Risk-oriented Approach in Implementation of Predictive Analytics into Control Systems of Production and Technological Complexes

Abstract

The integration of predictive analytical models into the control systems of production and technological complexes (PTC) involves multicomponent risks arising at the intersection of algorithmic, informational, and organizational factors. Traditional risk analysis methods, such as failure mode and effects analysis (FMEA) or expert surveys, do not fully account for the uncertainty of expert judgments and the system effects of implementing intelligent algorithms. This paper proposes a comprehensive methodology for the quantitative assessment of integration risks, based on the synthesis of three mathematical frameworks: fuzzy sets (triangular fuzzy numbers), the analytic hierarchy process (AHP) for determining the weights of risk groups, and the Monte Carlo method for statistical substantiation of the results. A formalized structure of nine key risks is presented, distributed across three groups: data, algorithms, and organization. Based on expert assessments for a typical PTC, composite indices for each risk are calculated, including median values, 95% confidence intervals, and variation coefficient, as well as an integral indicator of system risk classified as a medium threat level using quartile analysis. A polar diagram (risk wind rose) is developed to enable a clear comparison of risk profiles for three implementation scenarios: centralized, decentralized, and hybrid. It is shown that the hybrid architecture provides the most balanced risk profile. The high correlation of the results with the classical FMEA method (Spearman's rank correlation coefficient ρ = 0.97) confirms the adequacy of the developed methodology. The proposed approach is consistent with existing standards and can be used at the design stage of integrating predictive analytics systems into industrial PTCs.