Applications of different machine learning methods for water level predictions

• Main Author: Halla Marinósdóttir 1991-
• Other Authors: Háskólinn í Reykjavík
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: Rekstrarverkfræði; Meistaraprófsritgerðir; Vatnsaflsvirkjanir; Endurnýjanleg orka; Spálíkön; Tækni- og verkfræðideild; Engineering management; Hydroelectric power plants; Renewable energy sources; Predation; School of Science and Engineering; Iceland
• Online Access: http://hdl.handle.net/1946/32366

In recent years, more emphasis has been placed on the utilization of renewable energy sources in electricity generation around the world. One of the main tasks of many nations is to reduce the use of fossil fuels due to their negative impact on the environment. Icelanders have the uniqueness that virtually all of the electricity produced in Iceland can be attributed to renewable energy sources. The greenhouse gases emission from electricity production is therefore lower than in most countries. Despite the uniqueness that exists in Iceland, it is important to maximize the utilization of the resources that are available with the aim of maximizing production. Hydro power is one of two major resources, including geothermal power when it comes to electricity production in Iceland. The flow of water in the power plants in Iceland can be unpredictable and vary widely from seasons. With a good prediction model that predicts the water level of reservoirs, it would be possible to utilize the water better by minimizing the overflow. The power plant could be driven with more power just before the flood, knowing that the water level will reach a balance in a short period of time. In this paper, four different methods were used to predict the water level in Steinsvaðsflóa. The methods were as follows: multiple linear regression, random forest, support vector regression, and artificial neural networks. All forecast models are based on historical data from the Icelandic Meteorological Office and Landsvirkjun. The data contained flow measurements, water levels and various of weather variables from 2009 to 2017. The data was divided into a training set used to set the parameters of the models and the test set used to measure the models performances. The main results were that the forecast models yielded comparable results in terms of error. On the other hand, the main difference between the models was their prediction of the highest and lowest water levels. The model that yielded the lowest error was a random forest model based ...