Simulation and Forecasting for Wind Speed Modelling


Student thesis: Doctoral Thesis

View graph of relations


  • Jie TANG


Awarding Institution
Award date26 May 2016


Traditional power plants, which consume enormous amounts of fossil fuel have been known to be associated with serious health and environment risks. Therefore, more attention is being paid to renewable energy, among which wind energy is becoming the mainstream and reliable power form. One of the most important fields of wind technology research and development is wind simulation, which is very useful in choosing the best turbines for certain locations and optimising turbine design for the operations. Improving the accuracy of wind forecast is also a key field of wind technology development that is crucial for the operation of wind power plants, especially for electricity markets and the power system. This study focuses on the analysis, improvement and comparison of wind speed simulation and forecast methods.
First, this study introduced two improvements to the traditional Markov chain method for wind speed modelling. The first one is a new state categorization procedure that takes advantage of the empirical probability distribution function of the wind speeds time series. The second improvement relies directly on the empirical distribution of the wind speeds in each state. This thesis came up with a new simulation method to generate synthetic wind speeds from the state simulations. The applications and comparison conducted reveal that the improved method over-performs its traditional counterpart in term of simulation, and performs comparably with the traditional one in term of forecasting.
Second, this study constructed a simplified indexed semi-Markov model, which is designed to effectively preserve the autocorrelation feature of wind speeds with the aid of a new memory index. It simplifies the indexed semi-Markov model, which requires the generation of a four-dimensional matrix and the implementation of two sampling procedures. The simplified method is highly efficient in application, especially when the data involved are numerous and undergo many states. The comparative results indicate that the simplified method effectively preserves both the autocorrelation feature and the probability density function of wind speeds, with a much simpler structure than that in the indexed semi-Markov model. Further, it is shown that the simplified model is comparable with the indexed semi-Markov model with respect to forecasting performance.
Finally, comparisons among commonly used simulating and forecasting methods of wind speeds are given separately in this study. As for simulation comparisons, statistical features such as autocorrelation function (ACF), probability density function (PDF), mean and variance were examined whenever the second-order autoregression method—AR(2)—the simplified indexed semi-Markov chain method (SIM) and the Markov chain method (MC) are conducted. In terms of forecast comparison, on the other hand, root mean square error (RMSE) between data forecast and the observed were checked using the autoregression integrated moving average (ARIMA) method, exponential smoothing (ES) method and Markov chain method (MC). The comparison results show that: for simulation, SIM performs pretty well, in reproducing both the ACF and the PDF of the observed time series with different time scales. MC is comparable to SIM with respect to its ability to capture PDF features. Our findings also show that performances of ARIMA, single ES and MC are better in short- and medium-forecasting than in longer-term forecasting.