Comparative Study on Forecasting of Schedule Generation in Delhi Region for the Resilient Power Grid Using Machine Learning

• Published in: IEEE transactions on industry applications 2024-03, Vol.60 (2), p.2107-2116
• Main Authors: D, Lakshmi, Tiwari, Ravi Shekhar, Nagpal, Neelu, Kassarwani, Neelam, G, Vishnuvarthanan, Srivastava, Abhishek
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Analytical models; Calendars; Comparative studies; Data models; Deep learning; Electric power grids; electrical load forecasting; Electrical loads; Electricity distribution; Energy sources; ensemble learning; feature extraction; Forecasting; Interactive systems; Load forecasting; Load modeling; Machine learning; Mathematical models; Meteorological data; Meteorology; Predictive models; Recurrent neural networks; resilient power grid; Scheduling; Smart grid; time series analysis. short-term load forecasting
• ISSN: 0093-9994; 1939-9367
• DOI: 10.1109/TIA.2023.3316646

The increasing use of Renewable Energy Resources (RES) in energy generation has led to the transformation of the conventional electrical grid into a more adaptable and interactive system, and this has made electrical load prediction a crucial aspect of smart grid operation. Short-Term Load Forecasting (STLF) is the ultimate requirement for the essentialities, such as planning, scheduling, management, and trading of electricity. In the proposed work, a forecasting engine model is developed to figure out the load of the upcoming twelve months (2020) in the Delhi metropolis, and this is accomplished by integrating real and dynamic meteorological data, calendar data, and load patterns for the successive two years (2017-2018). It is performed using different ensemble models, such as XGBoost, Gradient Boosting, AdaBoost, Random Forest (RF) algorithms, and deep learning models such as Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and the Prophet algorithm. The simulation results of the proposed models are obtained on the Python platform using Delhi weather, load, and calendar data. Further, the STLF is analyzed using 14 different models on the basis of 78 scenarios, and 8 data sets are analyzed in conjunction. The train, validation, and test accuracy have been considered as validation metrics, both on hourly and daily load forecasting, to validate the overfitting in terms of the train, validation, and test loss. A comparative study is made to show that the predictions of LSTM and GRU outperform with 100% accuracy.