Figure 2
From: Machine learning discovery of cost-efficient dry cooler designs for concentrated solar power plants
The optimization process. In each optimization step our algorithm starts by selecting a dry cooler design (shown in circles). If the initial design does not satisfy constraints on temperature and pressure, it lengthens the cooler tubes until the constraints are satisfied. This can be seen as a projection step onto the feasible region of valid dry coolers. Given this, we obtain a design cost (shown by the colored contour lines). We use this cost to update a Gaussian process estimate of the global cost, and then use Bayesian optimization to select a new design to minimize the overall cost. This processes is repeated, continually reducing design cost.
