Modeling and optimization of photovoltaic systems under partially shaded and rapidly changing conditions

Abstract

This thesis focuses on maximum power point tracking (MPPT) techniques for PV systems under partial shading and rapidly changing irradiance conditions. However, during the initial investigation of this topic, we realized that parameter extraction and modeling of the photovoltaic (PV) cell/module/array were fundamental to our analysis and simulation of MPPT systems. Limitations to existing parameter extraction and modeling techniques for PV systems resulted in the scope of this thesis being extended to also cover these topics. In Chapter 1, the PV system is introduced and the motivations for the research on this topic are presented. In Chapter 2, various methods for modeling the PV cell/module/array and the MPPT techniques in the literature are reviewed. The performance of each method is evaluated and compared with each other in a table. Since the existing methods show either low accuracy in derived parameters of the PV array or too much complexity in their calculation, in Chapter 3, we proposed a Chebyshev functional link neural network (CFLNN) to model the PV module. Due to the absence of hidden layers in the network configuration, the complexity of the network based model is reduced and a better modeling accuracy is obtained using the proposed method. The results of current prediction using two other modeling methods – the two diode model and multilayer perceptron (MLP), are compared with that by the proposed method. The experimental results show that, compared to the analytical model, the CFLNN modeling method provides better prediction of the output current, and compared to the conventional MLP model, it has a reduced computational complexity. The determination of the parameters of the PV cell is of great importance to solar energy related research such as for investigating the performance of the MPPT system under various irradiance conditions, analyzing the performance of more complex grid-connected PV systems, and studying the insight to the known physical processes so that it can be used in quality control during the development of the devices and for fabrication process optimization etc. In the second part of Chapter 3, an improved adaptive differential evolutionary (IADE) algorithm is proposed to extract the parameters of the electrical circuit for PV modules. The principle is to apply the feedback of fitness value in the evolutionary process. Comparisons of parameter estimation results with existing methods for solar cell and module under various environmental conditions are conducted using both synthetic and experimental data. The proposed method eliminates the requirement for users to manually tune the control parameters of the differential evolutionary algorithms and offers better accuracy of the extracted parameters. After analyzing the disadvantages of the existing MPPT techniques, we proposed an ant colony optimization (ACO) based MPPT for the PV system to work under partial shading environments. The ACO based MPPT, which is introduced in Chapter 4, is then verified using simulations and an experimental setup. This is the first time that ACO has been directly used for MPPT in a PV system. As other related evolutionary algorithms can also be applied to the MPPT in a similar way, a uniform implementation scheme for other evolutionary algorithms under several different PV array topologies is proposed. Furthermore, a strategy to accelerate the convergence speed is also investigated. It provides an essential and useful guideline to implement the related evolutionary algorithms into the MPPTs. In Chapter 5, another type of MPPT technique, a hybrid MPPT which combines an artificial neural network (ANN)-based MPPT and a conventional MPPT, is proposed for PV systems operating under conditions with rapid irradiance change. The ANN is utilized to initially categorize the MPP region based on the irradiance pattern when irradiance sensors are available or based on spot current measurements along the I-V curve when irradiance sensors are not available. This classification is then used to force a conventional MPPT such as the perturb and observe (P&O) or extremum seeking control (ESC) to search near the identified MPP, resulting in the system tracking the global MPP. The direct prediction of the global MPP region in one step guarantees a fast system response. The implementation without irradiance sensors can accurately track the global MPP and features a relatively low cost. The effectiveness of the proposed hybrid MPPT is verified using both simulation and an experimental setup. The results show that it can effectively track the global MPP resulting in a significant power increase and as such, can be applied in locations with rapid irradiance change, such as in the tropics. Finally, in Chapter 6, the conclusions of this thesis are made and the possible areas that we can continue to work on in the future are discussed.