An evolutionary framework with adaptive recovery for online reconfiguration of multi-source measurement systems

Abstract

Multi-source measurement systems are expected to maintain estimation accuracy and operational robustness under abnormal disturbances. Such disturbances can alter source availability, observation quality, and synchronization conditions, causing a previously deployed configuration to become infeasible. Online reconfiguration therefore involves adapting source activation and fusion weights as operating conditions evolve. We formulate this task as a dynamic constrained multi-objective optimization problem in which disturbances induce migration of the feasible region. An effective algorithm should therefore restore feasible operating states promptly, limit cumulative feasibility loss and accumulated violation, and preserve terminal approximation quality. To meet these requirements, we propose an event identification and risk-adaptive scheduling multi-objective evolutionary algorithm (EIRAS-MOEA). Population-level feasibility and constraint violation determine recovery initiation, while a risk score regulates repair intensity through the proportion of infeasible offspring selected for repair. Across the Dynamic Generalized Time-Dependent (DGTD) disturbance stages, public dynamic constrained benchmarks, and an application case based on measured traces, EIRAS-MOEA restores feasibility faster than representative comparison algorithms. The method also accumulates less feasibility loss and constraint violation while maintaining competitive terminal approximation quality. The results further indicate that recovery latency and cumulative feasibility loss describe aspects of recovery that terminal indicators do not capture. These recovery metrics therefore provide timely information to support adaptive online reconfiguration under evolving disturbances.