Power System Inertia Estimation Using A Residual Neural Network Based Approach

The increasing penetration of non-synchronous generation into power grids is reducing the equivalent system inertia and leading to different frequency regulation and control challenges. Consequently, the monitoring and quantification of this inertia to implement actions that can keep it above critical levels have become a key issue for the stability of power systems. In this regard, a residual neural network (ResNet) based alternative is proposed and investigated in this paper to estimate the equivalent inertia of a sample system when synchronous generating units are displaced by converter-interfaced generators. The proposed ResNet model is trained according to the frequency of the center of inertia and the corresponding computed rates of change of frequency for a predefined time interval, where sudden generation outages and load step changes are considered under variations of total load demand and equivalent inertia reductions. The accuracy of the proposed approach is compared against the one achieved with the application of two traditional machine learning techniques, such as Support Vector Machine and Random Forest.