Artificial Intelligence Forecasting Techniques For Reducing Uncertainties In Renewable Energy Applications

Kivonat, rövid leírás

The work presented in this thesis provides an integrated view and related insights for solar and wind farm operators and renewable energy regulators regarding factors influencing electricity production using those resources. The findings help production planning and grid stability improvements through better energy forecasting to reduce uncertainty. With the high increase in energy demand expected in both the near and far future, generating energy from green sustainable resources has now become an imperative necessity. Renewable energy sources like wind and solar are among the most promising and environmentally-friendly energy generation options. However, wind and solar energy production are influenced by many variables which affect the reliability, stability, and economic benefits of wind and solar energy projects, therefore, forecasting the potential amount of energy from wind and solar resources is of great importance. Hence, the objective of the work reported here was to explore the possibility of using artificial intelligence methods to accurately predict the generated renewable power from solar and wind farms based on the available data. Specifically, this thesis reports on the following results: 1. At first, solar photovoltaic (PV) energy forecasting was studied. Operators of grid-connected PV farms do not always have full sets of data available to them, especially over an extended period of time as required by key forecasting techniques such as multiple regression (MR) or artificial neural network (ANN). Therefore, the work reported here considered these two main approaches of building prediction models and compared their performance when utilizing structural, time-series, and hybrid methods for data input. Three years of PV power generation data (of an actual farm), as well as historical weather data (of the same location) with several key variables, were collected and utilized to build and test six prediction models. Models were built and designed to forecast the PV power for a 24-hour ahead horizon with 15 minutes resolutions. Results of comparative performance analysis show that different models have different prediction accuracy depending on the input method used to build the model: ANN models perform better than the MR regardless of the input method used. The hybrid input method results in better prediction accuracy for both MR and ANN techniques while using the time-series method results in the least accurate forecasting models. Furthermore, sensitivity analysis shows that poor data quality does impact forecasting accuracy negatively, especially for the structural approach. 2. Then, wind energy forecasting was studied utilizing three machine learning techniques namely Artificial Neural Network (ANN), Support vector machine (SVM), and k-nearest neighbors (K-NN). The three mentioned techniques were used to design, train and test wind energy, forecasting models. Later, a hybrid model was proposed based on these three techniques. Real data obtained from a 2MW grid-connected wind turbine has been used to train and validate the different machine learning techniques. To compare the accuracy of the models over different performance measures with different scales, a comparative evaluation method was devised and used. Results show that the ANN model has great performance in forecasting long-term wind energy, but in contrast, it has very poor short-term performance. SVM model shows better short-term forecasting performance than ANN but presents weak long-term forecasting abilities. K-NN model shows very good short-term forecasting abilities and fair long-term performance. The suggested hybrid model was able to forecast both long and short-term wind energy with very good performance. To that degree, the suggested model can help grid and wind farm operators to forecast the potential amount of wind energy for both long and short term with a good degree of certainty. 3. Finally, the effect of the input data resolution on the forecasting accuracy was studied for both wind and solar. So, datasets were collected from a 546 KWp grid-connected PV farm and a 2 MW wind turbine for one full year. This data was used to train and test Artificial Neural Network models to forecast day-ahead PV and wind energy utilizing time-series input data with 15, 30, and 60 minutes resolutions. The models were able to forecast the PV energy accurately, while the same models trained for wind showed poor performance. Higher input data resolutions lead to slightly better forecasting performance for the PV farm. Utilizing data with higher resolution can improve the forecast by 1-5%. While for wind energy forecasting, the resolution has very minor effects, the 30-minute resolution shows a bit better forecasting performance.