Management of a cyber-physical system in an open environment based on the monitoring data analysis

A method of monitoring data analysis focused on the formation of control actions within cyberphysical systems has been developed and substantiated. The study is conditioned by the need to increase the efficiency of auto-mated control systems through the introduction of intelligent information processing methods that ensure adaptation to changing conditions of the open environment. The proposed method is based on the sequential execution of three stages: normalization of incoming data, prediction of system parameters using the architecture of the recurrent neural network LSTM and adaptation of control actions using the gradient method. As an example of the application of the developed method, a monitoring system for the parameters of the aquatic environment in fisheries was considered, where indicators of water temperature, pH level and dissolved oxygen content were collected. The experimental data obtained made it possible to train the model and implement a mechanism for automatic correction of control parameters in a mode close to real time. The quality of forecasting was assessed using the MAE and MAPE metrics, the values of which were less than 0.2%, which indicates the high accuracy of the model. The development also includes a module of empirical control rules that implement the logical interpretation of possible deviations and the formation of appropriate control signals. Field tests conducted at pond farming enterprises confirmed the effectiveness of the proposed approach: an increase in control accuracy of up to 86% was recorded, a 9% reduction in resource consumption and a 95% reduction in monitoring time. The developed method can be scaled and adapted for use in other fields: industrial automation, environmental monitoring, intelligent transport systems, as well as in the infrastructure of “smart” cities, which confirms its versatility and scientific novelty.