Systems and methods for generating a reduced-order surrogate model using genetic algorithms

A system for generating a reduced-order surrogate model that approximates a data set, comprising:

• a processor;
• a secondary storage device communicably coupled with the processor and configured to store the data set;
• and
• a memory communicably coupled with the processor and storing machine-readable instructions that, when executed by the processor, control the system to:
• determine, for each parent chromosome of a population of parent chromosomes, a fitness value by applying the data set to a chromosome function constructed from a sequence of genes of said each parent chromosome,
• randomly assign each parent chromosome into one of a first subpopulation of parent chromosomes and a second subpopulation of parent chromosomes,
• order the first subpopulation of parent chromosomes based on their fitness values,
• generate a population of child chromosomes by iteratively:
• (i) randomly selecting, from the first subpopulation, a first parent chromosome whose fitness value is within an upper echelon of the fitness values of the parent chromosomes of the first subpopulation,
• (ii) randomly selecting a second parent chromosome from the second subpopulation, and
• (iii) constructing a child chromosome of the population of child chromosomes by, for each gene starting from a first gene of a linked sequence of parent genes of the first parent chromosome, randomly selecting either (i) a gene of the first parent chromosome at a corresponding gene position or (ii) a gene of the second parent chromosome at the corresponding gene position, and copying the selected gene into a corresponding child gene of the child chromosome, and continuing until a last gene of the linked sequence of parent genes has been randomly selected and copied into a last child gene of the child chromosome;
• execute computer code to generate a plurality of reduced-order surrogate models based, at least in part, on the population of child chromosomes iteratively generated,
• identify the reduced-order surrogate model based, at least in part, on the generation of the plurality of reduced-order surrogate models, the reduced-order surrogate model identified further based on a plurality of outputs generated by executing the computer code, the reduced-order surrogate model represented by a fittest chromosome function constructed from a fittest child chromosome having a highest fitness value of the population of child chromosomes,
• parse the sequence of genes of the fittest child chromosome to determine the fittest chromosome function in a human-readable format, wherein the human-readable format comprises a mathematical equation representing the reduced-order surrogate model,
• load the mathematical equation into a computer program configured to parse mathematical expressions to perform validation or prediction, and
• perform, by the computer program, computer-implemented processing utilizing the loaded mathematical equation representing the reduced-order surrogate model to generate verification data for the loaded mathematical equation, the verification data including one or more predicted values determined by the loaded mathematical equation based on an alternative data set comprising one or more data samples partitioned from the data set prior to determining the fitness value for each parent chromosome of the population of parent chromosomes.