This paper presents the design and analytical modeling of the proposed solar photovoltaic standalone system under varying environmental conditions. The proposed system consists of a unique structure of a solar PV-tree, maximum power point tracking (MPPT) technique, and DC–DC converter. The output voltage acquired from the solar PV tree is low. A DC–DC boost converter is utilized to step-up the required amount of voltage level. In this paper, the appropriate duty cycle is obtained for extracting the optimum power from the solar PV tree by using various MPPT mechanisms such as perturb and observe (P&O), incremental conductance (INC), and a radial basis function network (RBFN)-based neural network (NN). The proposed solar photovoltaic tree-based energy harvesting system is designed and validated by using MATLAB/SIMULINK software and real-time application. The simulation results of the above-mentioned three techniques are compared with each other in order to show the effectiveness of the proposed system with RBFN. The RBFN-MPPT provides a significant improvement in tracking efficiency of 6.0% and 5.72% as compared with the P&O method and the INC method at 1000 W/m2 irradiance condition. From the simulation and real-time results, it is concluded that the RBFN-based NN provides better tracking efficiency and less oscillation as compared with the other two algorithms.
Electrical load demand forecasting plays a critical role in power system management that allows utilities to adequately plan resources and maximize operating efficiency. In this research, utilization of state-of-the-art machine learning algorithms, Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Convolutional Neural Network with LSTM (CNN- LSTM), Support Vector Machine (SVM) and AutoRegressive Integrated Moving Algorithm (ARIMA) has been explored for electrical load demand forecasting. The models are trained on past load data and tested on primary performance metrics, Root Mean Square Error (RMSE) and Co-efficient of determination ($\mathbf{R}^{\mathbf{2}}$). LSTM achieves the best result with an $\mathbf{R}^{\mathbf{2}}$ value of 0.978 and RMSE value of 27.46, proving its suitability to learn intricate temporal features in the data, closely followed by CNN-LSTM. RNN also delivers a competitive performance, with SVM not faring well as it results in increased error and worse accuracy. ARIMA did not produce the best and achieves high RMSE and negative $\mathbf{R}^{\mathbf{2}}$ demonstrating the lack of power in nonlinear and dynamic trend management. From this study, it is assured that there exists promise for smart and renewable energy systems via machine learning adoption by the energy sector aimed at limiting the upcoming difficulty of demand prediction.