Inertia Forecasting using Hybrid Machine Learning

High penetration of low-inertia Renewable Energy Sources (RESs) like wind and solar in power systems results in frequency stability issues. This necessitates the deployment of a Frequency Containment Reserve (FCR) to ensure frequency stability. The optimal deployment of FCR necessitates accurate inertia forecasts. Machine learning models like ANN are used for inertia forecasting. However, the accuracy of these models highly relies on their hyperparameters. This necessitates a suitable hybrid model to improve the forecasting accuracy. A hybrid model uses a machine learning algorithm to obtain forecasts and hyperparameters of the model are optimized by a suitable metaheuristic algorithm. Further, the missing values in the input data significantly affect the forecasting accuracy. The missing values can be easily replaced by a suitable data preprocessing technique. Therefore, this paper proposes two novel hybrid inertia forecasting models (ANN_ WaOA-CleanTS and SVR_ WaOA-CleanTS) using two powerful machine learning forecasting algorithms (ANN and SVR), a new metaheuristic optimization algorithm, Walrus Optimization Algorithm (WaOA), and a data preprocessing tool (cleanTS). ANN and SVR models are capable of obtaining accurate inertia forecasts due to their capability to identify nonlinear patterns in the data. The WaOA shows superior performance compared to other optimization techniques in various forecasting problems. The proposed work uses an R package, cleanTS, to replace the missing values as it shows superior performance in various forecasting problems. The performance analysis shows that the proposed hybrid models have an annual inertia forecasting accuracy above 97%.