Applications of Data-Driven Modelling and Machine Learning in Control of Water Resources

Traditionally, management and control of water resources is based on behavior-driven or physically based models based on equations describing the behavior of water bodies. Since recently models built on the basis of large amounts of collected data are gaining popularity. This modeling approach we will call data-driven modeling; it borrows methods from various areas related to computational intelligence—machine learning, data mining, soft computing, etc. The chapter gives an overview of successful applications of several data-driven techniques in the problems of water resources management and control. The list of such applications includes: using decision trees in classifying flood conditions and water levels in the coastal zone depending on the hydrometeorological data, using artificial neural networks (ANN) and fuzzy rule-based systems for building controllers for real-time control of water resources, using ANNs and M5 model trees in flood control, using chaos theory in predicting water levels for ship guidance, etc. Conclusions are drawn on the applicability of the mentioned methods and the future role of computational intelligence in modeling and control of water resources.

Bayesian Agencies in Control

Most agent architectures implement autonomous agents that use extensive interaction protocols and social laws to control interactions in order to ensure that the correct behaviors result during run-time. These agents, organized into multi-agent systems in which all agents adhere to predefined interaction protocols, are well suited to the analysis, design and implementation of complex systems in environments where it is possible to predict interactions during the analysis and design phases. In these multi-agent systems, intelligence resides in individual autonomous agents, rather than in the collective behavior of the individual agents. These agents are commonly referred to as “next-generation” or intelligent components, which are difficult to implement using current component-based architectures. In most distributed environments, such as the Internet, it is not possible to predict interactions during analysis and design. For a complex system to be able to adapt in such an uncertain and non-deterministic environment, we propose the use of agencies, consisting of simple agents, which use probabilistic reasoning to adapt to their environment. Our agents collectively implement distributed Bayesian networks, used by the agencies to control behaviors in response to environmental states. Each agency is responsible for one or more behaviors, and the agencies are structured into heterarchies according to the topology of the underlying Bayesian networks. We refer to our agents and agencies as “Bayesian agents” and “Bayesian agencies.”

Integrating Genetic Algorithms and the Finite Element Analysis for Structural Inverse Problems

A general method for integrating genetic algorithms within a commercially available finite element (FE) package to solve a range of structural inverse problems is presented. The described method exploits a user-programmable interface to control the genetic algorithm from within the FE package. This general approach is presented with specific reference to three illustrative system identification problems. In two of these the aim is to deduce the damaged state of composite structures from a known physical response to a given static loading. In the third the manufactured lay-up of a composite component is designed using the proposed methodology.

The Effect of Multi-Parent Recombination on Evolution Strategies for Noisy Objective Functions

In this chapter we apply (m / m, l)-ES to noisy test functions, in order to investigate the effect of multi-parent versions of both intermediate recombination and discrete recombination. Among the many formulations of ES, we test three in particular; Classical-ES (CES), i.e., Schwefel’s original ES (Schwefel, 1995, Bäck, 1996); Fast-ES (FES), i.e., Yao and Liu’s extended ES (Yao & Liu, 1997); and Robust-ES (RES), i.e., our extended ES (Ohkura, 2001). Computer simulations are used to compare the performance of multi-parent versions of intermediate recombination and discrete recombination in CES, FES and RES. We saw that the performance of the (m / m, l)-ES algorithms depended on the particular objective functions. However, the FES and RES algorithms were seen to be improved by multi-parent versions of discrete recombination applied to both object parameters and strategy parameters.

On the Modelling of a Human Pilot Using Fuzzy Logic Control

This chapter discusses the possibility to model the control behaviour of a human pilot by fuzzy logic control. For this investigation a special flight task is considered, the ILS tracking task, and an evaluation pilot has to perform this task in a ground based flight simulator. During the ILS tracking task all necessary flight data are stored in a database and additionally the pilot commands are recorded. The development of the described fuzzy controller (the fuzzy pilot) is based on cognitive analysis by evaluating the recorded flight data with the associated pilot comments. Finally the fuzzy pilot is compared with the human pilot and it can be verified that the fuzzy pilot and the human pilot are based on the same control concept.

