Dynamic Bayesian Network Construction And Prediction Of Pressure Of State Of Cement Grate Cooler

Bayesian network is a combination of graph theory and probability theory.It has a clear topological structure and a convenient decision-making mechanism.It can learn the probability among parameters and the change probability of parameters based on the data.Therefore,Bayesian network has obvious advantages on dealing with the linearity and uncertainty issues.Dynamic Bayesian network is the timing expansion of Bayesian network.It not only has the advantage of Bayesian network,but also can to deal with timing problems.Cement grate cooler is the key equipment when cooling the high temperature cement clinker and recycling heat in the cement production line.In actual production,the interior of grate cooler install less monitoring point.And parameters of grate cooler exist the character of nonlinear ? time-varying and coupling.Those shortcomings make it very difficult to model for grate cooler.In order to solve these problem,the thesis proposes the research on designing the best model of grate cooler based on Dynamic Bayesian network.The specific research work are as follows:Firstly,the classical structure learning algorithms of Bayesian network have K2 algorithm and Hill climbing algorithms,which exist different shortcomings respectively.This thesis propose two improving algorithms to solve these question named MAK algorithm and CPA algorithm.In the data condition of smooth and complete,the property of improving algorithms are verified by simulation experiments.Besides,the advantage and disadvantage of improving algorithms are analyzed in detail in this thesis.Secondly,A Dynamic Bayesian network structure learning algorithm is proposed aiming at the problem of Bayesian network don't deal with the time-varying data.The structure learning algorithm is named I-CPA-DBN.It builds initial structures of the prior network and transfer network by computing the mutual information and time series mutual information between parameters.Then,these initial structures will be optimized by CPA algorithm in order to design the best Dynamic Bayesian network.With simulation experiment,we can obtain the conclusion that I-CPA-DBN algorithm balances the running time and the accuracy rate.Thirdly,the process of cement cooler's cooling system is analyzed in order todescribe the relationship between under labyrinth pressure and the other key parameters of cement cooler.Based on these transcendental knowledge,the research variables are selected.According to the 3 ? standard,The research dare is screened to conserve the smooth and complete data.And these data are quantified using parameter standard point standard.Then,combining with I-CPA-DBN algorithm,2TBN of cement cooler is built which can achieve the goal for parameter state predication.Finally,taking under labyrinth pressure as an example to explain the process of parameter state production.In this these,under labyrinth pressure is assumed as the state node and the presents of under labyrinth pressure is assumed as the observation node.2TNB model are transformed into HMM model so that predicate state of under labyrinth in t time slice.The accuracy of cement cooler parameter state prediction is computed by analyzing predicted state with actual data state.