Research On The Application Of Reverse Model Method For Complex Thermal Systems

Energy saving of thermal power units can be achieved by equipment and technology's improvements and optimization of operation process.For the former,it means the request of condition monitoring,fault diagnosis,performance evaluation and intelligent control.Thermal power is a typical complex thermal system, quantitative analysis and study for its process is classified as modeling of thermal systems in essence.Taking into account the limitations and shortcomings of mechanism modeling,reverse modeling method provide new ideas and approaches for modeling of thermal systems.Based on the significance of feature variables extraction for modeling,proposed GL-MIV algorithm.GL-MIV algorithm divided the original variables into effective variables,invalid variables and redundant variables three categories.Removing invalid variable,lifting redundant variables and original variables'information doesn't required to reconstruction is the original intention of structuring the algorithm.Eature variables extraction is a important factor for model's performance and scale,Validity, interpretability and streamlining is the basic requirements for a method of feature variables extraction.Unlike principal component analysis and correlation analysis,GL-MIV algorithm could not only get a high compression ratio of information,but also can retain all the physical meaning of the original variables,rather than a linear combination of the original variables,which is important for the analysis of correlation between variables.In-depth study of generalized regression neural network structure,proposing the concept of performance factor δ for generalized regression neural network,which is used to describe our attentions for training accuracy and generalization ability.Where we no longer take training accuracy as network training's target,and add coefficient k which contain network threshold into the objective function,to limit the growth of this network's threshold,by which we can get a smoother network output.It's take attention to network generalization ability essentially,which providing theoretical basis for selection of key parameter of generalized regression neural networkFurther on,based on the the research results above and for the study to large thermal power units,introducing reverse modeling into complex thermal system modeling.Summary the application of reverse modeling in complex thermal system,which can be classified as two aspects,one is regression problems and the other is classification problem.Take the measurement of turbine main steam flow as a example of regression model based on reverse modeling,and take comprehensive condition assessment and fault diagnosis of turbine as examples of classification model based on reverse modeling.The former use generalized regression neural network algorithm and efectively improved model generalization ability through optimizing the network distribution density,and the latter introducing projection pursuit principle and Wavelet Analysis-Artificial Neural Networks model which used to condition assessment and fault diagnosis work for thermal system.Finally, providing comparative analysis of the model test results,which validated the effective application of reverse modeling in complex thermal system.