Evolutionary Learning Processes to Design the Dilation-Erosion Perceptron for Weather Forecasting

• Authors: Ricardo A. Araújo
• Venue: Neural Processing Letters
• Year: 2013

The weather forecasting is considered a rather difficult problem due to many complex features present in these time series. Several techniques have been proposed in the literature to solve this problem. In particular, the dilation-erosion perceptron (DEP), a model whose foundations are based on mathematical morphology and complete lattice theory, has been successfully used for time series forecasting. However, a drawback arises from the gradient estimation of morphological operators in the classical gradient-based learning process of the DEP, since they are not differentiable of usual way. In this sense, this work presents evolutionary learning processes, using a modified genetic algorithm, a particle swarm optimization, a modified differential evolution and a covariance matrix adaptation evolutionary strategy, to design the DEP model for weather forecasting. In addition, into the proposed learning processes we have included an automatic correction step that is geared toward eliminating time phase distortions that occur in some weather phenomena. An experimental analysis is presented using three non-linear forecasting problems from the Brazilian weather, and the obtained results are discussed and compared, according to five well-known performance metrics and an evaluation function, to results found using the DEP model with its classical gradient-based learning process.