class: center, middle, inverse, title-slide # A Short Review on Predictions for Wind Power Generation ## (Limitation of point estimates and future directions) ### Sim, Min Kyu & Jung, Jae-Yoon ### ICICIC2019 Conference ### August 28, 2019 --- # Contents 1. Background 1. Literature Classifications 1. Goal 1. Data & Method 1. Results 1. Conclusion --- # 1. Background + Wind power generation + 4.4% of global electricity generation in 2017 [1] + 14% of European Union’s electricity generation in 2018 [2] + Facts + Importance and popularity are continuously growing. + Controlability is a desirable element, but humans cannot. + Power amount depends on various climate factors including... + Wind speed + Temperature + Atmosphere pressure + Height of power plant + Many studies have conducted prediction studies using these predictors. --- # 2. Literature Classifications -- ## Criteria 1 - by Target Variable + Wind Speed + Useful for power plant operations such as turbine configuration and maintenance. + Wind Power Generation + Take up the larger portion [3] + More directly beneficial to electricity production. --- ## Criteria 2 - by Time Horizon + According to a review paper [4] + Very short-term (-30min) + Short-term (30min - 6h) + Medium-term (6-24h) + Long-term (24-72h) + Very long-term (72 h-) + Shorter time horizon + Beneficial for immediate operations + Turbine configuration + Medium time horizon + Production planning + Electricity trading + Longer time horizon + Plant installation planning + Maintenance scheduling --- ## Criteria 3 - by Approach + Physical + Weather agencies routinely adopt physical models + Generate Numerical Weather Predictions (NWPs) + NWPs on the speed and direction of wind + Statistical/Hybrid + Apply statistical forecasting methods + Use NWPs and historical wind power generation + Specific methods include... + Conventional time-series: AR, VAR, ARMA, and ARIMA + Statistical learning: SVM, boosted tree, random forest, k-NN + Neural-nets: ANN, MLP, RNN + A survey paper [4] provides an extensive lists. --- # 3. About this study + Apply popular prediction methods (mostly statistical learning) + Apply them on the actual dataset of NWPs and historical power generation records. + From the experiment results, identify the current challenges. + Suggest directions and requirements of future studies. --- # 4. Data & Methods -- ## Dataset + Han-Kyung Wind Power Plant + built in 2004 in the Jeju Island, Korea. + Four units, each 1500 KWH. + Hourly power generation data from 2014 to 2017 is provided by the operating company (Korea Southern Power Co. Ltd.) + Historical NWPs that include various climate variables. + Data from nearby weather stations are collected for the same period (2014-2017) + Power generation is determined by following. `$$P = 0.5 k C_p \rho A V^3,$$` where `\(P\)` is the amount of power output, `\(k\)` is a unit conversion constant, `\(C_p\)` is a dimensionless coefficient for the maximum power, `\(\rho\)` is air density, `\(A\)` is rotor swept area, and `\(V\)` is a wind speed. This study merges the two datasets and sets up a 4-hour ahead prediction problem. + Thus, wind speed, temperature, and humidity are used from NWPs. --- ## Utilized Methods + ANN + RF + Adaboost + Linear Reg. + SVR + Decision Tree + k-NN ## Performance Criteria + MSE (RMSE) + MAE + `\(R^2\)` --- # 5. Results #### **Table 1. Performance comparison** | Method | `\(R^2\)` | RMSE | MAE | Maximum Error | |:-------:|----:|------:|----:|--------------:| | ANN | 86.3 | 200.6 | 133.2 | 1,194 | | RF | 86.2 | 200.6 | 131.3 | 1,197 | | Adaboost | 86 | 207.8 | 150.3 | 1,156 | | Linear Reg. | 85.7 | 204.8 | 136.9 | 1,168 | | SVR | 85 | 211.6 | 133 | 1,273 | | Decision Tree | 84.5 | 212.2 | 139.9 | 1,263 | | k-NN | 82.8 | 225.2 | 151.6 | 1,347 | > Not large differences between methods. ---
#### **Figure 1. Forecast errors** > All methods generate very similar forecasts. --- #### **Table 2. Cross-Correlation** | | Lin. Reg | RF | SVR | KNN | Dec.Tree | Adaboost | ANN | |:------:|:--------:|:---:|:---:|:---:|:--------:|:--------:|:---:| | Lin. Reg | 1 | 0.95 | 0.94 | 0.90 | 0.91 | 0.93 | 0.99 | |RF | |1 |0.95 |0.91 |0.95 |0.95 |0.96| |SVR | | |1 |0.93 |0.91 |0.89 |0.96| |KNN | | | |1 |0.88 |0.89 |0.92| |Dec.Tree | | | | |1| 0.93| 0.92| |Adaboost | | | | | | 1 |0.93| |ANN | | | | | | | 1| |<img width=100/>|<img width=80/>|<img width=80/>|<img width=80/>|<img width=80/>|<img width=80/>|<img width=80/>|<img width=80/>| > Predicted values from different methods are highly correlated to each other. --- # 6. Conclusion ## Summary + The experiment utilized