Hugo Valadares Siqueira

UNORGANIZED MACHINES FOR SEASONAL STREAMFLOW SERIES FORECASTING

Modern unorganized machines--extreme learning machines and echo state networks--provide an elegant balance between processing capability and mathematical simplicity, circumventing the difficulties associated with the conventional training approaches of feedforward/recurrent neural networks (FNNs/RNNs). This work performs a detailed investigation of the applicability of unorganized architectures to the problem of seasonal streamflow series forecasting, considering scenarios associated with four Brazilian hydroelectric plants and four distinct prediction horizons. Experimental results indicate the pertinence of these models to the focused task.

Echo state networks and extreme learning machines: a comparative study on seasonal streamflow series prediction

Extreme Learning Machines (ELMs) and Echo State Networks (ESNs) represent promising alternatives in time series forecasting in view of their intrinsic trade-off between performance and mathematical tractability. Both approaches share a key feature: their supervised parameter adaptation is restricted to the output layer, the remaining synaptic weights being chosen according to a priori unsupervised schemes. This work performs a comparative investigation regarding the performances of a classic ELM and ESNs in the context of the prediction of monthly seasonal streamflow series associated with Brazilian hydroelectric plants. With respect to the ESN, two possible reservoir design approaches are tested, as well as the novel architecture of Boccato et al., which is characterized by the use a Volterra filter and PCA in the readout. Additionally, a MLP is included to establish a base for comparison. Results show the relevance of these architectures in modeling seasonal streamflow series.

Assessing the impact of PM2.5 on respiratory disease using artificial neural networks

Understanding the impact on human health during peak episodes in air pollution is invaluable for policymakers. Particles less than PM2.5 can penetrate the respiratory system, causing cardiopulmonary and other systemic diseases. Statistical regression models are usually used to assess air pollution impacts on human health. However, when there are databases missing, linear statistical regression may not process well and alternative data processing should be considered. Nonlinear Artificial Neural Networks (ANN) are not employed to research environmental health pollution even though another advantage in using ANN is that the output data can be expressed as the number of hospital admissions. This research applied ANN to assess the impact of air pollution on human health. Three well-known ANN were tested: Multilayer Perceptron (MLP), Extreme Learning Machines (ELM) and Echo State Networks (ESN), to assess the influence of PM2.5, temperature, and relative humidity on hospital admissions due to respiratory diseases. Daily PM2.5 levels were monitored, and hospital admissions for respiratory illness were obtained, from the Brazilian hospital information system for all ages during two sampling campaigns (2008-2011 and 2014-2015) in Curitiba, Brazil. During these periods, the daily number of hospital admissions ranged from 2 to 55, PM2.5 concentrations varied from 0.98 to 54.2 μg m-3, temperature ranged from 8 to 26 °C, and relative humidity ranged from 45 to 100%. Of the ANN used in this study, MLP gave the best results showing a significant influence of PM2.5, temperature and humidity on hospital attendance after one day of exposure. The Anova Friedmans test showed statistical difference between the appliance of each ANN model (p < .001) for 1 lag day between PM2.5 exposure and hospital admission. ANN could be a more sensitive method than statistical regression models for assessing the effects of air pollution on respiratory health, and especially useful when there is limited data available.

Echo state networks for seasonal streamflow series forecasting

The prediction of seasonal streamflow series is very important in countries where power generation is predominantly done by hydroelectric plants. Echo state networks can be safely regarded as promising tools in forecasting because they are recurrent networks that have a simple and efficient training process based on linear regression. Recently, Boccato et al. proposed a new architecture in which the output layer is built using a principal component analysis and a Volterra filter. This work performs a comparative investigation between the performances of different ESNs in the context of the forecasting of seasonal streamflow series associated with Brazilian hydroelectric plants. Two possible reservoir design approaches were tested with the classical and the Volterra-based output layer structures, and a multilayer perceptron was also included to establish bases for comparison. The obtained results show the relevance of these networks and also contribute to a better understanding of their applicability to forecasting problems.

