Um Modelo Estocástico baseado em Redes Neurais

his paper presents the proposal of a generic model of stochastic process based on neural networks, called Neural Stochastic Process (NSP). The proposal model can be applied to problems involving phenomena of stochastic behavior and / or periodic features. The NSP’s neural networks capture the behavior of the time series of these phenomena without requiring a priori information about the series, and generates synthetic time series with the same probabilities as the historical series. The NSP model was applied to the treatment of monthly inflows series and the results indicated that the synthetic series generated exhibit statistical characteristics similar to historical series