Reducing Electricity Cost for PV Prosumers by Load Forecast
Name
Artjom Vargunin
Abstract
The energy industry is experiencing nowadays a paradigm shift due to the switching to renewable energy sources, individual electricity storage and production systems, distributed energy production, electric cars, smart devices etc. The geopolitical situation in the world and climate policy force these changes to develop even more rapidly.
The thesis addresses this topic by considering photovoltaic (PV) prosumers, energy consumers able to produce PV energy, store it into the battery and sell back to the grid. It is assumed that such a household uses exchange-price electricity package from grid electricity provider which means that grid electricity price is not fixed, but changes each hour. The electricity market such as Nord Pool announces hourly electricity prices in advance for the next day. Under these constraints the ultimate goal for PV prosumer can be settled as follows: can household achieve zero electricity bill or even earn money? What would be the strategy for the most cost-effective energy usage?
The thesis is aimed to help solving this problem by providing machine-learning driven forecast for the electricity consumption. This prediction is then used to optimize battery usage and calculate electricity cost for PV prosumer. At that, the following requirements are chased: prediction results in the lowest cost, forecast model is good for all households considered. The outcome of the study is well-reasoned suggestion for the forecasting pipeline with statistically proven supremacy.
The thesis addresses this topic by considering photovoltaic (PV) prosumers, energy consumers able to produce PV energy, store it into the battery and sell back to the grid. It is assumed that such a household uses exchange-price electricity package from grid electricity provider which means that grid electricity price is not fixed, but changes each hour. The electricity market such as Nord Pool announces hourly electricity prices in advance for the next day. Under these constraints the ultimate goal for PV prosumer can be settled as follows: can household achieve zero electricity bill or even earn money? What would be the strategy for the most cost-effective energy usage?
The thesis is aimed to help solving this problem by providing machine-learning driven forecast for the electricity consumption. This prediction is then used to optimize battery usage and calculate electricity cost for PV prosumer. At that, the following requirements are chased: prediction results in the lowest cost, forecast model is good for all households considered. The outcome of the study is well-reasoned suggestion for the forecasting pipeline with statistically proven supremacy.
Graduation Thesis language
English
Graduation Thesis type
Master - Data Science
Supervisor(s)
Kristjan Eljand, Novin Shahroudi
Defence year
2023