A Dynamic Bayesian Network-Based Prediction Method for Cognitive Ability of Long-Duration Operators

The accurate prediction of cognitive ability variation over time is critical for operator performance and safety management. To address the time-dependent degradation of the cognitive ability of long-duration operators working in confined, shift-based and otherwise demanding environments, this paper proposes a prediction method based on a Dynamic Bayesian Network (DBN). Four cause-layer factors (confinement duration, shift cycle, circadian rhythm and sleep, and working environment) are first analyzed, and five result-layer indicators (perceptual stability, attention-and-vigilance stability, working-memory stability, reasoning-and-decision stability, and executive-function stability) are selected to characterize the cognitive ability of long-duration operators. After the indicators are quantified and normalized, a Dynamic Bayesian Network (DBN) is constructed under a first-order Markov assumption to describe the temporal evolution of cognitive ability. The inference process consists of two stages: state estimation using current observations, followed by prediction through the time-dependent transition. A cognitive ability level, denoted as Pr ∈ [0, 1], is treated as the target variable. A conditional-probability correction factor is further introduced to capture inter-operator heterogeneity, thereby establishing a prediction model for cognitive ability level. A 288-h simulated-task experiment was designed, and cognitive-test data were collected from 15 participants, who were divided into three groups (A1~A3). The model prediction accuracy was evaluated at both the individual and group levels. The results suggest that the prediction errors remain generally low. Furthermore, the predicted cognitive ability exhibits a continuous decline with task time, with partial recovery occurring during rest periods. The results show that the proposed DBN generally outperforms the Static Bayesian Network (SBN) baseline at both the individual and group levels. In the group-level evaluation, the Root Mean Square Error (RMSE) values for all three groups are below 0.011, the Mean Percentage Error (MPE) values are within −0.84%, and the Maximum Cross-Correlation Coefficient (MCCC) values exceed 0.759, indicating good numerical accuracy and stability. The DBN also outperforms the mean predictor, confirming that it captures meaningful temporal variation rather than merely fitting the overall mean level. At the individual level, the DBN generally achieves lower RMSE and MPE and higher MCCC than the SBN across the 15 participants. These results demonstrate the effectiveness of the proposed DBN framework for the dynamic monitoring and early warning of cognitive ability in long-duration operations.