Jan Braun

Interactive evolutionary multiobjective optimization for hydraulic valve controller parameters

This paper proposes a novel approach of interactive evolutionary multiobjective optimization. The approach combines an evolutionary algorithm with an instance based supervised online learning scheme for user preferences. Interactive preference articulation constitutes an alternative to a priori as well as a posteriori decision making in multiobjective optimization. The proposed scheme is applied to the interactive user guided parameter optimization of hydraulic valve controllers. The tight integration of expert knowledge into the design and optimization process results in an efficient search of feasible solutions. The proposed interactive scheme extends the scope and applicability of computational intelligence aided design to mechatronic and engineering problems.

Robust Identification of Nonlinear Frictional Dynamics for Advanced Controller Design

Abstract The robust identification of nonlinear frictional dynamics constitutes a significant challenge for the model based friction compensation in advanced control of electro-mechanical drives. This paper discusses the frictional phenomena in pre-sliding and gross sliding regime described by the advanced General-Maxwell-Slip friction model and proposes a robust identification approach to determine their distributed parameters. The nonlinear dynamics of a drive chain with multiple frictional surfaces is described. The appropriate control signals are designed to excite the friction dynamics in both pre-sliding and sliding regimes. The estimation of friction parameters relies on the recursive least square (RLS) technique with a suitable set of regressors. The identified model is compared with experimental data and applied in a tunable framework for the model based friction compensation.

Preference modeling and model management for interactive multi-objective evolutionary optimization

Multiobjective optimization and decision making are strongly inter-related. This paper presents an interactive approach for the integration of expert preferences into multi-objective evolutionary optimization. The experts underlying preference is modeled only based on comparative queries that are designed to distinguish among the non-dominant solutions with minimal burden on the decision maker. The preference based approach constitutes a compromise between global approximation of a Pareto front and aggregation of objectives into a scalar utility function. The model captures relevant aspects of multi-objective decision making, such as preference handling, ambiguity and incommensurability. The efficiency of the approach in terms of number of expert decisions and convergence to the optimal solution are analyzed on the basis of an artificial decision behavior with respect to optimization benchmarks.

Interactive Incorporation of User Preferences in Multiobjective Evolutionary Algorithms

This paper proposes a novel interactive scheme to incorporate user preferences into evolutionary multiobjective optimization. The approach combines an evolutionary algorithm with an instance based supervised online learning scheme for user preference modeling. The user is queried to compare pairs of prototype solutions in terms of comparability and quality. The user decisions form a preference model which induces a ranking on the population. The model acts as a critic in the selection among non-dominated solutions on behalf of the expert. The user preference is extrapolated from a minimal number of pairwise comparisons to minimize the burden of interactive expert decisions. The preference model includes the concept of comparability to allow simultaneous convergence to multiple disconnected regions of the Pareto front. The preference model comprehends the specific preference scenarios of scalar optimization, goal oriented scenarios, ranking of criteria and global approximation of the Pareto front. It thus represents a general scheme for interactive optimization that does not depend on prior assumptions on either the problem or user preference structure.

Gaussian Mixture Models vs. Fuzzy Rule-Based Systems for Adaptive Meta-scheduling in Grid/Cloud Computing

Adaptive scheduling strategies are about considering the state of computational grids to obtain efficient and reliable schedules and to prevent the system performance deterioration. In this work, emerging adaptive strategies in grid computing, namely Fuzzy Rule-Based Systems (FRBS) -based strategies and a new adaptive scheduling approach, gaussian scheduling founded on Gaussian Mixture Models (GMMs) are compared. Both types of strategies focus on modeling the state of resources and select the most convenient site of the grid at every scheduling step given the current conditions. FRBSs provide a fuzzy characterization of the grid state and the inference of a suitability index based on their own knowledge given in the form of fuzzy IF-THEN rules. Besides, a GMM can be trained to model a complex probability density distribution indicating the suitability of every site in the grid to be the target of the schedule with the current conditions of its resources. This way the GMM scheduler assigns a probability to every state of the site where a higher probability is associated to a higher suitability of selection. Simulations based on real grid facilities are conducted to test the FRBS and GMM-based models and results are analyzed in terms of accuracy and convergence behaviour of their associated learning processes.

