BP neural network-based prediction and optimal control of thermoelectric output of distributed solar PV/T systems

Abstract

This study employs An evolutionary algorithm to set up a multilayer BP neural network. The goal is to solve the issue that BP neural systems converge slowly and readily fall into local optimal solutions. The genetic algorithm reaches an initial set of evolutionary generations and outputs a BP neural network’s most appropriate starting weights and thresholds. In the GA-BP model, the fitness value is calculated as the sum of the total errors between the network’s output and the desired output. A communication protocol device was used to gather performance information for the global solar PV/T system, and the GA-BP neural network model parameters were adjusted. The GA-BP model had a mean absolute error of 0.7% lower than the BP model’s for the forecast of electrical efficiency. The GA-BP model has an average relative error of 0.21 in forecasting heating power.