Evolutionary Spatial Auto-Correlation For Assessing Earthquake Liquefaction Potential Using Parallel Linear Genetic Programming

The assessment of sites for liquefaction potential in earthquakes currently relies on the estimation of soil layer models which is laborious and standard regression techniques ineffectual. Although Parallel Linear Genetic Programming (PLGP) has proven to be an effective method for classification tasks it has not yet been applied to regression problems. This paper redefines a time-consuming, operator intensive process as an Evolutionary Computation (EC) regression task and designs a PLGP system that can produce candidate solutions for an operator to review. This paper introduces Evolutionary Spatial Auto-Correlation (ESPAC) which is an EC technique that uses a similar structure to PLGP programs to represent some layer models and evolve them using error matching against the target curve as a fitness function. The project achieves its goal of providing a working proof-of-concept with resultant curve matching being improved over that of a domain expert on four of the five datasets tested.