HARMONICS FORECASTING FOR WIND AND SOLAR RENEWABLE ENERGY RESOURCES-BASED ELECTRICAL POWER SYSTEMS

Abstract

The practice of harmonics forecasting plays an integral role in the development of mitigation devices aimed at lessening the adverse effects of harmonic disturbances in electrical systems. This doctoral research endeavours to contribute to this field by introducing a novel hybrid forecasting model capable of generating precise and reliable harmonics predictions for Renewable Energy Systems (RESs). To attain this objective, multi-layered Advanced Neural Networks (ANNs), the Adaptive Neuro Fuzzy Inference System (ANFIS), and the Long Short-Term Memory (LSTM) network were harnessed to formulate eight innovative hybrid forecasting models, which are the integral components of this study. Within the scope of the research, three distinct ANN structures featuring three layers each—Cascaded Recurrent Neural Network with Local feedback (3LCRNNL), Cascaded Recurrent Neural Network with Global feedback (3LCRNNG), and Cascaded Recurrent Neural Network with Local and Global feedback (CRNNLG)—are combined with ANFIS to create the initial six hybrid forecasting models (Models 1-6). The integration of the ANFIS-LSTM techniques results in the formulation of two additional hybrid models (Models 7-8). In conjunction with these modelling efforts, two renewable generator models are employed to generate harmonics. The first model involves a grid-connected Double-Fed Induction Generator (DFIG) driven by a wind turbine and integrated with a Solar Photovoltaic (PV)-based power generator. The second generator model combines a Solar-PV generator with a wind turbine-linked Permanent Magnet Synchronous Generator (PMSG) interconnected to a shared grid. The harmonics generated by these generator models are utilized to construct comprehensive training and testing datasets that are subsequently employed to generate forecasts using the novel hybrid forecasting models proposed in this research. To rigorously evaluate the performance and effectiveness of these models, a systematic comparison is conducted against benchmark studies available in the literature. The findings highlight the exceptional performance consistency of model-8, which not only outperforms all of the other proposed models in the study, but also significantly surpasses the capabilities of existing techniques in the literature. Moreover, this study underscores the superiority of hybrid forecasting models over individual forecasting techniques typically used as benchmarks, thereby reaffirming the value of hybrid modelling in the context of harmonics forecasting for RESs.