Juan J. Flores

The application of artificial neural networks to the analysis of remotely sensed data

Artificial neural networks (ANNs) have become a popular tool in the analysis of remotely sensed data. Although significant progress has been made in image classification based upon neural networks, a number of issues remain to be resolved. This paper reviews remotely sensed data analysis with neural networks. First, we present an overview of the main concepts underlying ANNs, including the main architectures and learning algorithms. Then, the main tasks that involve ANNs in remote sensing are described. The limitations and crucial issues relating to the application of the neural network approach are discussed. A brief review of the implementation of ANNs in some of the most popular image processing software packages is presented. Finally, we discuss the application perspectives of neural networks in remote sensing image analysis.

Evaluation of a lime-mediated sewage sludge stabilisation process. Product characterisation and technological validation for its use in the cement industry

This paper describes an industrial process for stabilising sewage sludge (SS) with lime and evaluates the viability of the stabilised product, denominated Neutral, as a raw material for the cement industry. Lime not only stabilised the sludge, raised the temperature of the mix to 80-100°C, furthering water evaporation, portlandite formation and the partial oxidation of the organic matter present in the sludge. Process mass and energy balances were determined. Neutral, a white powder consisting of portlandite (49.8%), calcite (16.6%), inorganic oxides (13.4%) and organic matter and moisture (20.2%), proved to be technologically apt for inclusion as a component in cement raw mixes. In this study, it was used instead of limestone in raw mixes clinkerised at 1400, 1450 and 1500°C. These raw meals exhibited greater reactivity at high temperatures than the limestone product and their calcination at 1500°C yielded clinker containing over 75% calcium silicates, the key phases in Portland clinker. Finally, the two types of raw meal (Neutral and limestone) were observed to exhibit similar mineralogy and crystal size and distribution.

Solving a School Timetabling Problem Using a Bee Algorithm

The timetabling problem consists in fixing a sequence of meetings between teachers and students in a given period of time, satisfying a set of different constraints. This paper shows the implementation of a Bee Algorithm (BA) to solve the Scholar Timetabling Problem. In the implemented BA, scout bees find feasible solutions while collector bees search in their neighborhood to find better solutions. While other algorithms evaluate every plausible assignment, the implemented algorithm only evaluates feasible solutions. This approach seems to be helpful to manage constrained problems. We propose a new measurement for replacing population that considers the evolutionary history of the bees as well as their fitness. Experimental results are presented for two real schools, where the algorithm shows promising results.

Models of performance of evolutionary program induction algorithms based on indicators of problem difficulty

Modeling the behavior of algorithms is the realm of evolutionary algorithm theory. From a practitioners point of view, theory must provide some guidelines regarding which algorithm/parameters to use in order to solve a particular problem. Unfortunately, most theoretical models of evolutionary algorithms are difficult to apply to realistic situations. However, in recent work (Graff and Poli, 2008, 2010), where we developed a method to practically estimate the performance of evolutionary program-induction algorithms (EPAs), we started addressing this issue. The method was quite general; however, it suffered from some limitations: it required the identification of a set of reference problems, it required hand picking a distance measure in each particular domain, and the resulting models were opaque, typically being linear combinations of 100 features or more. In this paper, we propose a significant improvement of this technique that overcomes the three limitations of our previous method. We achieve this through the use of a novel set of features for assessing problem difficulty for EPAs which are very general, essentially based on the notion of finite difference. To show the capabilities or our technique and to compare it with our previous performance models, we create models for the same two important classes of problems—symbolic regression on rational functions and Boolean function induction—used in our previous work. We model a variety of EPAs. The comparison showed that for the majority of the algorithms and problem classes, the new method produced much simpler and more accurate models than before. To further illustrate the practicality of the technique and its generality (beyond EPAs), we have also used it to predict the performance of both autoregressive models and EPAs on the problem of wind speed forecasting, obtaining simpler and more accurate models that outperform in all cases our previous performance models.

Gravitational interactions optimization

Evolutionary computation is inspired by nature in order to formulate metaheuristics capable to optimize several kinds of problems. A family of algorithms has emerged based on this idea; e.g. genetic algorithms, evolutionary strategies, particle swarm optimization (PSO), ant colony optimization (ACO), etc. In this paper we show a population-based metaheuristic inspired on the gravitational forces produced by the interaction of the masses of a set of bodies. We explored the physics knowledge in order to find useful analogies to design an optimization metaheuristic. The proposed algorithm is capable to find the optima of unimodal and multimodal functions commonly used to benchmark evolutionary algorithms. We show that the proposed algorithm (Gravitational Interactions Optimization - GIO) works and outperforms PSO with niches in both cases. Our algorithm does not depend on a radius parameter and does not need to use niches to solve multimodal problems. We compare GIO with other metaheuristics with respect to the mean number of evaluations needed to find the optima.

