Study On Compresser Units State Prediction Method Based On The Big Data Model Deep Learning Meachines

The different parts of compresser units are mutually related and tightly coupled,that makes the units' fault features uncertainty,nonlinear and concurrency.The learning ability of traditional models has a bad performance for units' features,while the deep learning is essentially an extraction technologyby multiple non-linear transform.Model of complex relationship among input data can be constructed by its hierarchical features.So on the basis of the deep learningtheory,the study on the prediction method for monitoring data of compresser units is theoretical significance and haspractical value.Then conduct the study on how to use the deep Boltzmann machine model in the prediction for the compressor units' monitoring data.(1)Study on the G-DBM prediction model for compresser units.Increase Gaussian filters to the visual layer in DBM,thus the improved model can process the real-valued data.Make the data preparation rules before the data processed and build single-step prediction,multi-step prediction models based on G-DBM.(2)Study on the optimization methods of G-DBM prediction model.The Ex-tremum Disturbed and Simple Particle Swarm Optimization algorithmis are presented to determine the model parameters.To accelerate the training of G-DBM model,a hybrid MLS Conjugate gradient methods is adopt.By comparison,the optimized model achieves better results.(3)Study on the bearing faultdiagnosis model formed by combining undecimated lifting hybrid Wavelet packet denoising method and Gaussian-DBM model.Denoise the data signal anddetect the bearing fault using the feature extraction G-DBM model,the combined model achieves better results.