Optimizacija cene električne energije pametne hiše z uporabo spodbujevalnega učenja

SIVEC, ANDRAŽ (author), Meža, Marko (mentor)

Zaradi prehoda v brezogljično družbo močno raste število malih sončnih elektrarn, ki jih nameščajo gospodinjstva. Zaradi nihanja v proizvodnji sončnih elektrarn se pogosto nameščajo tudi hranilniki energije, da se generirana energija lahko porabi tudi v času, ko sončna elektrarna ne deluje. Za razliko od klasičnih hiš, ki so samo porabniki, ima hiša s sončnimi paneli in hranilnikom energije mnogo več nadzora nad tem, kdaj kupuje energijo in kdaj odvečno energijo prodaja. To bo posebej pomembno v prihodnje, ko bo zaradi porasta sončnih elektrarn cena energije v času dneva, ko proizvajajo največ, predvidoma dodatno padla. V magistrski nalogi sem zgradil digitalni dvojček hiše z nameščenimi sončnimi paneli in hranilnikom energije. Dvojčka sem uporabil kot okolje za gradnjo agenta s spodbujevalnim učenjem. Uporabil sem 2 pristopa gradnje agenta, njune rezultate in zahtevnost delovanja sem primerjal med sabo in razložil prednosti in slabosti posameznega modela.

• Language: Slovenian
• Keywords: : strojno učenje, spodbujevalno učenje, pametna hiša, trgovanje z energijo, sončna elektrarna, baterija, zeleni prehod, elektrifikacija
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FE - Faculty of Electrical Engineering
• Year: 2024
• PID: 20.500.12556/RUL-161569
• COBISS.SI-ID: 215614723
• Publication date in RUL: 12.09.2024

Secondary language

• Language: English
• Title: Optimizing the electricity price of a smart home using reinforcement learning

Abstract

Due to the transition to a carbon-free society, the number of small solar power plants installed by households is growing rapidly. Due to fluctuations in the production of solar power plants, energy storage systems are often installed so that the generated energy can also be used when the solar power plant is not operating. Unlike traditional houses, which are only consumers, a house with solar panels and an energy storage system has much more control over when it buys energy and when it sells excess energy. This will be especially important in the future, as the price of energy is expected to further decrease during the day when solar power plants produce the most. In my master’s thesis, I built a digital twin of a house with installed solar panels and an energy storage system. Using this twin, I developed two principles of reinforcement learning (RL) operation. I compared their results and operational complexity and explained the advantages and disadvantages of each model.

• Keywords: machine learning, reinforcement learning, smart home, energy trading, solar power plant, battery, green transition, electrification