Abstract
The relevance of the research lies in the need to create a digital platform for the integration, storage, and intelligent processing of technological data in the mining industry. The insufficient number of telemetry devices and the lack of a closed loop adaptive learning limit the stability of predictive models. The purpose of the research is to develop a platform for data integration and forecasting. The methodology used is based on the Input–Processing–Storage–Analytics–Output architecture with integration via OPC UA, PostgreSQL/TimescaleDB for time series storage, and the Random Forest algorithm with automatic model degradation control. The scientific novelty of the research lies in the development of a closed digital circuit that combines streaming data integration, statistical stability monitoring, an automatic retraining mechanism, and a SCADA-oriented visualization interface within a digital platform architecture. Testing over 10 months (42,000 observations) showed an 82% reduction in data preparation time, an increase in data completeness from 78% to 96%, an increase in R2 from 0.71 to 0.93, and a 2-fold decrease in RMSE with an average delay of 4.7 seconds. The practical significance of the work is to increase the accuracy of ore quality control and reduce production risks. The research prospects are related to the development of the digital twin and the introduction of interpretable models.