A Method Combine Network with Random Subregion Gradient Descent Algorithm for the Design of the Heliostat Field Layout

Abstract This paper presents a method that combine a network with random subregion gradient descent algorithm, which we later called “RSNGD”, optimization of heliostat field layout for the design of a tower solar thermal power station. The algorithm divides the heliostat field into multiple parts, randomly samples in each section, calculates gradients for the sampled points, and then employs a network to estimate gradients for the remaining points. It provides an efficient means to optimize problems spanning continuous and discrete cases through the use of network. The algorithm trades precision for faster gradient computations, enabling more update steps and iterations within the same time budget. The algorithm successfully enhances heliostat field layout design approximately tenfold faster than particle swarm optimization. Results demonstrate that the algorithm is significantly faster than other optimization in a test case involving 1746 heliostats.