GPU-based Aim Point Optimization for Solar Tower Power Plants

Abstract

In solar tower power plants, aim point optimization is suitable to find aim point distributions resulting in intercept powers close to the theoretical maximum. However, the application in real time operation often faces the problem of long optimization duration. To counteract this issue, the convergence of an existing strategy, the ant colony optimization meta-heuristic, is enhanced. The raytracing is already replaced by pre-calculated flux maps of the individual heliostats in previous works to increase the optimization speed. In this work, the optimization is merged with a grouping strategy and implemented on a GPU to achieve further time reductions. Here, a k-means clustering algorithm performs the heliostats grouping. The use of groups reduces the solution space for the optimizer and additionally the amount of pre-calculated flux maps, so that the data fits in the global memory of the GPU. Over 100 billion flux values can be evaluated per second using this adapted approach. In this way, the algorithm finds suitable aim point distributions within a few seconds up to a minute. The achieved intercepts are 1 % to 4 % higher then those found by a single factor aiming strategy for the evaluated central receiver reference power plant. Moreover, the approach has proved its applicability in clouded environments that lead to spatially fluctuating solar radiation. There, a spillage reduction compared to the single factor aiming of 35 % is reached.