Designing Neural Network Ensembles by Minimizing Mutual Information

This chapter describes negative correlation learning for designing neural network ensembles. Negative correlation learning has been firstly analysed in terms of minimising mutual information on a regression task. By minimising the mutual information between variables extracted by two neural networks, they are forced to convey different information about some features of their input. Based on the decision boundaries and correct response sets, negative correlation learning has been further studied on two pattern classification problems. The purpose of examining the decision boundaries and the correct response sets is not only to illustrate the learning behavior of negative correlation learning, but also to cast light on how to design more effective neural network ensembles. The experimental results showed the decision boundary of the trained neural network ensemble by negative correlation learning is almost as good as the optimum decision boundary.

Design Wind Speeds Using Fast Fourier Transform

In this chapter, a study on the effects of transforming wind speed data, from a time series domain into a frequency domain via Fast Fourier Transform (FFT), is presented. The wind data is first transformed into a stationary pattern from a non-stationary pattern of time series data using statistical software. This set of time series is then transformed using FFT for the main purpose of the chapter. The analysis is done through MATLAB software, which provides a very useful function in FFT algorithm. Parameters of engineering significance such as hidden periodicities, frequency components, absolute magnitude and phase of the transformed data, power spectral density and cross spectral density can be obtained. Results obtained using data from case studies involving thirty-one weather stations in Malaysia show great potential for application in verifying the current criteria used for design practices.

Evolutionary Learning of a Box-Pushing Controller

In our research we use evolutionary algorithms to evolve robot controllers for executing elementary behaviours. This chapter focuses on the behaviour of pushing a box between two walls. The main research question addressed in this chapter is: how can a neural network learn to control the box-pushing task using evolutionary-computation techniques? In answering this question we study the following three characteristics by means of simulation experiments: (1) the fitness function, (2) the neural network topology and (3) the parameters of the evolutionary algorithm. We find that appropriate choices for these characteristics are: (1) a global external fitness function, (2) a recurrent neural network, and (3) a regular evolutionary algorithm augmented with the doping technique in which the initial population is supplied with a solution to a hard task instance. We conclude by stating that our findings on the relatively simple box-pushing behaviour form a good starting point for the evolutionary learning of more complex behaviours.

On Measuring the Attributes of Evolutionary Algorithms

In this chapter we compare the performance of two contrasting evolutionary algorithms addressing a similar problem, of information retrieval. The first, BTGP, is based upon genetic programming, while the second, MGA, is a genetic algorithm. We analyze the performance of these evolutionary algorithms through aspects of the evolutionary process they undergo while filtering information. We measure aspects of the variation existing in the population undergoing evolution, as well as properties of the selection process. We also measure properties of the adaptive landscape in each algorithm, and quantify the importance of neutral evolution for each algorithm. We choose measures of these properties because they appear generally important in evolution. Our results indicate why each algorithm is effective at information retrieval, however they do not provide a means of quantifying the relative effectiveness of each algorithm. We attribute this difficulty to the lack of appropriate measures available to measure properties of evolutionary algorithms, and suggest some criteria for useful evolutionary measures to be developed in the future.

Solving Two Multi-Objective Optimization Problems Using Evolutionary Algorithm

Being capable of finding a set of pareto-optimal solutions in a single run is a necessary feature for multi-criteria decision making, Evolutionary algorithms (EAs) have attracted many researchers and practitioners to address the solution of Multi-objective Optimization Problems (MOPs). In a previous work, we developed a Pareto Differential Evolution (PDE) algorithm to handle multi-objective optimization problems. Despite the overwhelming number of Multi-objective Evolutionary Algorithms (MEAs) in the literature, little work has been done to identify the best MEA using an appropriate assessment methodology. In this chapter, we compare our algorithm with twelve other well-known MEAs, using a popular assessment methodology, by solving two benchmark problems. The comparison shows the superiority of our algorithm over others.