Modelling Noisy Blast Furnace Data using Genetic Algorithms and Neural Networks

Helle, Mikko; Pettersson, Frank; Chakraborti, Nirupam; Saxén, Henrik

doi:10.1002/srin.200606357

Cited by 38 publications

(24 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…6, which shows that in all the cases almost all the inputs have certain importance, as shown in. 22) Here also it is seen that Cu and cooling rate have a significant role to play. The figures also reveal some additional information: The importance of Cu is found to be higher in the case of UTS than YS (cf.…”

Section: Modeling With Predator Prey Algorithmmentioning

confidence: 84%

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

et al. 2007

Self Cite

View full text Add to dashboard Cite

The properties of steels depend in a complex way on their composition and heat treatment and neural networks have therefore recently been widely used for capturing these relationships. Two different methods of reducing the network connectivity, viz a pruning algorithm and a multi-objective predator prey genetic algorithm, have been used for neural network modeling of the mechanical properties of high strength steels, so that relevant connections within the networks are revealed. This provides important understanding on the variables and their relationship with mechanical properties. In the pruning algorithm the lower layer of the network is gradually reduced by removing less significant connections. In the predator prey algorithm, a genetic algorithm based multi-objective optimization technique is used to train the neural network and a Pareto front is developed by minimizing the training error along with the network size. The results of both techniques reveal that they can extract more knowledge from the data, which is difficult to obtain from conventional neural models. The relative relevance of the composition and processing parameters detected could be used for designing steel with tailored property balance. The results developed by the two techniques are also found to be comparable.KEY WORDS: high strength multiphase steel; neural network model; pruning algorithm; genetic algorithm; multi-objective optimization; predator prey algorithm; alloy design. ISIJ International, Vol. 47 (2007), No. 8, pp. 1195No. 8, pp. -1203No. 8, pp. 1195 © 2007 ISIJ of metallurgy, but the quantitative relations are not always known. Artificial neural networks (ANN) are learning systems, which try to learn and map the existing input output relationship in an accurate way on the basis of experimental information, and have been found successful in developing prediction model for the properties of TMCP steels. 2,3) ANN methods have also been found superior compared to traditional regression techniques in case of prediction of properties of steel. 4,5) An ANN analysis usually depends on published data or data developed in the laboratories, and that often implies working with a diminutive databank. Sometimes, due to the large number of potential inputs the practical limitation may lead to over-fitting of the ANN model where the abilities of extracting meaningful information from the data are lost. An established method for avoiding over-fitting is to restrict the number of connectivity (weights) between the inputs and the hidden nodes and even to restrict the numbers of inputs and hidden nodes. 6,7) In addition to avoiding over-fitting, with its detrimental effect on model accuracy, a detection of the relevant connections within the network could provide important understanding on the variables and their relationship.In the present work two different methods of reducing the network connectivity have been used for modeling the mechanical properties of high strength steels. Firstly an intuitive pruning algorithm 8) with a large sigmo...

show abstract

Section: Modeling With Predator Prey Algorithmmentioning

confidence: 84%

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

et al. 2007

Self Cite

View full text Add to dashboard Cite

show abstract

“…The ironmaking blast furnace is an extremely complex metallurgical process, and information from it therefore lends itself perfectly to be used for tests of novel modeling techniques [41,42]. In this paper a difficult forecasting problem, i.e., the prediction of the silicon content of the hot metal, has been used as an example to demonstrate the potential of the method to both finding appropriate inputs and to evolving a sparse connectivity of the lower layer of the networks.…”

Section: Discussionmentioning

confidence: 98%

Evolving Nonlinear Time-Series Models of the Hot Metal Silicon Content in the Blast Furnace

Saxén

Pettersson

Gunturu

2007

Materials and Manufacturing Processes

Self Cite

View full text Add to dashboard Cite

Neural networks are versatile tools for nonlinear modeling, but in time-series modeling of complex industrial processes the choice of relevant inputs and time lags can be a major problem. A novel method for the simultaneous detection of relevant inputs and an appropriate structure of the lower part of the networks has been developed by evolving neural networks by a genetic algorithm, where the approximation error and the number of weights are minimized simultaneously by multiobjective optimization. The networks on the Pareto front are considered possible candidate models that are evaluated on an independent test set. In order to consider the problem of drift in the variables, which may cause parsimonious models to perform poorly on the test set, the weights in the upper layer of the networks are recursively estimated by a Kalman filter. The method is illustrated on a data set from ironmaking industry, where time-series models of the hot metal silicon content in a blast furnace are evolved. The technique is demonstrated to synthesize models with a choice of inputs in agreement with findings presented in the literature and process know-how.

show abstract

“…This approach proposed in some of our earlier work [9,19], evolves through genetic algorithms and has already been discussed in sufficient detail in a number of published The Physical Properties Studied: Valence-electron number, MartynovBatsanov electronegativity, atomic radius (Zunger's Psuedopotential radii of s and p electrons), principal quantum number, Pauling's electronegativity, Miedma's chemical potential [17], electron density in Wigner-Seitz atomic cell [17]. articles [19][20][21][22].…”

Section: The Evolving Neural Networkmentioning

confidence: 99%

“…In fact, the algorithm used here is capable of identifying the significant inputs through a simple probabilistic procedure [19] and use them in the network construction. f) The lower part of the network evolves through genetic algorithms.…”

Section: The Evolving Neural Networkmentioning

confidence: 99%

See 1 more Smart Citation

Identification and Optimization of AB₂Phases Using Principal Component Analysis, Evolutionary Neural Nets, and Multiobjective Genetic Algorithms

Agarwal

Pettersson

Singh

et al. 2009

Materials and Manufacturing Processes

Self Cite

View full text Add to dashboard Cite

Available data for a large number of AB 2 compounds were subjected to a rigorous study using a combination of Principal Component Analysis (PCA) technique, multiobjective genetic algorithms, and neural networks that evolved through genetic algorithms. The identification of various phases and phase-groups were very successfully done using a decision tree approach. Since the variable hyperspaces for the different phases were highly intersecting in nature, a cumulative probability index was defined for the formation of individual compounds, which was maximized along with Pauling's electronegativity difference. The resulting Pareto-frontiers provided further insight into the nature of bonding prevailing in these compounds.

show abstract

Modelling Noisy Blast Furnace Data using Genetic Algorithms and Neural Networks

Cited by 38 publications

References 22 publications

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

Evolving Nonlinear Time-Series Models of the Hot Metal Silicon Content in the Blast Furnace

Identification and Optimization of AB₂Phases Using Principal Component Analysis, Evolutionary Neural Nets, and Multiobjective Genetic Algorithms

Contact Info

Product

Resources

About

Modelling Noisy Blast Furnace Data using Genetic Algorithms and Neural Networks

Cited by 38 publications

References 22 publications

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

Designing High Strength Multi-phase Steel for Improved Strength–Ductility Balance Using Neural Networks and Multi-objective Genetic Algorithms

Evolving Nonlinear Time-Series Models of the Hot Metal Silicon Content in the Blast Furnace

Identification and Optimization of AB2Phases Using Principal Component Analysis, Evolutionary Neural Nets, and Multiobjective Genetic Algorithms

Contact Info

Product

Resources

About

Identification and Optimization of AB₂Phases Using Principal Component Analysis, Evolutionary Neural Nets, and Multiobjective Genetic Algorithms