Performance analysis of unorganized machines in streamflow forecasting of Brazilian plants

Abstract This work performs an extensive investigation about the application of unorganized machines – extreme learning machines and echo state networks – to predict monthly seasonal streamflow series, associated to three important Brazilian hydroelectric plants, for many forecasting horizons. The aforementioned models are neural network architectures which present efficient and simple training processes. Moreover, the selection of the best inputs of each model is carried out by the wrapper method, using three different evaluation criteria, and three filters, viz., those based on the partial autocorrelation function, the mutual information and the normalization of maximum relevance and minimum common redundancy method. This study also establishes a comparison between the unorganized machines and two classical models: the partial autoregressive model and the multilayer perceptron. The computational results demonstrate that the unorganized machines, especially the echo state networks, represent efficient alternatives to solve the task.

Neural Networks for Predicting Prices of Sugarcane Derivatives

Brazil is one of the largest sugarcane producers. It directly affects job creation and national gross domestic product, as well as bringing a large amount of foreign money to the country. Hence, this paper proposes investigating the application of neural networks—multilayer perceptron, extreme learning machines (ELM) and echo state networks (ESN)—for predicting the following cane derivatives prices: sugar, hydrous ethanol, and anhydrous ethanol. The main characteristic of ELM and ESN is their simple and fast training process, being based on the analytic calculation of the coefficients of a linear combiner. Their intermediate layer stands untuned, and the weights in this layer are randomly and independently defined. The computational results show that the application of the ELM achieved the best overall results, showing that they are viable candidates for these kinds of problems.

Unorganized neural networks applied to streamflow forecasting of Passo Real hydroelectric plant

Unorganized neural networks — or unorganized machines - are recent developed architectures in the field of computational intelligence, in which the supervised tuning of the free parameters is restricted to the weights of the output layer, by means of a linear least square solution. The remaining weights are randomly generated and stand untrained which become the adjustment process simple and fast. In this work, we considered the Extreme Learning Machines (ELMs) and Echo State Networks (ESNs) to predict the monthly seasonal streamflow series associated to Passo Real hydroelectric plant, located in Brazil. In view of to establish a performance comparison, a periodic autoregressive model was developed. The computational results obtained show the relevance of the proposed networks to solve the problem, extending the possibility of application of the unorganized networks.

Unorganized machines to predict hospital admissions for respiratory diseases

The impact of air pollution on human health is widely studied, especially at large cities in the world. The most used methodologies to access this impact are based on statistical regressions, such as Generalized Linear Models. A few researchers use neural networks. In this context, the present work intend to predict the number of hospital admissions for respiratory diseases in Campinas city, São Paulo State, Brazil, related to the concentration of particulate matter less than 10 µm in aerodynamic diameter (PM10) using neural networks. The used data set comprises daily measures of average temperature, relative humidity and PM10 concentration scattered in the air during 2007 to 2009. To do so, it is proposed the use of two neural networks architectures known as unorganized machines: the echo state networks and the extreme learning machines. The obtained computational results were compared to the performance of the multilayer perceptron. In general, it may be concluded that the proposed neural networks showed to be more relevant to solve problems of health risks, contributing to a better understanding of the applicability of these networks.

Modelos Lineares Realimentados de Previsão: Um Estudo Utilizando Algoritmos Evolucionários

Resumo A previsão das séries de vazões a usinas hidrelétricas é de vital importância para o planejamento energético em países como o Brasil, que possuem um parque gerador predominantemente hidráulico. Técnicas lineares são bastante utilizadas no âmbito desse problema. Desse conjunto de modelos possíveis, os modelos realimentados surgem como uma alternativa no ensejo de obter modelos mais acurados para fins de previsão e embora o processo de estimação dos parâmetros desse tipo de modelos seja significativamente mais complexo. Visando contribuir com o desenvolvimento de técnicas de otimização adequada a este problema, o presente trabalho busca a análise de algoritmos bio-inspirados, como técnicas de estimação de parâmetros de modelos ARMA e de filtros lineares com resposta ao impulso infinita. O estudo contempla três algoritmos bio-inspirados: algoritmo genético e duas propostas de algoritmos imunológicos, uma baseada em pequenas alterações do CLONALG e a opt-aiNet. Os resultados indicam a existência, do ponto de vista de otimização, de um ganho de desempenho trazido pelas meta-heurísticas bio-inspiradas e, do ponto de vista estrutural, revelam a validade da adoção de estruturas recorrentes.