Research On The Fouling Prediction Of Circulating Cooling Water Based On Relevance Vector Machine

Fouling is involved in many fields. It widely exists in nature, daily life and most of the industrial process. According to the investigation of Steinhangen, there are more than 90 percent heat exchangers which have the problem of fouling. Generally, the heat-transfer medium is usually cooling water. And the bad water quality of cooling water is the main reason of fouling problem. Just because of the university and huge damage of fouling, fouling is drawing more and more attention of researchers all over the world. Fouling formation is a complicated process. It includes physical, chemical process and the synthetic effect of deliver among momentum, energy and mass. In addition to a variety of difficulties in many overlapping branch of learning, all these are enhancing the difficulty of the research in heat transfer fouling. So the development of traditional prediction model based on mechanism of fouling formation is slow and far away from expected target.M. E. Tipping put forward the non-linear probabilistic sparse model based on Bayesian framework, and nominated it as Relevance Vector Machine. Relevance Vector regression method was established on the traditional kernel function and Bayesian inference framework was merged into it. Relevance Vector Machine can obtain predictive distributions outputs and have good generation ability. Relevance Vector Machine draws hyper-parameters into run, implements automatic estimation of'nuisance'parameters, and its kernel function doesn't need to satisfy `Mercer'condition. Its learning algorithm is simple, easy to implement so that the calculate complexity is dramatically reduced and the operating speed is increased. At the same time, the relevance vector is quite few.The development of prediction researching on heat exchanger fouling at home and abroad in recent years was summarized. Based on the experiment, each kind of parameters affecting the fouling resistance was analyzed. Relevance vector machine based on Bayesian theory was implemented to train the date and predict the fouling resistance. On the basis of radial kernel machine, wavelet kernel function, wavelet framework and wavelet packet analysis were also united with relevance vector machine to form WRVM and WPRVM. The simulation result showed that, WRVM and WPRVM have many advantages likes RVM, furthermore, they can avoid high frequency interference. So WRVM and WPRVM have good ability of anti-noise. The method introduced in this paper provides a new way for circulating cooling water fouling prediction.