An integrated mechanistic-neural network modelling for granular systems

A hybrid neural network model is designed to predict the micro-macroscopic characteristics of particulate systems subjected to shearing. The network is initially trained to understand the micro-mechanical characteristics of particulate assemblies, by feeding the results based on three-dimensional discrete element simulations. Given the physical properties of the individual particles and the packing condition of the particulate assemblies under specified loading conditions, the network thus understands the way contact forces are distributed, the orientation of contact (fabric) networks and the evolution of stress tensor during the mechanical loading. These relationships are regarded as soft sensors. Using the signals received from soft sensors, a mechanistic neural network model is constructed to establish the relationship between the micro-macroscopic characteristics of granular assemblies subjected to shearing. The macroscopic results obtained form this hybrid mechanistic neural network modelling for data that were not part of the training signals, is compared with simulations based on discrete element modelling alone and in general, the agreement is good. The hybrid network responds to their inputs at a high speed and can be regarded as a real-time system for understanding the complex behaviour of particulate systems under mechanical process conditions.     

A multipurpose parallel genetic hybrid algorithm for non-linear non-convex programming problems

The genetic hybrid algorithm (GHA) is a general-purpose algorithm, spanning several areas of mathematical problem solving. GHA makes calls to an (empty) accelerator function at key stages of the solution process, providing it with the current population of solution vectors in the argument list of the function. On return from the accelerator, GHA processes the population further. The user has control over the specific stage (generation of a new population, crossover, mutation etc.) and can modify the population of solution vectors, e.g., by making calls to special purpose algorithms through the accelerator channel. If needed, the steps of GHA can be partly or completely superseded by the special purpose mathematical/artificial intelligence based algorithm. The system can be used as a package for classical mathematical programming with the genetic sub-block deactivated. On the other hand, the algorithm can be turned into a machinery for stochastic analysis (e.g. for Monte Carlo simulation, time series modelling or neural networks), where the mathematical programming and genetic computing facilities are deactivated. Finally, pure evolutionary computation can be activated for studying genetic phenomena. As a completely new feature, we design and implement a flexible multicomputer framework for the basic GHA.     

Integrated decision support for aviation safety inspectors

This paper presents a systems viewpoint for developing an advanced decision support system for aircraft safety inspectors. Research results from a Federal Aviation Administration (FAA) sponsored project to use neural network and expert systems technology to analyze aircraft maintenance databases are summarized. One of the main objectives of this research is to define more refined “alert” indicators for national comparison purposes that can signal potential problem areas by aircraft type for safety inspector consideration.

Integration aspects are addressed on two levels: (1) integration of the various technical components of the decision support system, and (2) integration of the decision support system with individual behavior, management systems and organizational structure, as well as corporate culture across both formal and informal dimensions. The paper summarizes the creation of strategic “inspection profiles” for aging aircraft and reliability curve fitting for structural components both based upon using neural network technology. Also, the potential use of a model-based expert system to facilitate field inspection diagnostics is presented. Finally, a framework for developing an intelligent decision system to support aircraft safety inspections is proposed that links expert systems, neural networks, as well as a paradigm of the decision making process typically used in unstructured situations.     

Some contributions with Petri nets for the modelling, analysis and control of HDS

Petri nets (PN) are useful for the modelling, analysis and control of hybrid dynamical systems (HDS) because PN combine in a comprehensive way discrete events and continuous behaviours. On one hand, PN are suitable for modelling the discrete part of HDS and for providing a discrete abstraction of continuous behaviours. On the other hand, continuous PN are suitable for modelling the continuous part of HDS and for working out a continuous approximation of the discrete part in order to avoid the complexity associated with the exponential growth of discrete states. This paper focuses on the advantages of PN as a modelling tool for HDS. Investigations of such models for diagnosis and control issues are detailed.

Taking inspiration from the discrete event approach, sensor selection for diagnosis is discussed according to the structural analysis of the PN models. Faults are represented with fault transitions and a faulty behaviour occurs when a sequence of transitions is fired that contains at least one fault transition. Minimal sets of observable places are defined for detecting and isolating faulty behaviours.

