Water Leakage Detection Based on Variation Bayesian Neural Network Autoencoder

Abstract

Water pipe leakage detection is of great significance to the protection of water resources in our country. However, the detection accuracy of water pipe vibration signals is easily affected by external noise. Existing detection technologies are difficult to reduce the influence of noise signals, and some machine learning based outlier detection models are not very robust. Variation Bayesian neural network Autoencoder (VBAE) replaces the fully connected layer network in encoder and decoder with Bayesian neural network (BNN) on the basis of Variation Autoencoder (VAE). VAE has a strong generalization ability, while BNN has an uncertainty quantification ability. VBAE combines the advantages of the two models and is very robust. Experiments have proved that, compared with other models, VBAE for water leakage detection can greatly reduce the influence of noise on the output of the model, and has a better detection effect.