استخدام السلاسل الزمنية لمخرجات التحليل العنقودي مع التطبيق العملي

الملخص EN

Cluster analysis in time series data is one of the important topics in analyzing data and finding similar trends in time series, which represents a major challenge in various fields.

Scientists have increased interest in studying time series data clustering, as it has proven its effectiveness in providing important information in various fields.

Clustering time-series data to facilitate prediction of formed clusters and exploit time and effort.

Clustering time-series data has been used in various scientific fields to discover patterns that enable data analysts to extract valuable information from a large and complex data set.

Homogeneous clusters are grouped together on the basis of a specific similarity measure.

The monthly data of electrical energy productivity in Kirkuk was used to study its temporal behavior.

The hierarchical clustering method was used, and the method adopted in the linkage method is the hierarchical linking method, the ward's method, based on the similarity matrix, and we adopted the city-block distance scale manhaten) distance to find the similarity matrix between clusters and in order to reach homogeneous groups (clusters) that have common characteristics based on their productivity, hierarchical clustering, tree diagramming and prediction of future values of cluster productivity are used.

The most important findings of the research are the formation of four clusters and the construction of a time series model for each cluster, and through the analysis of the series, it was found that it is unstable and not random, and for the purpose of achieving stability and randomness, the necessary conversions were made, and the differentiation criteria (Akaik, Information Criteria: AIC) were used.

(Schwartz Bayesian Criteria: SBIC), (Hanna-Quinn Criterion: H-Q), (Root Mean Sguare Error RMSE).

To diagnose the significant models to choose the appropriate and efficient model, the prediction for the first cluster is in the ARIMA(0, 1, 1) model, the model that was used for the second cluster is SARIMA(2, 0, 0)x(1, 1, 2)12, the third cluster due to stopping the units About production since 2014 and until now the units have been idle, meaning that they do not produce electricity as a basis for leaving the rehabilitation works, the future values predicted here are zero, the fourth cluster was used ARIMA (2, 1, 0) model and the predictions were good and close to reality for the period from November 2020 to November 2022 for a period of two years.