, the frontend solution offers personalized responses. Behind the scenes, the bot is building the personalized experience for the user with its integration through the backend with the NILM offerings, the behavioral models and the pilot services. Given the requirements of the novel technologies that Eco-Bot integrates, user-feedback has been enabled to allow fine-tuning of algorithms and models. Furthermore, a mechanism for collecting metrics and findings from the user experience is built to drive the efforts for improved interactions over time while Artificial Intelligence continues breaking new grounds.