Adama II Wind Farm Long-Term Power Generation Forecasting Based on Machine Learning Models
No Thumbnail Available
Date
2022-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Currently, renewable energy production from wind farms with hundreds of megawatts connected to the grid is increasing. Wind energy is intermittent and random due to being highly dependent on wind speed nature, hence grid interconnection among nations of the Eastern Africa region plans face reliability issues. Accordingly, Ethiopian Electric Power needs to diligently plan ahead of time the allocation of generating units in its power plants to match its national and regional energy demand (MW) because if the demand is higher than the generation it can cause several blackouts resulting in a huge loss to the economy; on the other hand, if the generation is higher than the demand the extra electricity will be wasted and it can also create an unnecessary load on the transmission lines.
This thesis studies the power production performance analysis of the Adama II wind farm using MATLAB SIMULINK with scenarios of fault ride-through capability, short circuit fault, control fault, and wind speed variation impact. Furthermore, conducting long-term wind power forecasting to safely national and regional grid integration by applying basic time series models SARIMA and then extended to linear regression, random forests, and XGBoost to accurately forecast the Power output of the wind turbine.
Impact and performance analysis result has observed a short circuit and control fault on the grid interconnection point occur will bring a grid disturbance to the power system. But a rapid speed variation will be composited by the farm’s reactive power supply. Moreover, Adama II wind farm delivery 0MW upto 153MW power to gird, during maintenance and some grid and turbine faults it failed to delivery as its capacity. Main reason to downtime is due to grid fault, night time due to light load at night Adama II wind farm don’t have voltage regulator to cop up its power fluctuation. Furthermore, the thesis developed a one-year ahead forecasting model to improve the Adama II wind power plant grid integration impact for reliable plant operation and maintenance schedule. The study tests 1 hour, 1 week, and 12 months. 1-hour ahead and 1-week ahead forecasting using SARIMAX, Elastic net, and Random Forest regression give around 90% R2 score and 2~4% MAPE using lag variables for short-term 1-hour ahead and 1-week ahead forecasts. But for more than 1 month ahead forecasting XGBoost (with Fourier terms for seasonality) performs very well for longer forecast windows.
Description
Keywords
Variable Renewable Energy, Wind Farm, Performance, Long-Term Forecasting, grid integration.