Neural-evolutionary modelling of polish electricity power exchange

摘要

The paper contains selected results of studies and research of hybrid modeling which is comprised from neural modeling and evolutionary modeling. Neural modeling was focused on designing and teaching the Artificial Neural Network (ANN) of the Polish Electricity Power Exchange (PEPE) on the example of the data for the period from 01.01.2015 to 06.30.2015 of the Next Day Market, evolutionary modeling was used for improving the neural model parameters by using authorial modified System Evolutionary Algorithm (SEA). The paper also contains the results of simulation and sensitivity tests for neural-evolutionary model implemented in MATLAB. Modification of the evolutionary algorithm were treated, among the others, as the systemic approach to the initial population (IP) composed with two matrix weights and developed the system adaptation function (vigor function) defined as a discrepancy between the real vector output value (target values) and values of individual outputs represented by individual chromosomes (model output values). SEA algorithm implemented in MATLAB was used to improve the parameters of the PEPE neural model. Comparing the neural model to neuro-evolution model showed the improvement of the PEPE neural model on the level of 0.02%. Newly implemented in Simulink model was conducted for further research and comparative studies of sensitivity.