A Comparative Study on Time Series Prediction of Photovoltaic-Power Production Through Classic Statistical Techniques and Short-Term Memory Networks
The inherent variability in the power production of renewable energy sources (RES) limits the effectiveness of energy management systems (EMS) since optimal dispatch on power networks highly depends on the accuracy of predictors associated with the energy output and load demand. Consequently, power...
| Main Authors: | , |
|---|---|
| Format: | ARTÍCULO DE CONFERENCIA |
| Language: | es_ES |
| Published: |
IEEE
2024
|
| Subjects: | |
| Online Access: | http://dspace.ucuenca.edu.ec/handle/123456789/44125 https://www.scopus.com/inward/record.uri?eid=2-s2.0-85177461803&doi=10.1109%2fCoDIT58514.2023.10284303&partnerID=40&md5=2735e7893af85e561ae3e1df8573673a |
| Summary: | The inherent variability in the power production of renewable energy sources (RES) limits the effectiveness of energy management systems (EMS) since optimal dispatch on power networks highly depends on the accuracy of predictors associated with the energy output and load demand. Consequently, power prediction tools for variable time horizons allow for improving energy management decisions. In this context, this work presents a detailed methodology for the deployment of predictive models for the photovoltaic (PV) power output of a small solar farm. The prediction models process a PV power dataset's time series using statistical techniques and neural networks with long-short term memory (LSTM). Before the data fitting, we develop a data preprocessing system, which involves evaluating missing data in the time series and getting descriptive analysis of the data set to either complete portions or delete atypical data. The results strongly suggest that the LSTM network performs better than the statistical model in exchange for more considerable computation times for long-term predictions |
|---|