PRE2023 1 Group1

Group members

Name	Student number	Major
Sven Bendermacher	1726803	BAP
Marijn Bikker	1378392	BAP
Jules van Gisteren	1635530	BAP
Lin Wolter	1726927	BAP

Problem statement

The problem that will be addressed in this project is the development of an algorithm that finds the moment the price for electricity is the lowest for electric devices, reducing the amount of money one loses to its yearly electricity bill. With the variable fee of the energy contract, the timing of the use of your electric device matters for the price one pays for the electricity. At moments when the energy generation exceeds the demand of electricity, the price will be low. An smart algorithm putting on the dish washer or washing machine at the moment the price is the best during the night could be profitable. Today, people can already take advantage of the hourly fluctuating prices by watching the costs of a kilowatt in an app, and then setting the time of an high electricity demanding device at an inexpensive moment. This is a practice for some people, but not everyone likes put in the effort to do this consistently. For those this algortihm combined with an app that does the work for you can come in handy, in particular if the algoritm can be connected to smart electric devices to operate them, making sure the system operates on its own without the need of human activity.

Project Requirements, Preferences, and Constraints

Creating RPC criteria

Setting requirements

Setting preferences

Setting constraints

RPC-list

Requirements

• Safety
• User feedback/interaction
  • App should be easy to use
• Implementable
  • Relatively cheap
  • No infrastructural changes in the electric circuit for a device that is able to turn on electric device
  • Algorithm should be reliable

Preferences

• App should be easy to use
• App is looking good
• App tracking the total amount of saved money

Constraints

• Environment (house)
  • Not-smart electric devices
  • Times when the electric devices cannot be turned on, a not ready dishwater for example.

Users

The possible users for the algorithm for low energy pricing are quite vast, ranging from private homeowners to businesses. Private homeowners can use this app to lower their expenses on energy, which is especially important due to the surge in energy prices due to everything happening in geopolitics. Homeowners could thus use this to turn on their appliances at the right times leading to huge chances on saving large sums of money. Next to private homeowners even factories or businesses could look into using the algorithm, their sometimes-intensive use of energy could then also be better placed at more beneficial times. Energy intensive procedures needed to for example fabricate a certain product could then be done at better times lowering the costs of production leading to higher profits, which is in capitalism of course one of the main drivers in business. Thus, the users for the algorithm spread almost everyone, since almost no one lives without using electricity in this current era.

Private homeowners

As described above one of the most important users would be private homeowners, since the developed app/algorithm would enable them to save large sums of money. This users most important wish would encompass mostly a good working app which is easy to navigate as well as a trustable algorithm. The algorithm should be trustable since if the algorithm is wrong very often there would be no need to use the app, some errors can be accepted though since in no case would using the app cost more money than when using electricity on chosen times, only a perfect human being would be able to spread the electricity usage better than the algorithm.

Companies

The companies would similarly to the homeowners also want a good working app, with again a huge focus on the accuracy of the algorithm. Companies would also benefit from other built in functions such as overviews of electricty use as well as a method to make sure the responsible people are informed, which should thus need everyone to be connected to a single account. Companies would thus make profit from the use of the app/algorithm but would not need major alterations to the normal version for private homeowners.

Other institutions

Since the use of electricity is such a general thing in life, any group of people, company or institution having control over their electricity use could use the app to lower their electricty prices. An important question that then arises however is, in the case that many people start using the app, can the algorithm adapt to correctly spread the use of electricity. Since the increase in usage would also lead to an increase in electricity use on times that would otherwise be seen as off peak.

Deliverables

The deliverables of this project will exist of an algorithm that is implanted in an app. This algorithm will be substantiated with documentation in which a literature review is embedded and the capabilities/results of the algorithm will be compared with other energy pricing options.

State-of-the-art

Papers on algorithms for optimal energy consumption.

