M H Refan, A Dameshghi, M Kamarzarrin,
Volume 9, Issue 4 (December 2013)
Abstract
Differential base station sometimes is not capable of sending correction information for minutes, due to radio interference or loss of signals. To overcome the degradation caused by the loss of Differential Global Positioning System (DGPS) Pseudo-Range Correction (PRC), predictions of PRC is possible. In this paper, the Support Vector Machine (SVM) and Genetic Algorithms (GAs) will be incorporated for predicting DGPS PRC information. The Genetic Algorithm is employed to feature subset selection. Online training for real-time prediction of the PRC enhances the continuity of service on the differential correction signals and therefore improves the positioning accuracy in Real Time DGPS. Given a set of data received from low cost GPS module, the GASVM can predict the PRC precisely when the PRC signal is lost for a short period of time. This method which is introduced for the first time for prediction of PRC is compared to other recently published methods. The experiments show that the total RMS prediction error of GASVM is less than 0.06m for on step and 0.16m for 10 second ahead cases
M. Kamarzarrin, M. H. Refan, P. Amiri, A. Dameshghi,
Volume 18, Issue 2 (June 2022)
Abstract
One of the major faults in Doubly-Fed Induction Generator (DFIG) is the Inter-Turn Short Circuit (ITSC) fault. This fault leads to an asymmetry between phases and causes problems to the normal state between current lines. Faults diagnosis from non-stationary signals for the Wind Turbine (WT) is difficult. Therefore, the strategy of fault diagnosis must be robust against instability. In this paper, a new intelligent strategy based on multi-level fusion is proposed for diagnosis of DFIG inter-turn stator winding fault. Firstly, to overcome the non-stationary nature of the vibration signals of the WT, empirical mode decomposition (EMD) method is performed in time-frequency domains to extract best fault features from information power sensor and information current sensor. Moreover, a feature evaluation technique is used for the input of the classifier to choose the best subset features. Secondly, Least Squares Wavelet Support Vector Machines (LS-WSVM) classifier is trained to classify fault types based on feature level fusion (FLF) from different sensors. The main parameters of SVM and the kernel function are optimized by Genetic Algorithm (GA). Finally, Dempster-Shafer evidential reasoning (DSER) is used for fusing the GA-LS-WSVM results based on decision level fusion (DLF) of individual classifiers. In order to evaluate the proposed strategy, a DFIG WT test rig is developed. The experimental results show the efficiency of the proposed structure compared to other ITSC fault diagnosis methods. The results show that the classification accuracy of DSER-GA-LS-WSVM is 98.27%.
