Optimizing solar cell efficiency with neural networks

Recent years have witnessed tremendous growth in the usage, research, and development of green energy, with further advancement doubtlessly on the horizon. One alternative energy technology of particular interest, currently being used and explored is the photovoltaic effect more commonly known as solar power [4]. Defined as the conversion of optical energy into electrical energy, the photovoltaic effect was first discovered in 1839 by Edmond Becquerel and may be brought to realization through the utilization of silicon semiconductor solar cells [8]. Build characteristics, such as diffusion time and temperature, alter the performance of the solar cell. The multiple parameters and conditions that are used in the building and design process can impact their efficiency [7]. These parameters can be put into datasets and used as initial conditions in a neural network. A neural network is a way to model the method by which the human brain processes information, using a large number of simulated processing units that resemble neurons. These processing units are typically arranged in layers such as an input layer, a hidden layer, and finally an output layer. Input values are usually propagated from each neuron in a layer to the neurons in the next layer before a result arrives at the output layer. The output typically has a target unit that is connected with varying weights. The network typically learns by examining records and data and generating predictions for them. Each time the network makes an incorrect prediction, it alters and adjusts the weights each iteration until the stopping criteria has been met. As the training progresses, the network becomes more accurate in replicating known outcomes. Once the known outcomes have been confirmed, the network can be applied to cases where the output is unknown. The neural network will then be able to learn by combining all of the data that is inputted [3]. The output of the neural network will be the optimal dopant diffusion time under which to produce the most efficient solar panel. In our study, we trained neural networks using different dopant diffusion times and the corresponding currents and voltages in order to determine the optimal time for producing the most power with a solar cell.