• Dynamic energy scheduling and routing of a large fleet of electric vehicles using multi-agent reinforcement learning.
• Residential demand response: Dynamic energy management and time-varying electricity pricing.
• Implementation of a dynamic energy management system using real time pricing and local renewable energy generation forecasts.
• Impact of dynamic energy pricing schemes on a novel multi-user home energy management system.

A paper from 2021 analysing many papers on electricity price forcasting sets out to find the state-of-the-art electricity price forecasting models, it describes problems that make comparing of different models hard and also state that there is no clear benchmark to check the performance of models to. The paper states that there are three main models, statistical models, machine learning models and hybrid models. The comparison of these 3 is very hard thus leading to the authors stating not one single state-of-the-art method but choosing multiple. For the statistical models the authors decide that the LEAR model is very accurate, while for the machine learning models the DNN model is most state-of-the-art. The hybrid models they state to be not compared enough to other models thus they decide to leave them out of consideration.  

LEAR stands for Lasso Estimated Auto Regressive, where Lasso is a regression analysis method that performs both variable selection as well as regularization, which is useful to increase the quality of the dataset used for the model. Auto regressive just points to the type of model being based on time series analysis.  

DNN stand for Deep Neural Network, which is a type of machine learning with the objective of trying to replicate the way a human brain thinks. The deep part stands for it being multiple levels of machine learning. These models can be better in analysis and prediction than the statistical models but do use much more computing power.  

Appendix

Planning and logbook

Planning

Week	Day	Date	Occasion	Contents
1	Monday	04-09	Instruction meeting	Group formation, Brainstorm about subject
1	Thursday	07-09	Group meeting	Decide on a subject, divide tasks among group members
1	Sunday	10-09	Deadline	Update the wiki on the first progress
2	Monday	11-09	Feedback meeting	Receive feedback on choice of the subject
2	Thursday*	14-09	Group meeting	Brainstorm about what involved parties to contact
2	Sunday	17-09	Deadline	Finish literature study, Write summaries of the literature study on the wiki
3	Monday	18-09	Feedback meeting	Receive feedback on literature study and choice of involved parties
3	Thursday*	21-09	Group meeting	Start working on the product, contact involved parties
3	Sunday	24-09	Deadline	Finish contacting involved parties, update the wiki on the first design/idea of the product
4	Monday	25-09	Feedback meeting	Receive feedback on contact with involved parties and product
4	Thursday*	28-09	Group meeting	Implement information of the involved parties, work on the product
4	Sunday	01-10	Deadline	Update the wiki on the progress
5	Monday	02-10	Feedback meeting	Recieve feedback about the implementation of the information of the involved parties and current version of the product
5	Thursday*	05-10	Group meeting	Implement the feedback, continue working on the product
5	Sunday	08-10	Deadline	Update the wiki on the progress
6	Monday	09-10	Feedback meeting	Receive feedaback on the progress
6	Thursday*	12-10	Group meeting	Work on the first draft of the final version of the wiki and the product
6	Sunday	15-10	Deadline	Finish the first draft of the final version of the wiki and product
7	Monday	16-10	Feedback meeting	Recive feedback on the drafts
7	Thursday*	19-10	Group meeting	Implement feedback, Check each others work, start working on the presentation
7	Sunday	22-10	Deadline	Finish the final version of the wiki and the product, finish the presentation
8	Monday	23-10	Presentation

Logbook

Week	Name	Break-down of hours	Total hours spent
1	Sven Bendermacher	Searing for ideas (2h), Meeting about subject (1h), Writing deliverables section and mail teachers (0.5h), finding/scanning some promising literature [1-4] (2.5h).	6
	Marijn Bikker
	Jules van Gisteren	Searching for ideas (1.5h), Preparing meeting (0.5h), Meeting about subject (1h), Creating the logbook and planning (2h)	5
	Lin Wolter
2	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
3	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
4	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
5	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
6	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
7	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter
8	Sven Bendermacher
	Marijn Bikker
	Jules van Gisteren
	Lin Wolter

Approach

• Literature study
• Contacting involved parties, interviews