(169e) The Alamo Software for Model Building, Constrained Regression, and Intelligent Experimental Design

The purpose of ALAMO (Automated Learning of Algebraic Models using Optimization) is to generate algebraic surrogate models of black-box systems for which a simulator or experimental setup is available.  Consider a system for which the outputs z are an unknown function of the system inputs x.   The software identifies a function z=f(x), i.e., a relationship between the inputs and outputs of the system, that best matches data (pairs of x and corresponding z-values) that are collected via simulation or experimentation.  ALAMO can:

• build an algebraic model of a simulation or experimental black-box system
• use previously collected data for model building
• call a user-specified (simulation) function to collect measurements
• enforce response variable bounds, physical limits, and boundary conditions
• use a preexisting data set for model validation

The problems addressed by the software have long been studied in the fields of statistics, design of experiments, and machine learning.  Whereas existing techniques from this literature can be used to fit data to models, ALAMO is distinctive in that it determines where to run the simulations or experiments, what models to fit, and how to determine if a model thus produced is accurate and as simple as possible.  A distinguishing feature of the software is that it provides models that are as simple as possible and still accurate.  These model attributes are achieved by using a recently developed optimization methodology [1].  Moreover, ALAMO is uniquely in that it utilizes theory-driven insights alongside data in its constrained regression module [2]. 

Following a brief review of the algorithms implemented in the software, we describe ALAMO’s software design and a number of applications that ALAMO facilitates.  The ALAMO models can be used to facilitate subsequent system analysis, optimization, and decision making.

References cited

1. Cozad, A., N. V. Sahinidis and D. C. Miller, Automatic learning of algebraic models for optimization, AIChE Journal, 60, 2211-2227, 2014.
2. Cozad, A., N. V. Sahinidis and D. C. Miller, A combined first-principles and data-driven approach to model building, Computers and Chemical Engineering, 73, 116 - 127, 2015.