A data-driven based adaptive fault diagnosis scheme for nonlinear stochastic distribution systems via 2-step neural networks and descriptor model

A data-driven based adaptive sensor fault diagnosis (FD) and compensation scheme for stochastic distribution control (SDC) systems is studied in this paper, where an augmented descriptor model is employed. Unlike traditional SDC systems, the driven information is the output probability density function (OPDF), which is a kind of image mapping information to the true output values. A mixed 2-step adaptive neural network (NN) framework is studied, where the static NN is to describe the OPDF while the dynamic NN is to identify nonlinearity, uncertainty of system and to refine the OPDF model based on data of the input and statistic information of the output. To identify the sensor fault, an augmented descriptor system is employed, where the augmented state includes the plant state and the sensor fault. As a result, an adaptive strategy is given for nonlinear parameter estimation and sensor fault identification simultaneously. A sensor compensation rule is given to restore the plant by adding it to output feedback controller. The simulation examples are given to verify the effectiveness of the presented algorithm.