Equivalent inertia prediction for power systems with virtual inertia based on PSO-SVM

Abstract

Inertia prediction for power systems with a high proportion of renewable energy units can help coordinate inertia support methods, guide power system planning, and lower grid operational risk. Existing inertia prediction methods rarely use machine learning to predict the equivalent inertia of the power system, and there is also little consideration of the virtual inertia of the renewable energy units; some of the prediction methods rely on massive volumes of system data and suffer from issues such as data redundancy and complex pre-processing procedures. A method for predicting the equivalent inertia for power systems based on particle swarm optimization support vector machines (PSO-SVM) is proposed for this purpose. The method initially creates a database of system-equivalent inertia, which regards the power change and system frequency rate of change as feature inputs and the system-equivalent inertia as an output. Then, the optimal prediction model is matched using the feature difference matrix, and the PSO-SVM prediction method is utilized to predict the power system's equivalent inertia. The method proposed in this paper is validated by an improved three-machine nine-node power system, and the prediction accuracy is better than that of GA-BP neural network and SVM algorithms, and then the applicability in complex scenarios is validated by a ten-machine, thirty-nine-node power system as well as a site-specific power system under real-time wind speeds. The PSO-SVM prediction method reduces the maximum error by 23.64% compared to the GA-BP neural network and 68.27% compared to the SVM algorithm and the results show that the method proposed in this paper can more accurately predict inertial changes and inertial information of the system when a loading accident occurs.