Mode Decomposition Support Vector Research Articles

An EMD-SVM model with error compensation for short-term wind speed forecasting

In this paper, we propose an empirical mode decomposition-support vector machine (EMD-SVM) model with error compensation in order to reduce the cumulative error and improve the prediction accuracy of short-term wind speed forecasting. The essential idea behind the proposed approach is that the error of the current prediction is highly correlated with the previous prediction errors, and the forecasted speed should be compensated in terms of the errors incurred from previous predictions. Specifically, we first predict the historical data by the EMD-SVM model so as to obtain the corresponding prediction errors. Then, we establish the error compensation mechanism. Finally, we combine the EMD-SVM model with error compensation to obtain the final prediction results. The error compensation strategy is validated by a series of actual 10 min wind speed data collected from New Zealand. Experimental results demonstrate that the proposed EMD-SVM model with error compensation can be successfully applied to short-term wind speed forecasting, and it has higher accuracy and stronger robustness compared with the method without error compensation.

Study of Models to Forecast the Radio-electric Spectrum Occupancy

Objectives: The analysis of spatial opportunities to reuse frequencies by secondary users (SU) in a Cognitive Radio (CR) network is the main objective of this work. Methods/Statistical Analysis: Here lineal and no-lineal models are developed and evaluated to forecast the received power of different channels base on measurements performed in Bogota Colombia for the global system for mobile communication (GSM) bands. Seasonal autoregressive integrated moving average (SARIMA), generalized autoregressive conditional heteroskedastic (GARCH), Markov, empirical mode decompositionsupport vector regression (EMD-SVR) and wavelet neural models were utilized for the forecasting of the channel occupancy. Findings: The analysis performed shows that the wavelet neural model presents less error for the received power forecast compared to the other models analyzed and developed in this work. This is in particular for the different types of mean absolute error evaluated. In addition, the accuracy percentage reach by the wavelet neural model is greater than 95% for the forecast of the availability and occupancy times of channels. Application/Improvements: Accordingly, this research indicates that wavelet neural model depicts a hopefuloption to CR systems, in order to forecast received power for the detection of spectral opportunities.