Fault Diagnosis Method of Industrial Boiler Based on Physical Knowledge Adversarial Generation Network

Abstract

【Problem】To provide a method for diagnosing faults in an industrial boiler based on a physics knowledge adversarial generation network even when boiler fault data is of poor quality or insufficient. 【Solution】First, step 1 of obtaining historical operation data of the industrial boiler and performing preprocessing of the data; step 2 of constructing a physics knowledge neural network model, constructing a mass and energy conservation formula related to the boiler, and simulating the operation of the boiler by the physics knowledge neural network; step 3 of constructing an improved adversarial generation network model, where the adversarial generation network generates good data using the competition between a generator and a discriminator, and its parameters are alternately trained by the generator and the discriminator; step 4 of training a diagnostic model using a training sample set, performing a fault diagnosis test using a test set, and verifying the effectiveness and superiority of the method by algorithm comparison. 【Selected Drawing】Figure 1

Images