Abstract and keywords
Abstract (English):
The review of methods for solving the forecasting problem on the basis of neural network technologies, neural network and fuzzy neural network, is carried out. The adaptive algorithm of fuzzy network self-organization is considered, which is used to determine the parameters and training of the network. The basis of the self-organization algorithm is the concept of grouping, i.e. clustering of data. In this, each cluster is associated with an inference rule of the fuzzy network. At the initial moment of time, the center of the cluster is set equal to the value of the first input vector. The amount of data belonging to one cluster depends on the Euclidean distance between the newly arrived input vector and the center of the cluster. Experimental studies are carried out to determine the optimal value of the Euclidean distance, providing the required approximation accuracy and acceptable computational complexity, as well as to determine the prediction period. The considered algorithm allows to determine in real time all the required parameters of the fuzzy network: the number of clusters, the center of each cluster and their powers, as well as the value of the accumulated function assigned to the cluster. The network uses a Gaussian membership function. The fuzzy forecasting model is built using a software package implemented in the Python programming language. The sample is constructed based on the given function. The network is trained on 50% of the whole compiled data sample and the remaining 50% is used for testing. A table with the results of comparative analysis of prediction error depending on the value of Euclidean distance and δ . The results of the study have shown the effectiveness of applying the adaptive fuzzy network self-organization algorithm for solving prediction problems.

Keywords:
FORECASTING, TIME SERIES, A FUZZY NEURAL NETWORK, SELF-ORGANIZATION ALGORITHM, CLUSTERS
Text
Text (PDF): Read Download
Login or Create
* Forgot password?