Taking inspiration from the continuous time approach, flow control of HDS modelled with continuous PN is also investigated. Gradient-based controllers are introduced in order to adapt the firing speeds of some controllable transitions according to a desired trajectory of the marking. The equilibria and stability of the controlled system are studied with Lyapunov functions.     

Approximating Unknown Mappings: An Experimental Evaluation

Different methodologies have been introduced in recent years with the aim of approximating unknown functions. Basically, these methodologies are general frameworks for representing non-linear mappings from several input variables to several output variables. Research into this problem occurs in applied mathematics (multivariate function approximation), statistics (nonparametric multiple regression) and computer science (neural networks). However, since these methodologies have been proposed in different fields, most of the previous papers treat them in isolation, ignoring contributions in the other areas. In this paper we consider five well known approaches for function approximation. Specifically we target polynomial approximation, general additive models (Gam), local regression (Loess), multivariate additive regression splines (Mars) and artificial neural networks (Ann).Neural networks can be viewed as models of real systems, built by tuning parameters known as weights. In training the net, the problem is to find the weights that optimize its performance (i.e. to minimize the error over the training set). Although the most popular method for Ann training is back propagation, other optimization methods based on metaheuristics have recently been adapted to this problem, outperforming classical approaches. In this paper we propose a short term memory tabu search method, coupled with path relinking and BFGS (a gradient-based local NLP solver) to provide high quality solutions to this problem. The experimentation with 15 functions previously reported shows that a feed-forward neural network with one hidden layer, trained with our procedure, can compete with the best-known approximating methods. The experimental results also show the effectiveness of a new mechanism to avoid overfitting in neural network training.     

Vibrational resonance in neuron populations with hybrid synapses

In this paper, vibrational resonance in excitable neuron populations with synapses is investigated by numerical simulation. In particular, the effect of the hybrid synapses on the signal detection and transmission in neural system is studied. Different topologies from regular and random networks to small-world networks are considered to analyze the dependence of vibrational resonance on the network structure and parameters. It is shown that there exists an optimal amplitude of high-frequency driving, enhancing the response of coupled neuron populations to a subthreshold signal. We find that chemical synaptic coupling is more efficient than the electrical coupling in signal detection and electrical synaptic coupling is better in signal transmission. Neuron populations with hybrid synapses compromise the merits of the two types of coupling and have an advantage in information communication.     

Crack Detection in Experimental Data by Means of a Spiking Response Model

In engineering tasks, multiple types of neural networks, such as e.g. feed-forward neural networks (FNN) or radial basis function neural networks (RBFN) are common solution methods for a wide scope of applications [1, 2]. Beside the different kinds of artificial neural networks, spiking neural networks (SNN) represent a continuous development in information processing within the computational units of a net. In the contribution, the Spiking Response Model (SRM) [3], a derivative of the Hodgkin-Huxley model, is utilized, whereas the specific properties of this neuron type are used in order to facilitate the evaluation of uniaxial tension test data sets of carbon reinforced concrete specimens. The evaluation procedure of the experimental data targets the identification of major crack appearance based on the load cell data. For reasons of comparison and standardization, numerous uniaxial experiments are performed within the collaborative research project Carbon Concrete Composite C³. Crack detection is considered as showcase for further development of evaluation methods based on SNNs with focal point to engineering experiments. Essential for the evaluation is the transition of the experimental data (load cell data) to the temporal domain of the neuron. Therefore, rate coding [4] is applied resulting in a pre-synaptic spike train. The actual detection of cracks in the pre-synaptic spike train is carried out by an evaluation neuron. The heuristic approach is validated by multiple examples and presents a sufficient accuracy as well as it retains the temporal information regarding the crack occurrence. A feed-forward network structure containing additional evaluation neurons in combination with extra experimental data (e.g. strain gauge data, load cell data) enables extended information extraction such as determining crack location with respect to the attached strain gauges. (© 2017 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)     

Stability and robustness of quasi-linear impulsive hybrid systems

Both hybrid dynamical systems and impulsive dynamical systems are studied extensively in the literature. However, impulsive hybrid systems are not yet well studied. Nonetheless, many physical systems exhibit both system switching and impulsive jump phenomena. This paper investigates stability and robust stability of a class of quasi-linear impulsive hybrid systems by using the methods of Lyapunov functions and Riccati inequalities. Sufficient conditions for stability and robust stability of those systems are established. Some examples are given to illustrate the applicability of our results.     

