Reinforcing the electricity grid with Evolutionary Algorithms

In our modern daily life, many activities require electricity, for example, the use of domestic appliances, manufacturing, communication, and transportation. It is therefore essential to maintain a reliable supply of electricity to ensure the operation of such activities.

Publication date: 09-10-2018

In our modern daily life, many activities require electricity, for example, the use of domestic appliances, manufacturing, communication, and transportation. It is therefore essential to maintain a reliable supply of electricity to ensure the operation of such activities. The electricity supply, in a large part, depends on the underlying electrical networks that transfer electricity from power plants to meet the demand of end users. In the past, electricity consumption has grown over time and, at some point, the electricity demand will exceed the current capacity of certain network assets, causing overloads on parts of the networks.

Functioning under overload conditions reduces the reliability of the networks and also damages network assets. Network reinforcement is thus required. This incurs substantial investment costs and time-consuming activities, such as acquisitions of new assets, constructions of substations, and installations of suitable cables and other electrical devices. Network operator companies, therefore, need to make suitable expansion plans to enhance the capacity of their networks.

In his thesis 'Design and Application of Scalable Evolutionary Algorithms in Electricity Distribution Network Expansion Planning’  Hoang Luong of CWI’s Life Science & Health group and Intelligent and Autonomous Systems group, focuses on medium-voltage distribution networks. The aim of his research is to develop robust computational methods to assist network operators in tackling the distribution network expansion planning (DNEP) problem. Evolutionary algorithms (EAs) are a promising type of optimization algorithm to tackle the DNEP problem. However, there exist two major challenges in applying EAs to real-world optimization tasks. First, proper control parameter settings for an EA are very hard to be determined. Practitioners often need to perform multiple optimization runs with different parameter settings in a time-consuming trial-and-error manner. Second, EAs that are employed for solving real-world problems like DNEP should be designed with good scalability in mind to ensure being able to efficiently handle a wide range of network sizes and structures.

The contribution of this thesis is twofold. First, it shows how the DNEP problem with real-world constraints can be modeled effectively and efficiently. The problem formulation is highly customizable such that it can be straightforwardly modified by network operators to suit their needs: static or dynamic planning, employing only traditional assets or also smart grid technologies, and optimizing with respect to one objective or handling multiple objectives at the same time. Second, the thesis proposes guidelines for constructing scalable and user-friendly EAs that do not require users to tune their control parameters. Following the guidelines, we have designed the Multi-Objective Gene-pool Optimal Mixing Evolutionary Algorithm (MO-GOMEA), which is shown to be capable of efficiently and effectively tackling real-world optimization tasks, such as different versions of the DNEP problem considered in this thesis.

More information

Everyone is welcome to attend the public defense of Hoang Luong on Wednesday, 17 October 2018, from 14.30 hrs in the Senaatszaal of the Auditorium (Aula), Delft University of Technology, Mekelweg 5, Delft.

PhD-thesis: 'Design and Application of Scalable Evolutionary Algorithms in Electricity Distribution Network Expansion Planning’. Promotors: Han La Poutré (CWI, TU Delft) and Peter Bosman (CWI, TU Delft).

18 October Kalyanmoy Deb (member of the reading committe of Hoangs' dissertation) will give a seminar about Evolutionary Algorithms at CWI (Title: Breaking the Billion-Variable Barrier Using Customized Evolutionary Optimization), more information and registration via our website