Wind Speed Forecasting Using a Hybrid Neural-Evolutive Approach

The design of models for time series prediction has found a solid foundation on statistics. Recently, artificial neural networks have been a good choice as approximators to model and forecast time series. Designing a neural network that provides a good approximation is an optimization problem. Given the many parameters to choose from in the design of a neural network, the search space in this design task is enormous. When designing a neural network by hand, scientists can only try a few of them, selecting the best one of the set they tested. In this paper we present a hybrid approach that uses evolutionary computation to produce a complete design of a neural network for modeling and forecasting time series. The resulting models have proven to be better than the ARIMA and the hand-made artificial neural network models.

Time-Invariant Dynamic Systems identification based on the qualitative features of the response

The problem of Systems Identification starts with a time-series of observed data and tries to determine the simplest model capable of exhibiting the observed behavior. This optimization problem searches the model from a space of possible models. In this paper, we present the theory and algorithms to perform Qualitative and Quantitative Systems Identification for Linear Time-Invariant Dynamic Systems. The methods described here are based on successive elimination of the components of the systems response. Sinusoidals of high frequencies are eliminated first, then their carrying waves. We continue with the process until we obtain a non-oscillatory carrier. At this point, we determine the order of the carrier. This procedure allows us to determine how many sinusoidal components and exponential components are found in the impulse response of the system under study. The number of components determines the order of the system. The paper is composed of two important parts, the statement of some mathematical properties of the responses of Linear Time Invariant Dynamic Systems, and the proposal of a set of filters that allows us to implement the recognition algorithm. We present the application of the proposed methodology to analyze and model the electrical circuits and electrical power systems.

Evolutionary computation solutions to the circle packing problem

In this work, we present an evolutionary omputation-based solution to the circle packing problem (ECPP). The circle packing problem consists of placing a set of circles into a larger containing circle without overlaps: a problem known to be NP-hard. Given the impossibility to solve this problem efficiently, traditional and heuristic methods have been proposed to solve it. A naïve representation for chromosomes in a population-based heuristic search leads to high probabilities of violation of the problem constraints, i.e., overlapping. To convert solutions that violate constraints into ones that do not (i.e., feasible solutions), in this paper we propose two repair mechanisms. The first one considers every circle as an elastic ring and overlaps create repulsion forces that lead the circles to positions where the overlaps are resolved. The second one forms a Delaunay triangulation with the circle centers and repairs the circles in each triangle at a time, making sure repaired triangles are not modified later on. Based on the proposed repair heuristics, we present the results of the solution to the CPP problem to a set of unit circle problems (whose exact optimal solutions are known). These benchmark problems are solved using genetic algorithms, evolutionary strategies, particle swarm optimization, and differential evolution. The performance of the solutions is compared to those known solutions based on the packing density. We then perform a series of experiments to determine the performance of ECPP with non-unitary circles. First, we compare ECPP’s results to those of a public competition, which stand as the world record for that particular instance of the non-unitary CPP. On a second set of experiments, we control the variance of the size of the circles. In all experiments, ECPP yields satisfactory near-optimal solutions.

k-Nearest-Neighbor by Differential Evolution for Time Series Forecasting

A framework for time series forecasting that integrates k-Nearest-Neighbors (kNN) and Differential Evolution (DE) is proposed. The methodology called NNDEF (Nearest Neighbor - Differential Evolution Forecasting) is based on knowledge shared from nearest neighbors with previous similar behaviour, which are then taken into account to forecast. NNDEF relies on the assumption that observations in the past similar to the present ones are also likely to have similar outcomes. The main advantages of NNDEF are the ability to predict complex nonlinear behavior and handling large amounts of data. Experiments have shown that DE can optimize the parameters of kNN and improve the accuracy of the predictions.

Reducing the search space in evolutive design of ARIMA and ANN models for time series prediction

Evolutionary design of time series predictors is a field that has been explored for several years now. The levels of design vary in the many works reported in the field. We decided to perform a complete design and training of ARIMA models using Evolutionary Computation. This decision leads to high dimensional search spaces, whose size increases exponentially with dimensionality. In order to reduce the size of those search spaces we propose a method that performs a preliminary statistical analysis of the inputs involved in the model design and their impact on quality of results; as a result of the statistical analysis, we eliminate inputs that are irrelevant for the prediction task. The proposed methodology proves to be effective and efficient, given that the results increase in accuracy and the computing time required to produce the predictors decreases.