Intelligent systems in finance

Business sectors ranging from banking and insurance to retail, are benefiting from a whole new generation of ‘intelligent’ computing techniques. Successful applications include asset forecasting, credit evaluation, fraud detection, portfolio optimization, customer profiling, risk assessment, economic modelling, sales forecasting and retail outlet location. The techniques include expert systems, rule induction, fuzzy logic, neural networks and genetic algorithms, which in many cases are outperforming traditional statistical approaches. Their key features include the ability to recognize and classify patterns, learning from examples, generalization, logical reasoning from premises, adaptability and the ability to handle data which is incomplete, imprecise and noisy. This paper is the first in a series to appear in Applied Mathematical Finance;here we introduce the reader to the basic concepts of intelligent systems, describe their mode of operation and identify applications of the techniques in real world problem domains. Subsequent papers will concentrate on neural networks, genetic algorithms, fuzzy logic and hybrid systems, and will investigate their history and operation more rigorously.     

Learning of Chua's circuit attractors by locally recurrent neural networks

Many practical applications of neural networks require the identification of strongly non-linear (e.g., chaotic) systems. In this paper, locally recurrent neural networks (LRNNs) are used to learn the attractors of Chua's circuit, a paradigm for studying chaos. LRNNs are characterized by a feed-forward structure whose synapses between adjacent layers have taps and feedback connections. In general, the learning procedures of LRNNs are computationally simpler than those of globally recurrent networks. Results show that LRNNs can be trained to identify the underlying link among Chua's circuit state variables, and exhibit chaotic attractors under autonomous working conditions.     

Stochastic stabilization of hybrid neural networks by periodically intermittent control based on discrete-time state observations

This paper is concerned with stabilization of hybrid neural networks by intermittent control based on continuous or discrete-time state observations. By means of exponential martingale inequality and the ergodic property of the Markov chain, we establish a sufficient stability criterion on hybrid neural networks by intermittent control based on continuous-time state observations. Meantime, by M-matrix theory and comparison method, we show that hybrid neural networks can be stabilized by intermittent control based on discrete-time state observations. Finally, two examples are presented to illustrate our theory.     

Municipal Water Demand Forecasting: Tools for Intervention Time Series

This article introduces some approaches to common issues arising in real cases of water demand prediction. Occurrences of negative data gathered by the network metering system and demand changes due to closure of valves or changes in consumer behavior are considered. Artificial neural networks (ANNs) have a principal role modeling both circumstances. First, we propose the use of ANNs as a tool to reconstruct any anomalous time series information. Next, we use what we call interrupted neural networks (I-NN) as an alternative to more classical intervention ARIMA models. Besides, the use of hybrid models that combine not only the modeling ability of ARIMA to cope with the time series linear part, but also to explain nonlinearities found in their residuals, is proposed. These models have shown promising results when tested on a real database and represent a boost to the use and the applicability of ANNs.     

An airfoil optimization technique for wind turbines

Optimization algorithms coupled with computational fluid dynamics are used for wind turbines airfoils design. This differs from the traditional aerospace design process since the lift-to-drag ratio is the most important parameter and the angle of attack is large. Computational fluid dynamics simulations are performed with the incompressible Reynolds-averaged Navier–Stokes equations in steady state using a one equation turbulence model. A detailed validation of the simulations is presented and a computational domain larger than suggested in literature is shown to be necessary. Different approaches to parallelization of the computational code are addressed. Single and multiobjective genetic algorithms are employed and artificial neural networks are used as a surrogate model. The use of artificial neural networks is shown to reduce computational time by almost 50%.     

