Double level method of data verification based on neural network predictor and pipeline characteristics

To make elbow-pipe measuring system have tolerant ability for sensor fault, sensor data verification based on neural network predictor and pipeline characteristics is studied, taking a given heat and power plant for example. Neural network predicting model is built with sensor output sequence. The data validity is determined on the basis of whether the error between the predicting value and the sensor output is larger than the threshold. For the invalid data, the further verification is conducted with the pipeline character to avoid false alarm due to the inadequate learning knowledge of neural network. If it is finally proved to be invalid, the data will be replaced with the predicting value to keep the system in operation. Simulation results show that the method can verify the sensor data validation, and has good diagnostic and recovery capability for the sensor failure.