Machine Learning Loss Of Excitation Protection For Synchronous Motor Based On Identification Of Measured Impedance Trajectory

Loss of excitation of all or part of a large synchronous motor is a common and serious fault,which requires more precise operation of loss of excitation protection.Traditional loss of excitation protection for hydroelectric generator based on end-measured impedance static boundary and synchronous condenser based on reactive power and voltage can only judge whether the loss of excitation is based on the final static local information after the fault,and cannot reflect the changes of the measured impedance of various disturbances in the complex grid environment,which is difficult to meet the selectivity and rapidity at the same time.Recently,either some mechanism methods that reflect the change of electrical quantity or some machine learning methods are difficult to adapt to unknown scenarios.In order to improve the generalization ability of machine learning loss of excitation protection,a new scheme of loss of excitation protection for synchronous motor based on multiple kernel support vector machine(MKL-SVM)for measuring impedance trajectory identification was proposed in this paper.Firstly,feature extraction was performed from the angle of the measured impedance trajectory.Typical electrical quantities and measured impedance trajectories at the machine end after loss of excitation of hydro-generators and synchronous condensers were analyzed theoretically,and the shortcomings of traditional loss of excitation protection in engineering application were pointed out through performance analysis.In view of the difference of the dynamic time sequence motion characteristics of the impedance trajectory measured at the machine end in the fixed time window,the track features of the hydro-generator and synchronous condenser were extracted respectively.For the former,the statistical idea was introduced to describe the distribution of the time-series characteristics,and minimal-redundancy-maximal-relevance criterion(mRMR)based on mutual information was used to sort the features to remove the feature redundancy,while the Wrapper strategy was used to determine the final input features;For the latter,the trajectory features with strong interpretation were manually extracted from the statistical point of view.Secondly,MKL-SVM was selected as the classifier to form the loss of excitation protection of the synchronous motor.The good fitting of local RBF kernel function and the good extrapolation characteristics of global Poly kernel function were briefly described.Then the kernel function was formed by the linear weighted combination of the two to train MKL-SVM to take both global and local features into account and to improve the generalization ability of the model.An adaptive genetic algorithm with elitist preservation criteria(IGA)was proposed to optimize the model parameters.On this basis,considering the influence of the severity of the loss of excitation fault of the synchronous motor on the measured impedance trajectory,a two-time window identification principle based on the distance of the classification function was proposed to improve the reliability of the loss of excitation protection.Finally,the simplified and extended systems of the hydro-generator and synchronous condenser were simulated and verified respectively.The results show that the protection scheme improves the speed while ensuring selectivity,and still has excellent adaptability in the face of complex changes in the power grid.