1. Key Laboratory of Power System Intelligent Dispatch and Control of Ministry of Education (Shandong University), Jinan 250061, China; 2. State Grid Jiangsu Electric Power Co. Ltd., Nanjing 210024, China

With integration of high voltage direct current(HVDC)transmission with large capacity and long distance and large scale of renewable generation, the risk of frequency security at the receiving-end of power systems is rising. Aiming at severe disturbances such as large-scale HVDC blocking which may trigger under-frequency load shedding(UFLS), a method for fast estimation of maximum transient frequency deviation is proposed based on machine learning. The problem is decomposed into two sub-problems i. e. UFLS response judgment and maximum frequency deviation estimation, respectively. The maximum transient frequency deviation is estimated by solving the sub-models alternately. Support vector regression method is used to establish the sub-model of maximum frequency deviation estimation, the Bagging ensemble learning method based on support vector machine is used to establish the sub-model of UFLS response judgement. Operation condition and disturbance information are regarded as inputs. ReliefF method and principal component analysis are introduced to select and extract input features to reduce the model complexity. A receiving-end power system with multiple HVDC links is taken as an example to build a maximum transient frequency deviation model and verify the accuracy and adaptability of the proposed method.