widely used data sources. + The results of prediction are very similar even though the different algorithms are applied. + Unless the scope of dataset is significantly widened, it is difficult to enhance prediction accuracy. + This striking similarity indicates an inherent limitation of the predictability of wind power generation. + Presumably, point estimation techniques for wind power generation may have reached its limit of improvement. --- ## Current Trends Due to the limitation of point estimations, recent studies have focused on... 1. Study on wind ramps + Investigate wind ramp (sudden increases/decreases in the wind speed) that degrade the accuracy of predictions [11] + Wind ramp is generally defined in a binary form and efforts are made to identify climate conditions that cause ramps. + (Cons) The binary definition can be easily understood, mixing it up with parametric information may be tricky. + (Cons) The results tend to be qualitative rather than quantitative. 2. Probabilistic forecasts + Generate the interval estimate of power generation instead of point estimates [12]. + Apply statistical methods such as quantile regression and bootstrapping methods. --- ## Future directions + Identify the conditions where the error of point estimate becomes large or small. + Utilize both NWPs and recent power generation when producing interval estimates. + Generates interval estimates also using the findings from studies on wind ramp. + Generate interval estimates according to different confidence levels. + Interval estimates beneficial to decision makers more than are point estimates. + Nonparametric approaches are preferred. --- # References + [1] BP statistical review of world energy 2018, BP p.l.c., London, UK, 67th Ed., Jun 2018. [Online]. Available: https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2018-full-report.pdf + [2] C. Walsh and I. Pineda, Wind energy in Europe in 2018 - Trends and statistics, Wind Europe, Brusselsm Belgium, Feb 2019. [Online]. Available: https://windeurope.org/wp-content/uploads/files/about-wind/statistics/WindEurope-Annual-Statistics-2018.pdf + [3] I. Colak, S. Sagiroglu and M. Yesilbudak, Data mining and wind power prediction: A literature review, Renewable Energy, vol. 42, pp. 241-247, 2012, DOI: `10.1016/j.renene.2012.02.015` + [4] I. Okumus and A. Dinler, Current status of wind energy forecasting and a hybrid method for hourly predictions, Energy Conversion and Management, vol. 123, pp. 362-371, 2016, DOI: `10.1016/j.enconman.2016.06.053` + [5] C. Croonenbroeck and D. Ambach, A selection of time series models for short-to medium-term wind power forecasting. Journal of Wind Engineering and Industrial Aerodynamics, vol. 136, pp. 201-210, 2015, DOI: `10.1016/j.jweia.2014.11.014` + [6] T. Senjyu, A. Yona, N. Urasaki and T. Funabashi, Application of recurrent neural network to long-term-ahead generating power forecasting for wind power generator. Proc. IEEE PES Power Systems Conference and Exposition, Atlanta, GA, USA, pp. 1260-1265, 2006, DOI: `10.1109/PSCE.2006.296487` --- + [7] J. Wang and J. Hu, A robust combination approach for short-term wind speed forecasting and analysis-Combination of the ARIMA (Autoregressive Integrated Moving Average), ELM (Extreme Learning Machine), SVM (Support Vector Machine) and LSSVM (Least Square SVM) forecasts using a GPR (Gaussian Process Regression) model. Energy, vol. 93, pp. 41-56, 2015, DOI: `10.1016/j.energy.2015.08.045` + [8] A. Kusiak and Z. Zhang, Short-horizon prediction of wind power: A data-driven approach. IEEE Transactions on Energy Conversion, vol. 25, no. 4, pp. 1112-1122, 2010, DOI: `10.1109/TEC.2010.2043436` + [9] A. M. Foley, P. G. Leahy, A. Marvuglia and E. J. McKeogh, Current methods and advances in forecasting of wind power generation. Renewable Energy, vol. 37, no. 1, pp. 1-8, 2012, DOI: `10.1016/j.renene.2011.05.033` + [10] T.J. Hastie, Ch.7 Generalized Additive Models, In Statistical Models in S, Routledge, pp. 249-307, 2017, ISBN-13: 978-0412830402 + [11] C. Gallego-Castillo, A. Cuerva-Tejero and O. Lopez-Garcia, A review on the recent history of wind power ramp forecasting. Renewable and Sustainable Energy Reviews, vol. 52, pp. 1148-1157, 2015, DOI: `10.1016/j.rser.2015.07.154` + [12] A. Khosravi, S. Nahavandi, and D. Creighton, Prediction intervals for short-term wind farm power generation forecasts, IEEE Transactions on sustainable energy, vol. 4, no. 3, pp. 602-610, 2013, DOI: `10.1109/TSTE.2012.2232944` --- background-image: url('data/powerplant.jpg') background-position: 50% 50% background-size: cover > Thank you for listening