Core Concepts and Methods in Load Forecasting
With Applications in Distribution Networks
This book is an open access publication.
Open Access This book is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this book are included in the book's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the book's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
One of the oldest and most cherished desires for any quantitative scientific and technological field is the ability to develop forecasts of the, as yet, unforeseen. Of course, whatever has been observed already is, in most applications, a moving conveyor belt. Hence, this ambition raises fundamental questions, such as What actually is a good forecast?, and What is the value of older data?.
Many realistic challenges involve spikey data, perhaps within nonstationary (drifting and step change) settings. In turn, this fact modifies the answers to the questions above. If we aim to predict realistic forward profiles, as opposed to point estimates, we shall need to generate realistic structures (with spikes, gaps, and so on). This is problematical for methods that treat errors as a noise component that is to be minimised (in some way). It presents a meta problem.
This book is timely and very much needed. From my own experience, I see that whenever novel concepts are posited so as to address some of these issues, they are eagerly feasted upon and further developed by many scientists within a plethora of applied fields. The breadth of ideas and methods here is what gives this book its power: readers will return to it again and again. For data science applications, there is a need for options that address “what works” as well as “why it works”. In many fields of highly regulated sciences pertaining to the interests of public individuals, it is simply unethical to present opaque methods; why is as essential as what. This provides transparency and assurance to the subject who feels the consequences of the forecasting outputs.
No serious professional data scientist can be oblivious to the contents presented here. The interests of readers should be both refined and peaked by dipping into this book. Open it and read at random (like a grasshopper), or crawl through it (like an ant): your investment of both interest and effort will be rewarded. Now that is a forecast!
Electricity networks around the world are rapidly moving towards digitalisation, producing an ever-increasing amount of data. This data is opening up vast opportunities to decarbonise the energy system, as well as helping us to increase the efficiency of the energy we use. In turn, this improves the chances of addressing some of the most urgent problems causing climate change.
To implement the necessary data analytical and modelling techniques for the future, low carbon economy requires a wide range of skills, knowledge, and data literacy. Without these, there is a genuine threat of a skills gap where there is insufficient personnel who can put into practice these methods and models. One of the main goals of this book is to help support these vital skills by providing an accessible but thorough introduction to the techniques required for household and low voltage load forecasting.
The area of load forecasting at the Low Voltage (LV) level is relatively immature compared to high voltage or national level forecasting, and there are many important and exciting areas still to be explored. The volatile and spiky nature of LV level demand provides many challenges within applications which utilise forecast inputs, but also within the forecast themselves. Hence, a second major aim of this book is to lay a strong foundation for researchers and innovators to develop the future novel methods and advanced algorithms that can produce ever more varied products and services.
The seeds of this book started over 10 years ago when the authors began their research into forecasting smart meters and low voltage demand. At the time there was very little data available in this area, and its unique challenges such as the “double penalty effect” were hidden or ignored. Much research simply applied the techniques which had been successfully applied at the national or system level, with very little thought to their appropriateness to the LV system. The area is now rapidly progressing and starting to embrace the much-needed probabilistic techniques and advanced time series machine learning methods. Additionally, more data is becoming available all the time, further supporting the development of robust benchmarks and new applications.
Although data science techniques are rapidly developing in LV level load forecasting (and will continue to), this book provides the fundamental techniques which serve as the foundation for anyone interested in this area (and load forecasting more generally), as well as the timeless, but necessary, principles and approaches underlying them. Discussed in this book are the core concepts of time series forecasting, the unique features of LV level demand, fundamental data analytics, methods for feature engineering, a plethora of statistical and machine learning forecast models, as well as a demonstration of them in a case study applied to real-world data. At the end of this book, the reader should be well versed on the complete load forecast process, as well, perhaps, ready to develop some novel models of their own!
It was our desire to produce a book “we wished was available when we were first starting out in this field”. We feel that we have achieved this, and we hope that it supports you on your journey into load forecasting.
We would like to thank the University of Reading, who funded the Open Access publication of this book. The authors would also like to thank all the colleagues who have contributed to the content of this book either directly or indirectly. Much of the knowledge and work presented here have come from the authors’s research, collaborations, and discussions. In particular a special thanks to Peter Grindrod, Danica Vukadinovic Greetham, Colin Singleton, Billiejoe Charlton, Florian Ziel, Ben Potter, Timur Yunusov, JonathanWard, Laura Hattam, Lauren Barrett, Matthew Rowe, Bruce Stephen, David O’Sullivan, Gideon Evans, Maciej Fila, Wayne Travis, Rayner Mayer, Asmaa Haja, Marcel Arpogaus, Alexander Elvers, Brijnesh Jain, and Daniel Freund. Furthermore, the authors would like to give special thanks to Siddharth Arora, Charlotte Avery, Jethro Browell, Alison Halford, Sam Young, and especially Georgios Giasemidis for their time to provide invaluable feedback on drafts of the book.
The Case Study in the book will be based on real data from low voltage residential feeders from the Thames Valley Vision project. We would like to thank Scottish and Southern Electricity Networks and our partners who helped support this work throughout the project.
The author would also like to thank the following journals that have allowed the use of figures and material published by the authors in this book:
-
International Journal of Forecasting [35] which provides the basis for the LV residential feeder forecasting case study in Chap. 14.
-
Springer International Publishing for permitting the use of material from Chapter [85] which is used to demonstrate the storage control example in Chap. 15.
-
Applied Energy for use of the Network diagram in Chap. 2 which was originally plotted in [34].