Augmented Genetic Algorithm for Decoupling Capacitor Optimization in Power Distribution Network Design Through Improved Population Generation

In power distribution network designs, a large number of decoupling capacitors (decaps) may be needed to satisfy target impedance limits. Many algorithms have been proposed and implemented for finding the optimal decap placement, including genetic algorithms (GA), and machine learning methods. In this work, an improved GA is proposed for finding the decap placement pattern that can satisfy a target impedance using the minimum number of decaps. The distribution of capacitors expected to appear in the global minimum solution is first predicted by determining how effective each decap type is toward satisfying certain critical impedance points. This estimation is used to inform the generation of initial solutions in order to put the initial search space nearer the global minimum and ensure certain solution characteristics appear. GA search using this improved population generation is found to be an improvement over a Canonical GA implementation, by finding solutions where the latter could not, or finding a solution using fewer decaps.