Energy meters in DH Substations for Heat Consumption Characterization and Prediction Using Machine-Learning Techniques

The use of smart energy meters enables the monitoring of a large quantity of data related to heat consumption patterns in buildings connected to DH networks. This information can be used to understand the interaction between the building and the final users´ without accurate information about building characteristics and occupational rates. In this paper, presented at the Beyond 2020 digital event, an intuitive and clarifier data-driven model is presented, which couples heat demand and weather variables. This model enables the disaggregation of Space-Heating & Domestic Hot water demand, characterization of the total heat demand and the forecasting for the next hours. Simulations for 53 buildings have been carried out, with satisfactory results for most of them, reaching R2 values above 0.9 in some of them.

Authors: Mikel Lumbreras, Roberto Garay and Antonio Garrido.

Introduction

Information and Communication Technologies (ICTs) are changing the way of working in many sectors within technology, revolutionizing energy management. The computational capacity has rapidly increased in the last few decades, enabling a novel way of running simulations, named as data-driven models based on a large amount of information. Regarding energy systems, the core technology that is currently used for collecting the necessary data is smart energy meters. These devices enable the gathering of energy consumption, among other thermal variables such as temperature, with a specific time-frequency, typically 15 min or 1 hour [1]. The drivers behind the employment of smart metering include energy load management, peak or demand reduction, fraud reduction, accurate billing and water conservation [2].

In the past decade, the interest in energy meters was focused on electricity smart grids where the introduction of these devices at the household level enabled the detailed recording of consumption. The study of the gathered data enables the electricity operators to identify patterns of use at the consumer level and this way optimizing electricity grid functioning and allowing to integrate flexibility techniques into the entire system. However, this study is focused on the possibilities that recent energy meters offer to District-Heating (DH) system and the heat flow management.

This study is focused on the possibilities that recent energy meters offer to District-Heating (DH) system and the heat flow management.

DH networks are responsible for covering 13% of the total heat demand in the EU [3]. Although most of the DH networks are located in the northern countries due to their large load for Space-Heating (SH), [4] & [5] show the cost-effectiveness of these heat supply systems in other locations. Traditionally, DH systems have been based on a large production plant, typically a Combined Heat & Power (CHP) plants, due to the advantages of producing electricity and heat at the same time. DH systems, as well as other energy systems, need to evolve and adapt to new requirements, reaching the 4th Generation DH systems or 4GDH.

The so-called 4GDH introduce the design and operation improvements necessary to reach the decarbonization objectives of the heat supply within the EU. The introduction of Renewable Energy Sources (RES) into the heat production mix is the core of the 4GDH. These new-generation DH networks are also known as Ultra Low Temperature or ULT DH because the temperature of the supply line is supposed to be reduced up to 45ºC when the current network supply heat is typically above 75ºC. The reduction of supply temperature enables the introduction of low-grade RES such as the solar thermal system, geothermal heat pumps, waste heat streams from Data-centres etc. Moreover, a lower temperature gradient in the distribution pipelines reduces heat losses, optimizing the heat distribution system.

DH networks, as well as electricity grids, needs to evolve into active systems where the instant heat load that needs to be satisfied controls the heat production and vice versa.

DH networks, as well as electricity grids, needs to evolve into active systems where the instant heat load that needs to be satisfied controls the heat production and vice versa. The gradual introduction of RES into the heat production mix also incorporates a degree of uncertainty, due to the changing conditions of weather variables. This makes even more important the information that can be achieved from the energy meters about the consumption patterns in the different buildings that conform to the overall network.

This paper presents a data-driven model based on machine-learning techniques, such as classification and regression, that allow the characterization of the total heat load in a building and enabling the heat load forecast for the next time-steps.

The rest of the paper is divided into 5 sections. In the next chapter, a literature review that summarizes the state of the art of data-driven models in terms of DH networks is presented. Then, the different methods used in the study are shown, followed by the analysis and results chapter. To conclude, the conclusions from the study are drawn, indicating as well, which are the possibilities for the future that can follow this study.