Research On Modeling Method Of Combustion Process Of Coal-fired Heating Boiler

The boiler combustion process is a complex,time-varying nonlinear system.Establishing an accurate model to reflect the combustion situation is an effective means to guide actual production and improve combustion efficiency.The re-learning ability of the model is a reliable guarantee to reflect the dynamic changes of the system conditions in real time.It is often difficult to accurately describe the characteristics of the combustion system using the mechanism method,and the historical model cannot reflect the time-varying characteristics of the system in time.In response to the above questions,the main research contents are as follows:Firstly,a nonlinear function is designed to simulate the load change during the combustion process of the boiler.The nonlinear function model established by the least squares support vector machine(LSSVM)has low prediction accuracy.Therefore,the improved particle swarm optimization algorithm(PSO-LSSVM)is used to improve the model prediction accuracy.Due to the redundancy between the samples in the training set,the model training time is too long.The pruning algorithm is used to thin the model and reduce the training time of the model.Simulation experiments on sliding time window type,forgetting weighting type and model contribution type iterative method show that the model contribution method has better practicability.Secondly,the boiler combustion model is established by taking the actual data as a sample.Compared with the LSSVM model,the PSO-LSSVM model has a 43.5%improvement in model prediction accuracy.After sparsification of the model,the training time of the BP A model is reduced by 14.3%;the training time of the FPA model is reduced by 17.9%.Analyze the difference and consistency between simulation and reality,and find that it has certain applicability to the PSO-LSSVM modeling method of stror:tg nonlinear system.The model sparse method can improve the training efficiency of the model and has certain practicability.Finally,in view of the time-varying character:istics of the combustion system,three model iterative methods are experimentally studied.The results show that the sliding time window is to replace the existing information in the "first in,first out" manner and train the model in time;although it can reflect the changes of the actual working conditions in real time,it ignores the influence of historical characteristics on the model.The forgetting weighting type is based on the iterative method of forgetting thought to fill the new information in turn,and according to the duration of the current time,the forgotten weights that are positively related to the relevant information are eliminated,and the historical characteristics of the system are included to some extent,but the model is iteratively The short-term prediction accuracy is reduced and the training model needs to be repeated periodically.The model contribution degree type dynamically collects the information exceeding the contribution threshold in real time,and eliminates the information with low contribution degree.This method can retain the effective information of the model to the maximum extent.To improve the effectiveness and generalization of the model,and to reduce the number of training sessions.The PSO-LSSVM method is used to establish a boiler combustion model with high efficiency and high generalization ability and iterative correction can timely and accurately describe the characteristic changes in the combustion process of the boiler can better describe the characteristic changes in the combustion process of the boiler,can serve the optimal control of the combustion system,and can operate and manage the heating production.Provide positive guidance and assistance to the operation and management of heating production.