The simulation and optimization of steady state process circuits by means of artificial neural networks

[摘要] ENGLISH ABASTRACT: Since the advent of modern process industries engineers engaged in themodelling and simulation of chemical and metallurgical processes have had tocontend with two important dilemmas. The first concerns the ill-defined natureof the processes they have to describe, while the second relates to thelimitations of prevailing computational resources.Current process simulation procedures are based on explicit process models inone form or another. Many chemical and metallurgical processes are notamenable to this kind of modelling however, and can not be incorporatedeffectively into current commercial process simulators. As a result manyprocess operations do not benefit from the use of predictive models andsimulation routines and plants are often poorly designed and run, ultimatelyleading to considerable losses in revenue.In addition to this dilemma, process simulation is in a very real way constrainedby available computing resources. The construction of adequate process modelsis essentially meaningless if these models can not be solved efficiently - asituation occurring all too often.In the light of these problems, it is thus not surprising that connectionistsystems or neural network methods are singularly attractive to processengineers, since they provide a powerful means of addressing both thesedilemmas. These nets can form implicit process models through learning byexample, and also serve as a vehicle for parallel supercomputing devices. In thisdissertation the use of artificial neural networks for the steady state modellingand optimization of chemical and metallurgical process circuits is consequentlyinvestigated.The first chapter is devoted to a brief overview of the simulation of chemicaland metallurgical plants by conventional methods, as well as the evolution andimpact of computer technology and artificial intelligence on the processindustries.Knowledge of the variance covariance matrices of process data is of paramountimportance to data reconciliation and gross error detection problems, andalthough various methods can be employed to estimate these often unknown variances, it is shown in the second chapter that the use of feedforward neuralnets can be more efficient than conventional strategies.In the following chapter the important problem of gross error detection inprocess data is addressed. Existing procedures are statistical and work well forsystems subject to linear constraints. Non-linear constraints are not handledwell by these methods and it is shown that back propagation neural nets can betrained to detect errors in process systems, regardless of the nature of theconstraints.In the fourth chapter the exploitation of the massively parallel informationprocessing structures of feedback neural nets in the optimization of processdata reconciliation problems is investigated. Although effective andsophisticated algorithms are available for these procedures, there is an everpresent demand for computational devices or routines that can accommodateprogressively larger or more complex problems. Simulations indicate that neuralnets can be efficient instruments for the implementation of parallel strategiesfor the optimization of such problems.In the penultimate chapter a gold reduction plant and a leach plant are modelledwith neural nets and the models shown to be considerably better than the linearregression models used in practice. The same technique is also demonstratedwith the modelling of an apatite flotation plant. Neural nets can also be used inconjunction with other methods and in the same chapter the steady statesimulation and optimization of a gravity separation circuit with the use of twolinear programming models and a neural net are described.

[发布机构] Stellenbosch University