A Method for Predicting the State of Sensors and Meters Based on Improved Generative Adversarial Networks

Sensors and meters are used in a wide range of applications, the effects of anomalies or failures are increasing. Therefore, accurate and efficient state prediction methods are required. In this paper, the main problem of state prediction— imbalanced classification is studied. The traditional GAN and CatGAN are analyzed, and an improved model CatGAN-SP is proposed for the problem of state prediction. The relevant objective functions are deduced, and the specific expressions of the objective functions are given. The experimental verification was carried out using measured data of ultrasonic flowmeter. The effectiveness of the model in sensors state prediction problem is proved by comparison with related algorithms. In particular, the model further solves the class imbalance problem and improves the accuracy of sensors state prediction effectively.