Numerical solution of a Fredholm integro-differential equation modelling neural networks

We compare piecewise linear and polynomial collocation approaches for the numerical solution of a Fredholm integro-differential equations modelling neural networks. Both approaches combine the use of Gaussian quadrature rules on an infinite interval of integration with interpolation to a uniformly distributed grid on a bounded interval. These methods are illustrated by numerical experiments on neural networks equations.     

Invariance principles for hybrid systems with memory

Hybrid systems with memory are dynamical systems exhibiting both delayed and hybrid dynamics. Such systems can be described by hybrid functional inclusions. Classical invariance principles play an instrumental role in proving stability and convergence of dynamical systems. Invariance principles for general hybrid systems with delays, however, remain an open topic. In this paper, we prove invariance principles for hybrid systems with memory, using both Lyapunov–Razumikhin function and Lyapunov–Krasovskii functional methods. These invariance principles are then applied to derive two stability results as corollaries.     

The impact of special days in call arrivals forecasting: A neural network approach to modelling special days

A key challenge for call centres remains the forecasting of high frequency call arrivals collected in hourly or shorter time buckets. In addition to the complex intraday, intraweek and intrayear seasonal cycles, call arrival data typically contain a large number of anomalous days, driven by the occurrence of holidays, special events, promotional activities and system failures. This study evaluates the use of a variety of univariate time series forecasting methods for forecasting intraday call arrivals in the presence of such outliers. Apart from established, statistical methods, we consider artificial neural networks (ANNs). Based on the modelling flexibility of the latter, we introduce and evaluate different methods to encode the outlying periods. Using intraday arrival series from a call centre operated by one of Europe’s leading entertainment companies, we provide new insights on the impact of outliers on the performance of established forecasting methods. Results show that ANNs forecast call centre data accurately, and are capable of modelling complex outliers using relatively simple outlier modelling approaches. We argue that the relative complexity of ANNs over standard statistical models is offset by the simplicity of coding multiple and unknown effects during outlying periods.     

Pricing and trading European options by combining artificial neural networks and parametric models with implied parameters

We compare the ability of the parametric Black and Scholes, Corrado and Su models, and Artificial Neural Networks to price European call options on the S&P 500 using daily data for the period January 1998 to August 2001. We use several historical and implied parameter measures. Beyond the standard neural networks, in our analysis we include hybrid networks that incorporate information from the parametric models. Our results are significant and differ from previous literature. We show that the Black and Scholes based hybrid artificial neural network models outperform the standard neural networks and the parametric ones. We also investigate the economic significance of the best models using trading strategies (extended with the Chen and Johnson modified hedging approach). We find that there exist profitable opportunities even in the presence of transaction costs.     

Artificial neural network modeling techniques applied to the hydrodesulfurization process

Reduction of harmful emissions in the combustion of fossil fuels imposes tighter specifications limiting the sulfur content of fuels. Hydrodesulfurization (HDS) is a key process in most petroleum refineries in which the sulfur is mostly eliminated. The modeling and simulation of the HDS process are necessary for a better understanding of the process operation; it is also a requirement to optimize process operation. The objective of this work is to explore the use of different artificial neural network (ANN) architectures in creating various models of the HDS process for the prediction of sulfur removal from naphtha. A database was build using daily records of the HDS process from a Mexican refinery. Accuracy of the predictions was quantified by the root of the mean squared difference between the measured and the predicted sulfur content in the desulfurized naphtha, along with the coefficient of correlation as a measure of the goodness of fit. Results show that the ANN models can be used as practical tools for predictive purposes. One particular example is the ability to anticipate such situations, in the process, that could increase alertness because some variables are deviating from acceptable limits.     

Two‐group classification using the Bayesian data reduction algorithm

The Bayesian data reduction algorithm (BDRA) is compared to traditional classification methods as well as feed forward artificial neural networks through a rigorous experiment. The BDRA performs comparably to alternative techniques and approaches theoretical optimal classification rates. Furthermore, it has a fundamentally different method for determining class membership. This study is novel in that it explores how the BDRA relates to established techniques, how it might be used in an explanatory manner, and how best to use it. © 2009 Wiley Periodicals, Inc. Complexity, 2010