Development of deep learning based methods for short-term wind speed forecasting for Meiktila in Myanmar | |
| Author | Kaung Myat San |
| Call Number | AIT Thesis no.ET-20-01 |
| Subject(s) | Wind power--Forecasting Machine learning--Development |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Energy, School of Environment, Resources and Development |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. ET-20-01 |
| Abstract | Energy demand has become very vital for the world and it has been increasing because of world economic development. Therefore, wind power is popular and is extensively used and has been growing quickly in recent years because one of the hopeful clean renewable and green resources and also helps solve environmental issues. However, accurate wind speed prediction is required for wind power generation to promote the reliability and performance of wind turbine and power systems because the wind speed is intermittent, and it is difficult to be forecasted. Therefore, this thesis proposes two signal decomposition methods (Wavelet Packet Decomposition-WPD and Butterworth filter) and three deep neural networks (CNN (Convolutional Neural Network), LSTM (Long Short-term Memory) and hybrid CNNLSTM (hybrid model)) for short-term prediction. Moreover, this study describes the performance of individual deep neural networks and that of deep neural networks combined with signal decomposition procedures. Thus, this study has developed three deep neural models without signal decomposition methods and two deep neural networks with signal decomposition methods for one step ahead of short-term wind speed prediction. They are CNN, LSTM, CNN-LSTM, WPD-CNN and Butterworth-CNN. The historical data of wind speed, temperature and relative humidity is collected based on five-time of a daily basis (6:30, 9:30, 12:30, 15:30 and 18:30) from the metrological stations of Mandalay and Meiktila, Myanmar for 5 years and 11 months. This study applies five performance criteria methods for accurate measurement. They are MAE (Mean Absolute Error), MSE (Mean Square Error), RMSE (Root Mean Square Error), correlation and regression. The promoting percentage of error (PMAE, PMSE, and PRMSE) of the experimental tests are applied in order to compare the performance of individual deep neural networks (CNN, LSTM and CNN-LSTM) and that of deep neural networks combined with signal decomposition procedures (WPD-CNN and Butterworth-CNN). MSE is defined as a loss when the training process. The key contribution to this study is that it proposed two prediction methods combined with signal decomposition methods (WPD-CNN and Butterworth-CNN) and it compared the performance results of CNN, LSTM, CNN-LSTM and WPD-CNN, ButterworthCNN. This study illustrates five forecasting models that utilize input data from January 2014 to November 2019 (10800 samples). After evaluating the performance of five models, WPD-CNN has better than the other four models and WPD-CNN outperformed the other four models. |
| Year | 2020 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ;|vno. ET-20-01 |
| Type | Thesis |
| School | School of Environment, Resources, and Development (SERD) |
| Department | Department of Energy and Climate Change (Former title: Department of Energy, Environment, and Climate Change (DEECC)) |
| Academic Program/FoS | Energy Technology (ET) |
| Chairperson(s) | Singh, Jai Govind; |
| Examination Committee(s) | Salam, P. Abdul;Weerakorn Ongsakul; |
| Scholarship Donor(s) | Loom Nam Khong Pijai (Greater Mekong Subregion) Scholarships; |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2020 |