Performance sensitivity analysis of linear alarm filters

Abstract

This paper conducts the performance sensitivity analysis for linear alarm filters. With a derivative-based local sensitivity measure, models are proposed to assess the sensitivity of the system detection errors to changes in the trip point and uncertainties in the collected data. Then, sensitivity formulations for a general linear alarm filter framework with unknown data distributions are studied. Extensive simulations with Gaussian and Gamma data using two typical linear filters, the moving average (MA) filter and the linear weighted (LW) filter, are conducted to verify the analytical results and to show the trade-offs among different system parameters.