Structure and parameter identification of nonlinear systems with an evolution strategy

Modeling and identification of dynamic systems often is a prerequisite for the engineering of technical solutions, for example control system design. This paper presents an multi-objective evolutionary approach for identification of dynamic systems of variable structure. The evolutionary algorithm employs domain specific operators in order to evolve the block oriented structure of the model and simultaneously optimize its parameters. Based on the observed inputs and outputs the multi-objective method identifies an entire set of optimal compromise models which contrast model accuracy against complexity. The models are constructed from a set of basic blocks that capture phenomenons such as linear transfer functions, nonlinear gains and hysteresis that typically occur in mechanical, hydraulic and electrical systems. This representation enables the incorporation of domain knowledge in terms of building blocks and the interpretation of the identified model for further analysis and design. The feasibility of the proposed method is validated in the identification of an artificial dynamic system as well as a hydraulic proportional valve.

Struktur- und Parameteridentifikation mit Evolutionären Algorithmen

Zusammenfassung Modellierung und Identifikation dynamischer Systeme sind wichtige Aspekte der Regelungstechnik. Dieser Beitrag stellt einen Ansatz zur Optimierung von strukturvariablen Modellen dynamischer Systeme mit multikriteriellen evolutionären Algorithmen vor. Die Methode verwendet spezifische evolutionäre Operatoren zur simultanen Optimierung der Struktur eines Modells und ihrer Parameter. Basierend auf Ein- und Ausgangsdaten des realen Systems ermittelt die Strategie eine Menge von optimalen Modellen, die einen Kompromiss zwischen Qualität und Komplexität der Modellierung darstellen. Exemplarisch angewandt wird das Verfahren für die Identifikation eines künstlichen Testsystems sowie eines hydraulischen Proportionalventils und eines elastischen Roboterarms. Abstract Modeling and identification of dynamic systems are crucial in the fields of control engineering. This paper presents an approach for optimization of dynamic system models with variable structure using multi-objective evolutionary algorithms. The algorithm uses specific operators to simultaneously evolve the structure of a model and optimize its parameters. Based on input and output data of the real system the method aims at finding a set of optimal models that constitute a compromise between model quality and complexity. The method is exemplary applied to the identification of an artificial test system as well as of a hydraulic proportional valve and an elastic robot arm.

Multi-objective optimization with controlled model assisted evolution strategies

Evolutionary algorithms perform robust search in complex and high dimensional search spaces, but require a large number of fitness evaluations to approximate optimal solutions. These characteristics limit their potential for hardware in the loop optimization and problems that require extensive simulations and calculations. Evolutionary algorithms do not maintain their knowledge about the fitness function as they only store solutions of the current generation. In contrast, model assisted evolutionary algorithms utilize the information contained in previously evaluated solutions in terms of a data based model. The convergence of the evolutionary algorithm is improved as some selection decisions rely on the model rather than to invoke expensive evaluations of the true fitness function. The novelty of our scheme stems from the preselection of solutions based on an instance based fitness model, in which the selection pressure is adjusted to the quality of model. This so-called -control adapts the number of true fitness evaluations to the monitored model quality. Our method extends the previous approaches for model assisted scalar optimization to multi-objective problems by a proper redefinition of model quality and preselection pressure control. The analysis on multi-objective benchmark optimization problems not only confirms the superior convergence of the model assisted evolution strategy in comparison with a multi-objective evolution strategy but also the positive effect of regulated preselection in contrast to merely static preselection.

Active covariance matrix adaptation for multi-objective CMA-ES

This paper proposes a novel approach for a derandomized covariance matrix adaptation for multi-objective optimization. Common derandomized multi-objective algorithms only utilize the information gained from successful mutations. However in case of optimization problems with a limited budget for fitness evaluations inferior mutations provide additional information to adjust the search. The proposed algorithm, called active-(μ+λ)-MO-CMA-ES, extends previous approaches as it reduces the covariance along directions of unsuccessful mutations. In experiments on a set of commonly accepted multi-objective test problems the presented algorithm outperforms other derandomized evolution strategies.

Multiobjective evolutionary structural optimization for system identification and controller design

This paper presents a multiobjective evolutionary optimization scheme for incremental structural design of solutions. The approach is successfully applied to identification of nonlinear dynamic models and optimal controller design. The experimental results demonstrate that the approach simultaneously identifies the correct nonlinear structure in conjunction with its optimal parameters.