Intelligent Agent for Weather Parameters Prediction

Abstract

The paper shows how the typical and not sophisticated topology of the neural network trained by easily implemented gradient method can fulfil the practical needs of the intelligent agent to be useful for weather parameters prediction. If we are able to accumulate the significant set of weather events recording temperature, atmospheric pressure, wind speed, etc. we have the real input for correct prediction in the future. The size of the training vectors can be limited as well as the number of the training epochs. Better results of prediction we can expect when we use the combination of weather events for the training vectors creation. It is possible to create the type of intelligent agent to predict the value of the weather parameters with acceptable low-level error at different climate zones. This way the idea of the weather Complex Event Processing systems seems to be sensible where Event Processing Agents (EPAs) can typical sensors to test the values of the weather parameters as well as intelligent tools created based on big data sets stored year by year.