Abstract: Aiming at the problems of low diagnostic accuracy, poor robustness and easy to fall into over fitting in the fault diagnosis of gas pipeline valves, combining with the deep learning theory, a MLP neural network model based on Google artificial intelligence learning system Keras is proposed to predict the fault degree of valves. Eight characteristic parameters of valve fault are selected as the original input of the model. After feature extraction, parameter reconstruction, Adam optimization, Softmax classification of multi-layer perceptron, and adding dropout module to avoid over fitting, the multi-layer perceptron model with high prediction accuracy is finally obtained. The multi-layer perceptron model is applied to the gas pipeline valve fault diagnosis system in the laboratory. 

Key words: MLP, valve fault, artificial intelligence, gas pipeline