Short-Term Wind Speed Forecasting Using Regularization Extreme Learning Machine Da-cheng XING 1, Ben-shuang QIN 1,* and Cheng-gang LI 2

Transcription

1 27 International Conference on Mechanical and Mechatronics Engineering (ICMME 27) ISBN: Short-Term Wind Speed Forecasting Using Regularization Extreme Learning Machine Da-cheng XING, Ben-shuang QIN,* and Cheng-gang LI 2 College of Electrical Engineering, Northeast Dianli University, Jilin 322, Jilin, China 2 Electric Power Research Institute, Jilin Electric Power Company Limited, Changchun 32, China *Corresponding author Keywords: ACF, Wind speed, Short-Term forecasting, R, Regularized. Abstract. High efficient and accurate wind speed prediction is the basis of the wind farm power prediction. So it is helpful for the control of the wind power and has great importance to the parallel operation of the wind farms. Wind speed time series has strong nonlinear and volatility. Besides, it is very difficult to accurately predict. A new method of short-term wind speed forecasting is proposed based on regularized extreme learning machine (regularized extreme learning machine, R). First of all, the autocorrelation function (ACF) is used to analyze the correlation of wind speed time series. After that the number embedded in the time dimension is gotten. And the forecast networ parameters such as inputs, output and so on are determined. That is to say the R model is set up. Then, using the training set trains the networ parameters to get the R prediction model trained. Finally, prediction results are obtained with the test set data. And the wind speed data from the American wind energy technology center is carried out on the experiment. It shows that the new method has better prediction precision compared with the standard and the traditional neural networ. Introduction With the mass consumption of fossil energy and global environmental pressure increasing, the wind energy as a renewable and clean energy brought to the attention of the countries [-3]. Using short-term wind speed forecasting to predict wind farm output and short-term scheduling all have important value. In order to improve the predictive accuracy of wind speed short-term proposed a short-term prediction based on R wind speed method. First of all, according to ACF autocorrelation function of wind speed time series correlation analysis, the former delay time history of wind speed as forecast input; Then, based on R networ short-term wind speed forecasting model is set up; Finally, using the model of short-term wind speed forecasting. In order to verify the effectiveness of the new method, the article using the measured wind speed data to carry out the experiment. Wind Speed Time Series Analysis The wind is in the air pressure imbalance and formation of the air flow. Wind speed, therefore, are susceptible to the influence of weather factors such as temperature, humidity, air pressure, has great randomness and instability [4-5]. By statistical nowledge, the autocorrelation function ACF is commonly used to describe the same sequence and the correlation between the time apart. Existing research does not mature wind speed feature selection method [6], ACF has certain reference value. Autocorrelation function is: ρ γ = () γ 77

2 γ = P ZtZt+ (2) P i= where: Is Since the covariance; P is the time sequence length; Z t and Z t+ are respectively said time away from the two time series. Delay correlation curve ACF Delay time/(min) Figure. ACF of wind speed series. Figure for wind speed sequence of the autocorrelation function of ACF with delay variation. The figure shows that historical data perimeter for predicting time close time correlation data and to predict the wind speed value larger [7]. Regularized Extreme Learning Machine Extreme learning machine () is the single hidden layer feedforward neural networ, which is proposed on the basis of the structure as shown in figure 2. x ω β y x 2 y 2 x 3 y 3 x n ω L β Lm y m Figure 2. The structure of. of objective function solved by networ is: m i i (3) j= E = y t is through the networ looing for the optimal parameters to mae the minimum value of the objective function. min E = min Hβ - T (4) β In algorithm, the input weights and bias in the networ initialization time random assignment is given, and the corresponding output hidden layer for a certain matrix. Therefore, the training of the is the least squares solution of solving the optimal networ parameters of process. 78

3 Hβ- T = min Hβ -T (5) β The output weights matrix is β = H T (6) R In order to enhance the generalization ability of networ, some scholars introduced on the basis of regularization coefficient, build R. R of objective function is λ 2 2 min E = min{ ε + β } (7) β 2 2 The Lagrange equations is λ 2 2 L( αεβ,, ) = ε + β α( Hβ -T - ε ) (8) 2 2 Get the fitting regression model based on R wind speed forecasting is: L y = β g( ω x + b ) (9) i= i i i Build Prediction Model Before the new method, the selected historical moment of wind speed as the input, output to predict time point wind speed, the prediction step way, prediction model is established. As shown in figure 3, forecasting process is as follows: Wind speed history attributes Normalized processing Training data set Test data set R model formed the full-year data Carried out the wind speed forecasting Gain the wind prediction Figure 3. Flowchart of R wind forecasting. Using the data interval for min. Selection and prediction correlation moment wind speed larger before time history of wind speed value as a predictive input, moment to forecast wind speed value as a predictive output, form the training data set. Will stay to predict the wind speed at the corresponding input attributes to have trained model, the output is the moment to predict the wind speed data. Experimental Verification Tae NWTC wind field experiment quarter of 25 a day for short-term wind speed forecasting experiment, and USES the NN model, model and R 3 inds of prediction model to 79

4 forecast model. The prediction of RMSE and MAPE is the results of evaluation index. Experimental results are shown in figure 4, the three methods all can very good forecast wind speed change trend. In figure 4 (a) change in wind speed slow min period of 3-3 (see chart 4) (b) and 7-76 min (see figure 4) (c), three inds of model predicted results were similar; And in the large variations in wind speed stage, R and model can obviously fast tracing out the change of wind speed. Combined with MAPE indicators can be seen in table 2 R and predicted results were similar, but are better than networ R Wind/(m/s) (a) R Wind speed/(m/s) (b) R 9 Wind speed/(m/s) (c) Figure 4. The forecasting results on January 25, 8. Conclusion To improve the short-term wind speed forecasting precision, the paper puts forward a ind of short-term wind speed forecasting method based on R. R, in standard wind speed prediction model, the process of building a model considering empirical ris and structural ris, got a better objective function and improves the generalization ability of short-term wind speed forecasting model. 8

5 Reference [] Xue Yusheng, Yu Chen, Zhao Junhua.et.al. A review on short-term and ultra-short-term wind power prediction[j]. Automation of Electric Power Systems, 25, 39(6):4-5. [2] Gao Yujie, Zhang Yongai, Li Rongrong, et al. Maintenance reserve capacity evaluation and its application considering wind power and maintenance demand uncertainty[j]. High Voltage Apparatus, 25, 5(6):3-38. [3] Peng Zhanggang, Zhou Buxiang, Feng Chao. Analysis of low frequency oscillation for power system with large wind power integration based on PMSG-PSS[J]. High Voltage Apparatus, 26, 52(6): [4] Liu Xingjie, Cen Tianyun, Zheng Wenshu, et al. Neural Networ Wind Speed Prediction Based on Fuzzy Rough Set and Improved Clustering[J]. Proceedings of the Csee, 24, 34(9): [5] Tian Zhongda, Li Shujiang, Wang Yanhong, et al. Short-term wind speed combined prediction for wind farms based on wavelet transform[j]. Transactions of China Electrotechnical Society, 25, 3(9):2-2. [6] Tian Zhongda, Li Shujiang, Wang Yanhong, et al. Chaotic characteristics analysis and prediction for short-term wind speed time series[j]. Acta Physica Sinica, 25, 64(3): [7] Liu Xingjie, Zheng Wenshu. Study on real-time forecasting method of wind speed based on STCP-[J]. Acta Energiae Solaris Sinica, 25, 36(8):