Machine-learning algorithms to categorize consumers

The local distribution grid operators have been facing technical, economic and regulatory challenges when operating and planning their grids to accommodate increased integration of renewable generation and electrification of vehicles. Fig. 1 shows an example of two 10 kV distribution grids connected in parallel to the 40 kV
upper-stream grid. The local grid A is dominated with wind power and industrial and commercial consumers, whereas the local grid B is dominated with residential consumers, with/without solar PV, battery storage and electric vehicles, and with different heating systems. During the day time of a working day, there may be net
excess generation from solar PV in local grid B and net load consumption in local grid A; whereas during the night, there may be net excess generation from wind power in local grid A, and net load consumption in grid B instead. However, these excess generation has to be transmitted up to the 40 kV grid and then distributed down again to the local grid with net load consumption. This way of power flow induces additional losses in the grid. An alternative path for the power flow is to interconnect grid A and grid B locally through power electronic converters as shown in Fig.1. Consequently, not only the power losses can be reduced, but also the total peak power flow exchange between the local grids and the upper-stream grid can also be reduced, which can significantly minimize the need to expand the size of the substation and the upper-stream grid. Another benefit of such an interconnection is that if the upper-stream grid is unavailable due to faults or regional black out, the local grids can still operate in island mode by sharing their generation and flexibility resources including energy storages. This is a typical function in the so-called microgrid operation and can significantly increase the reliability and resiliency of the power grids. Thus, in order to assess and quantify the benefits of joint operation of two or multiple
adjacent local grids, it is essential to understand the electrical characteristics of consumers, producers and prosumers within these local grids. In this way, the extent of the complimentary power flow behavior can be identified between different types of consumers, and between consumers and producers located at the different adjacent
grids. Consequently, through their interconnected and coordinated operation, it is possible to increase the grid efficiency and reduce the peak power flow exchange between the upper-stream and the local grid, and to reduce the size of energy storage needed to enable the island operation of the local power grids. As a first step, this project focuses on the electrical characteristics of different types of
consumers. Based on the discussions with the local grid operators and literature study, there is a general lack of knowledge in the electrical characteristics of different types of consumers. The main information available for a consumer is its monthly consumed
energy for the billing purpose and a rough annual peak load estimation according to the Velander formula developed for a given consumer type. One reason for this is that there is no obligation for the consumers to notify the grid operator about energy efficiency measures or which type of heating system that is used. This leads to a low customer knowledge, and together with a too rough customer categorization, it adds uncertainty for planning decisions. On the other hand, the large-scale rolling out of smart meters for consumers provide great opportunity to develop methods to
characterize detailed electrical characteristics of the consumers, in which the method of pattern recognition and machine learning including neutral networks is considered to have a great potential to be applied.

Partner organizations

  • Göteborg Energi AB (Private, Sweden)
Start date 01/01/2019
End date The project is closed: 30/06/2019

Published: Thu 25 Jul 2019.