Francesco Barcellona

Forecasting Energy Commodity Prices Using Neural Networks

A new machine learning approach for price modeling is proposed. The use of neural networks as an advanced signal processing tool may be successfully used to model and forecast energy commodity prices, such as crude oil, coal, natural gas, and electricity prices. Energy commodities have shown explosive growth in the last decade. They have become a new asset class used also for investment purposes. This creates a huge demand for better modeling as what occurred in the stock markets in the 1970s. Their price behavior presents unique features causing complex dynamics whose prediction is regarded as a challenging task. The use of a Mixture of Gaussian neural network may provide significant improvements with respect to other well-known models. We propose a computationally efficient learning of this neural network using the maximum likelihood estimation approach to calibrate the parameters. The optimal model is identified using a hierarchical constructive procedure that progressively increases the model complexity. Extensive computer simulations validate the proposed approach and provide an accurate description of commodities prices dynamics.

Subband prediction of energy commodity prices

The dynamics of commodity prices has become a major field of analysis in the last 20 years. Standard econometric procedures to describe the behavior of prices have not been able to provide accurate description of the real dynamics. In this paper we apply filter banks to predict prices of specific energy commodities: crude oil, natural gas and electricity, which play a crucial role in the international economic and financial context. Given the high volatility of energy commodity prices, an accurate short term prediction allows to set adequate risk management strategies for producers, retailers and consumers. Filter banks for subband decompositions of the sequences to be predicted are proposed in the paper, allowing the implementation of a parallel computing system to get faster and more accurate implementation. The prediction system is based on a neural model trained on each subband according to specific training and prediction techniques.

A study on crude oil prices modeled by neurofuzzy networks

In the last decade the increasing volatility of petroleum markets has challenged time series analysts to produce highly predictive models. Crude Oil is a major driver of the global economy and its price fluctuations are a key indicator for producers, consumers and investors. With investors following the longerterm upward trend in Energy prices Commodity investments, we believe this will drive an increasing importance for methodologies like neurofuzzy networks for risk quantification, measurement and management. The data used is Crude Oil prices for both Brent and WTI in the 10 year period from 2001 to 2010. We will prove that the neurofuzzy approach based on ANFIS networks compare favorably with respect to other standard and neural models and it is able to achieve useful performances in terms of accurate prediction of prices and their probability distribution.

Neural network in modeling glucose-insulin behavior

In this paper we propose a neural network identification of a mathematical model called MINMOD, which describes the interactions between glucose and insulin in human subjects, in order to realize an adequate model for patients suffering from Diabetes Mellitus Type 2. The model has been tested on the basis of clinical data and it can correctly reproduce glucose and insulin reply and temporal evolution, according to experimental data test. Using neural networks, we can predict the glucose temporal evolution without invasive technique for patients, with the aim to determine the clinical effects to be made in case of pathological behaviors.

Neural Processing of Biomedical Data for Improving Driving Safety

Driving ability can be impaired by fatigue, drowsiness, drugs and alcohol, all of which have been implicated in causing road traffic accidents. Acute hypoglycaemia, which is the most common side effect of insulin therapy in individuals with diabetes, may also compromise driving skills. Other than by forbidding people to drive, the potential danger can be reduced by monitoring health and consciousness of drivers, by providing them with feedback on their conditions using, eventually, an emergency centre or biofeedback. In this paper, we propose the use of a signal processing system based on neural networks for system modelling and prediction. In particular, using neural networks we will reproduce the glucose temporal evolution without invasive technique for drivers, with the aim of preventing loss of consciousness while driving and hence improving road safety. Some illustrative trials will be shown in this regard. This research work is supported by the “CTL Excellence Centre (Centro di Ricerca sul Trasporto e la Logistica)” co-funded by the Italian Ministry of University, Education and Research and by the University of Rome “La Sapienza”.

Modeling Glucose-Insulin Behavior in Ill Patients (DM Type2)

In this paper we analyze a mathematical model (called MINMOD) that describes the interactions between glucose and insulin in human subjects, in order to realize an adequate model for ill patients, suffering from Diabetes Mellitus (DM) Type 2. Our model has been tested on the basis of clinical data and it has correctly reproduced glucose and insulin reply and temporal evolution, according to experimental data test. This model could, in the future, contribute to predict glucose and insulin behavior in ill patients and suggest the adequate treatment.