Model Synthesis Algorithms for a Monitoring Software Agent

The article examines current algorithms for synthesizing models in order to usage in monitoring software agents. Neural network and statistical models, which use additional (exogenous) variables, were tested for time series forecasting. The focus was on predicting the number of COVID-19 cases in the Vinnytsia region in 2021. Various neural network models are considered and tested, including those based on a multi-layer perceptron, recurrent neural networks, convolutional neural networks, and the Transformer architecture. The study compared models that make predictions for the entire horizon at once to those that predict each day of the horizon separately. Also, in study was compared error metrics of usage neural network models and statistical methods for solution time-series forecasting task. The result of the study is a set of model synthesis algorithms that can be used in monitoring software agents based on the analysis of obtained error metrics for each day of the forecasting horizon.