Flor A. Castillo

Application Issues of Genetic Programming in Industry

This chapter gives a systematic view, based on the experience from The Dow Chemical Company, of the key issues for applying symbolic regression with Genetic Programming (GP) in industrial problems. The competitive advantages of GP are defined and several industrial problems appropriate for GP are recommended and referenced with specific applications in the chemical industry. A systematic method for selecting the key GP parameters, based on statistical design of experiments, is proposed. The most significant technical and non-technical issues for delivering a successful GP industrial application are discussed briefly.

Using Genetic Programming in Industrial Statistical Model Building

The chapter summarizes the practical experience of integrating genetic programming and statistical modeling at The Dow Chemical Company. A unique methodology for using Genetic Programming in statistical modeling of designed and undesigned data is described and illustrated with successful industrial applications. As a result of the synergistic efforts, the building technique has been improved and the model development cost and time can be significantly reduced. In case of designed data Genetic Programming reduced costs by suggesting transformations as an alternative to doing additional experimentation. In case of undesigned data Genetic Programming was instrumental in reducing the model building costs by providing alternative models for consideration.

Pareto front genetic programming parameter selection based on design of experiments and industrial data

Symbolic regression based on Pareto Front GP is the key approach for generating high-performance parsimonious empirical models acceptable for industrial applications. The paper addresses the issue of finding the optimal parameter settings of Pareto Front GP which direct the simulated evolution toward simple models with acceptable prediction error. A generic methodology based on statistical design of experiments is proposed. It includes statistical determination of the number of replicates by half-width confidence intervals, determination of the significant inputs by fractional factorial design of experiments, approaching the optimum by steepest ascent/descent, and local exploration around the optimum by Box Behnken or by central composite design of experiments. The results from implementing the proposed methodology to a small-sized industrial data set show that the statistically significant factors for symbolic regression, based on Pareto Front GP, are the number of cascades, the number of generations, and the population size. A second order regression model with high R2 of 0.97 includes the three parameters and their optimal values have been defined. The optimal parameter settings were validated with a separate small sized industrial data set. The optimal settings are recommended for symbolic regression applications using data sets with up to 5 inputs and up to 50 data points.

A methodology for combining symbolic regression and design of experiments to improve empirical model building

A novel methodology for empirical model building using GPgenerated symbolic regression in combination with statistical design of experiments as well as undesigned data is proposed. The main advantage of this methodology is the maximum data utilization when extrapolation is necessary. The methodology offers alternative non-linear models that can either linearize the response in the presence of Lack or Fit or challenge and confirm the results from the linear regression in a cost effective and time efficient fashion. The economic benefit is the reduced number of additional experiments in the presence of Lack of Fit.

Genetic Programming Transforms in Linear Regression Situations

The chapter summarizes the use of Genetic Programming (GP) inMultiple Linear Regression (MLR) to address multicollinearity and Lack of Fit (LOF). The basis of the proposed method is applying appropriate input transforms (model respecification) that deal with these issues while preserving the information content of the original variables. The transforms are selected from symbolic regression models with optimal trade-off between accuracy of prediction and expressional complexity, generated by multiobjective Pareto-front GP. The chapter includes a comparative study of the GP-generated transforms with Ridge Regression, a variant of ordinary Multiple Linear Regression, which has been a useful and commonly employed approach for reducing multicollinearity. The advantages of GP-generated model respecification are clearly defined and demonstrated. Some recommendations for transforms selection are given as well. The application benefits of the proposed approach are illustrated with a real industrial application in one of the broadest empirical modeling areas in manufacturing - robust inferential sensors. The chapter contributes to increasing the awareness of the potential of GP in statistical model building by MLR.

Competitive advantages of evolutionary computation for industrial applications

Defining the technical and business competitive advantages of evolutionary computation (EC) is critical for successful marketing of this technology in industry and other research communities. The key competitive advantages of EC, based on industrial applications in the chemical industry are presented in the paper. Gaining competitive advantage by integrating EC with statistical methods, neural networks, and support vector machines is recommended. Several examples of application areas in the chemical industry with demonstrated competitive advantage of EC are given. The most important areas are inferential sensors, empirical emulators of mechanistic models, accelerated new product development, complex process optimization, and effective industrial design of experiments

Split-Split-Plot Experimental Design in a High-Throughput Reactor

ABSTRACT In the last few years, high-throughput (HT) reactors have received significant attention due to the potential for fast material development. Split-plot experimental design plays a critical role in this type of application given randomization restrictions often imposed by equipment constraints. A case study in a parallel polymerization reactor is presented.

Empirical Models with Self-Assessment Capabilities for On-Line Industrial Applications

Self-assessment capabilities are critical for the longevity of online empirical models in industrial settings. A generic structure of an on-line model supervisor, consisting of within-the-range indicator, confidence of prediction, performance indicator, novelty/outlier detector, and model fault detector, is proposed in the paper. Several methods for confidence limits calculations, such as ensembles of analytic neural networks and symbolic regression models generated by genetic programming, linearized models based on transforms, derived by genetic programming, and a strangeness measure, based on support vector machines for regression, have been explored and their performance was compared in a case study for emission estimation on-line model. Some of the self-assessment capabilities for detection of unacceptable on-line performance and model and process faults are illustrated with industrial applications in the chemical industry.

Using evolutionary algorithms to suggest variable transformations in linear model lack-of-fit situations

When significant model lack of fit (LOF) is present in a second-order linear regression model, it is often difficult to propose the appropriate parameter transformation that will make model LOF insignificant. This paper presents the potential of genetic programming (GP) symbolic regression for reducing or eliminating significant second-order linear model LOF. A case study in an industrial setting at The Dow Chemical Company is presented to illustrate this methodology.

Symbolic regression in multicollinearity problems

In this paper the potential of GP-generated symbolic regression for alleviating multicollinearity problems in multiple regression is presented with a case study in an industrial setting. The main advantage of this approach is the potential to produce a simple and stable polynomial model in terms of the original variables.