| Title |
Application of AI-based methods for electrical load forecasting in power systems |
| Translation of Title |
Dirbtiniu intelektu grįstų metodų taikymas apkrovų prognozavimo uždaviniuose elektros energetikos sistemose. |
| Authors |
Morkūnas, Dovydas |
| Full Text |
|
| Pages |
64 |
| Keywords [eng] |
artificial intelligence ; clustering ; forecasting model ; electrical load forecasting |
| Abstract [eng] |
This project focuses on application of AI-based methods for forecasting electrical load. Hourly electrical load data from the year 2020 of 176 users connected to the Lithuanian electrical power distribution grid was used to create forecasting models. Dynamic time warping and Euclidean distance matrices, together with k-means and hierarchal clustering methods were used for clustering. A filter was implemented to filter out large irregularities. Forecasting model structure was created, aiming to forecast 24 hours into the future. A linear regression model was applied. TensorFlow was used for creation of forecasting models using temporal convolutional networks and feed-forward neural networks. Statistical model evaluation metrics were calculated to determine performance of the models. Results indicate that AI-based load forecasting models did not outperform linear regression models. The best performing model was found to be a linear model using DTW distance matrix and k-means clustering (R2 = 0,5438). The worst performing model was found to be a linear model using Euclidean distance matrix and k-means clustering (R2 = 0,3601). |
| Dissertation Institution |
Kauno technologijos universitetas. |
| Type |
Master thesis |
| Language |
English |
| Publication date |
2023 |
