Evaluation of Deep Learning-based prediction models in Microgrids

Györi, Alexey; Niederau, Mathis; Zeller, Violett; Stich, Volker

doi:10.1109/cencon47160.2019.8974844

Cited by 2 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [28], demand, generation and battery state of charge of a microgrid are forecast with ARIMA, using only past data of the variable to be forecast among photovoltaic (PV) power production, load consumption and battery state of charge in a building, and a control algorithm is applied afterwards. However, in [29,30], which deals with microgrid demand, it is argued that ANNs are more appropriate for time series forecasting than ARIMA, especially in cases where the signals exhibit abrupt fluctuations and nonlinearity. In [30] especially, season (with arbitrary numbering) and time themselves instead of demand average over season/time have been inserted as input to the model, thus leading to non-linear instead of linear relations between input and output, in contrast to our approach.…”

Section: Related Work and Our Contributionmentioning

confidence: 99%

Multi-Step Energy Demand and Generation Forecasting with Confidence Used for Specification-Free Aggregate Demand Optimization

2021

View full text Add to dashboard Cite

Energy demand and generation are common variables that need to be forecast in recent years, due to the necessity for energy self-consumption via storage and Demand Side Management. This work studies multi-step time series forecasting models for energy with confidence intervals for each time point, accompanied by a demand optimization algorithm, for energy management in partly or completely isolated islands. Particularly, the forecasting is performed via numerous traditional and contemporary machine learning regression models, which receive as input past energy data and weather forecasts. During pre-processing, the historical data are grouped into sets of months and days of week based on clustering models, and a separate regression model is automatically selected for each of them, as well as for each forecasting horizon. Furthermore, the multi-criteria optimization algorithm is implemented for demand scheduling with load shifting, assuming that, at each time point, demand is within its confidence interval resulting from the forecasting algorithm. Both clustering and multiple model training proved to be beneficial to forecasting compared to traditional training. The Normalized Root Mean Square Error of the forecasting models ranged approximately from 0.17 to 0.71, depending on the forecasting difficulty. It also appeared that the optimization algorithm can simultaneously increase renewable penetration and achieve load peak shaving, while also saving consumption cost in one of the tested islands. The global improvement estimation of the optimization algorithm ranged approximately from 5% to 38%, depending on the flexibility of the demand patterns.

show abstract