Health Assessment for a Sensor Network With Data Loss Based on Belief Rule Base

As the complexity of a system increases, the use of sensor networks becomes more frequent and the network health management becomes more and more important. When sensor networks are applied to complex environments, they are influenced by the disturbance factors in engineering practice and observation data may be lost. This will decrease the accuracy of the health state assessment. Moreover, due to the disturbance factors and complexity of the system, observation data and system information cannot be adequately gathered. To deal with the above problems, a new health assessment model is developed based on belief rule base (BRB). The BRB model is one of the expert systems in which the quantitative data and qualitative knowledge can be aggregated simultaneously. In the new health assessment model for a sensor network, a new missing data compensation model based on BRB is constructed first, in which the historical data of the monitoring indicators are used. In addition, the expert knowledge for the historical working state of the sensor network is also applied in the constructed missing data compensation model. Then, based on the compensated data and the observation data of the sensor network, the health state can be estimated by the developed health assessment model based on BRB. Given the uncertainty of expert knowledge, the initial health assessment model cannot assess the health state of the sensor network in an actual working environment. Thus, in this paper, an optimization model is constructed based on the projection covariance matrix adaption evolution strategy (P-CMA-ES). To illustrate the effectiveness of the new proposed model, a practical case study of a sensor network in a laboratory environment is conducted.