Henrik Saxén

A genetic algorithms based multi-objective neural net applied to noisy blast furnace data

A genetic algorithms based multi-objective optimization technique was utilized in the training process of a feed forward neural network, using noisy data from an industrial iron blast furnace. The number of nodes in the hidden layer, the architecture of the lower part of the network, as well as the weights used in them were kept as variables, and a Pareto front was effectively constructed by minimizing the training error along with the network size. A predator-prey algorithm efficiently performed the optimization task and several important trends were observed.

Analyzing Leaching Data for Low-Grade Manganese Ore Using Neural Nets and Multiobjective Genetic Algorithms

Existing acid leaching data for low-grade manganese ores are modeled using an evolving neural net. Three distinct cases of leaching in the presence of glucose, sucrose and lactose have been considered and the results compared with an existing analytical model. The neural models are then subjected to bi-objective optimization, using a predator–prey genetic algorithm, maximizing recovery in tandem with a minimization of the acid concentration. The resulting Pareto frontiers are analyzed and discussed.

Genetic Programming Evolved through Bi-Objective Genetic Algorithms Applied to a Blast Furnace

In this study, a new Bi-objective Genetic Programming (BioGP) technique was developed that initially attempts to minimize training error through a single objective procedure and subsequently switches to bi-objective evolution to work out a Pareto-tradeoff between model complexity and accuracy. For a set of highly noisy industrial data from an operational ironmaking blast furnace (BF) this method was pitted against an Evolutionary Neural Network (EvoNN) developed earlier by the authors. The BioGP procedure was found to produce very competitive results for this complex modeling problem and because of its generic nature, opens a new avenue for data-driven modeling in many other domains.

Genetic Algorithm-Based Multicriteria Optimization of Ironmaking in the Blast Furnace

A method has been developed for optimizing ironmaking in the blast furnace with the aim to minimize costs and CO2 emissions. These two goals are pursued by a genetic algorithm yielding states of operation on a Pareto-optimal front with nondominated solutions. The blast furnace process is described mathematically by a thermodynamic simulation model, where realistic operational constraints are imposed. The states on the Pareto-optimal fronts evolved are analyzed in more detail, considering the constraints of the process. The solutions are found to give rise to clearly different specific emissions but very similar specific costs as long as the production stays within the limits of the granted CO2 emissions allowances of the plant. However, this also implies that the costs of ironmaking may rise considerably along with increased prices of the allowances or reduced emission rights. The findings of the work are expected to be valuable in the strategic evaluation of future ironmaking options.

Method for the selection of inputs and structure of feedforward neural networks

Feedforward neural networks of multi-layer perceptron type can be used as nonlinear black-box models in data-mining tasks. Common problems encountered are how to select relevant inputs from a large set of variables that potentially affect the outputs to be modeled, as well as high levels of noise in the data sets. In order to avoid over-fitting of the resulting model, the input dimension and/or the number of hidden nodes have to be restricted. This paper presents a systematic method that can guide the selection of both input variables and a sparse connectivity of the lower layer of connections in feedforward neural networks of multi-layer perceptron type with one layer of hidden nonlinear units and a single linear output node. The algorithm is illustrated on three benchmark problems.

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

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.

Analyzing Sparse Data for Nitride Spinels Using Data Mining, Neural Networks, and Multiobjective Genetic Algorithms

Nitride spinels are typically characterized by their unique AB2N4 structure containing a divalent cation A, a trivalent cation B, and an anion N. Numerous such species may exist as metals, semiconductors, or semimetals leading to their extensive usage in diverse scientific and engineering fields. Experimental and theoretical data on the physical or material properties of nitride spinels are, however, severely limited for coming up with a data driven, generic description for their material properties. In this study we have attempted to establish a methodology for handling such sparse data where the various features of some of the state of the art soft computing tools like Genetic Algorithms, Data Mining, and Neural Networks are used in tandem to construct some generic predictive models, in principle applicable to the nitride spinel structures at large, irrespective of their electronic characteristics. The paucity of the available data was circumvented in this work with a data mining strategy, important inputs were identified through an evolving neural net, and finally, the best possible tradeoffs between the bulk moduli and the relative stabilization energies of the nitride spinels were identified by constructing the Pareto-frontier for them through a Genetic Algorithms-based multiobjective optimization strategy.

Analysing blast furnace data using evolutionary neural network and multiobjective genetic algorithms

Abstract Approximately one years operational data of a TATA Steel blast furnace were subjected to a multiobjective optimisation using genetic algorithms. Data driven models were constructed for productivity, CO2 content of the top gas and Si content of the hot metal, using an evolutionary neural network that itself evolved through a multiobjective genetic algorithm as a tradeoff between the accuracy of training and the network complexity. The final networks were selected using the corrected Akaike information criterion. Bi-objective optimisation studies were subsequently carried out between the productivity and CO2 content with various constraints at the Si level in the hot metal. The results indicate that a productivity increase would entail either a compromise of the CO2 fraction in the top gas or the Si content in the hot metal. The Pareto frontiers presented in this study provide the best possible parameter settings in such a scenario.

Identification of Factors Governing Mechanical Properties of TRIP-Aided Steel Using Genetic Algorithms and Neural Networks

Mechanical properties of transformation induced plasticity (TRIP)-aided multiphase steels are modeled by neural networks using two methods of reducing the network connectivity, viz. a pruning algorithm and a predator prey algorithm, to gain understanding on the impact of steel composition and treatment. The pruning algorithm gradually reduces the complexity of the lower layer of connections, removing less significant connections. In the predator prey algorithm, a genetic algorithm based multi-objective optimization technique evolves neural networks on a Pareto front, simultaneously minimizing training error and network size. The results show that the techniques find parsimonious models and, furthermore, extract useful knowledge from the data.

Data-Driven Time Discrete Models for Dynamic Prediction of the Hot Metal Silicon Content in the Blast Furnace—A Review

A review of black-box models for short-term time-discrete prediction of the silicon content of hot metal produced in blast furnaces is presented. The review is primarily focused on work presented in journal papers, but still includes some early conference papers (published before 1990) which have a clear contribution to the field. Linear and nonlinear models are treated separately, and within each group a rough subdivision according to the model type is made. Within each subsection the models are treated (almost) chronologically, presenting the principle behind the modeling approach, the signals used and the main findings in terms of accuracy and usefulness. Finally, in the final section the approaches are discussed and some potential lines of future research are proposed. In an Appendix , a list of commonly used input and output variables in the models is presented.