Recurrent neural network for faulty data identification in smart grid

The accuracy of the control data from different sensors in a system is evaluated by embedding a recurrent neural network with layer feedback for each sensor. The accuracy of the sensor output is calculated by comparing the values from neighboring sensor output. Here non-linear sensor model using Hammerstein-Wiener was used and the amount of sensor data fault is estimated by using kalman filter. This value will be considered as an actual output in case of sensor failure. The performance is analyzed with and without extended kalman filter learning algorithm by introducing a step size fault.