An Improved Water Cycle Optimized Extreme Learning Machine with Ridge Regression towards Effective Maximum Power Extraction for Photovoltaic based Active Distribution Grids

摘要

Accurate Maximum Power Point Tracking (MPPT) is addressed in this paper for Photovoltaic (PV) based Distributed Generation (DG) by virtue of an error optimized (by Improved Water Cycle) Extreme Learning Machine with Ridge Regression (IWC-ELMRR). For robust performance against this DG integration DC bus voltage ($V_{dc}$) stability is obtained by considering the PV system to be single-staged (DC-AC). As solar power is independent with respect to grid operation, The MPPT output is considered for stable grid response in terms of DC-AC control parameter calculation. The power demand at grid side is considered with a provision to be compensated by Battery Energy Storage (BES), operated parallel to PV-DG at DC bus. The MPPT schemes are prone to estimation error ($Er_{MPP}$), especially while considering PV part (e.g. shading) and grid part (e.g. short-circuit faults) uncertainties. Thus robust depletion in $Er_{MPP}$ profile is targeted to be achieved irrespective of grid/ DG contingencies by mean of proposed IWC-ELMRR. Here randomness and data dependent performance of Artificial Neural Network (ANN) based MPPT approach is overcome by estimating optimal initial weight parameters through IWC. A sinusoidal chaotic map is used for accurate, nonlinear assessment of MPP values under grid/ DG uncertainties. The fastness in computation is achieved by non-iterative Moore Penrose pseudo- Inverse (MPI). The efficacy of the proposed technique is evidenced in terms of improvement in $Er_{MPP}$ profile, reduced dynamic oscillation at DG coupling bus. Various cases are considered in MATLAB environment against such superiority performance, compared with recent MPPT techniques.