Institutional Repository of Vilnius University
Efektyvus smulkių saulės elektrinių panaudojimas /
Abstract
dc:descriptionDuring this study, a prototype of the effective use of a small solar power plant was created. The main idea of it was to turn on certain electrical devices when the overflow of the generated energy is forecasted. Devices should not be necessary at the moment but their operation should be beneficial, for example - charging the electrical vehicle. When a solar power plant is generating more energy than the household that is connected to it can utilize, overflow energy is being sent to the main grid of the electricity provider, but when a household is experiencing an energy shortage, the same energy is being bought back at the higher price which means some part of the generated energy is being wasted during this process. During this study, universal microcontroller software was created to collect various information using the Modbus communication protocol from the solar inverter that is connected to the solar power plant. Collected data was gathered from solar inverters and historical weather data API and prepared later on. The study describes the data preparation and feature engineering process that was later used to create various models. Two main parts of the created prototype are different models that forecast the energy generation of certain solar power plants and the energy consumption of a household. Both of these models were tested with different neural network architectures - Feed-forward, Long short-term memory, and Gated recurrent unit networks. All of them were tested with different parameter sets - optimization algorithms, loss functions, learning rates, sequence lengths, number of layers, and number of epochs. To use created models without any additional tools, for example - cloud services, it was tried to move them into a microcontroller. Several tools for this usage were examined - Tensorflow Lite for Microcontrollers, Neural Network on Microcontroller, and Keras2C. The study describes problems that were encountered during the process and how the Keras2C library was adapted to work in a microcontroller. To sum up, the study helped to understand the current situation of using neural networks on microcontrollers and to create a working prototype that could be used in microcontroller based smart switches for efficient energy utilization.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gaidukevičius, Julius,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146236906&prefLang=en_US