Research On Complex Data Preprocessing Technology For Big Data Platform Of Thermal System

With the development of science and technology,measurement technology and advanced equipment,online monitoring points of thermal system are increasing,and history data can be provided as many as oceans.By building a big data platform supporting the thermal system,a large amount of raw data can be pooled together.Through intelligent analysis,data mining and other technical analysis of the potential law in the data,the operating status of power plant equipment and system can be forecast,which can help staff make the right decisions in a timely manner.Thus the operational efficiency can be enhanced,and greater returns can be achieved.The massive data of thermodynamic system brings not only huge information and benefits,but also a lot of challenges,such as data storage,integration,usability analysis and etc.How to identify poor quality information from the explosive growth of massive data,to retain valuable information,has become a huge challenge taken by a big data.Data preprocessing,including cleaning,integration,conversion and reduction,is an important way to solve these problems.Data reduction is one of the key steps in the preprocessing of big data.It can reduce the data processing quantity and improve the efficiency of data processing by reducing the attributes of big data.The amount of data provided by the thermal system is large and the dimension is high.Furthermore,different from test data,operation data has many features that are not favorable for modeling,for example,there are multiple correlations among variables,uneven distribution of working conditions,and the process is nonlinear.All these problems seriously hinder the development and application of the modeling method of historical data.In order to reduce the information overlap between the data of the thermal system,as far as possible to provide variables with strong independence,the dimensionality reduction of the data attributes has been made.Reference patents at home and abroad,in view of the problem of multiple colinearity between the variables,a comprehensive dimension reduction screening has been carried out based on calculation of the conditional number of the variables and the variance inflation factor(VIF)values of the auxiliary regressi on equation.When modeling application,rough set method will be used for the intelligent selection of model input variables,and then the attribute reduction will be further realized.In order to verify the pretreatment effect,the historical operation da ta of a 600 MW unit in a power plant were preprocessed,and the BP neural network was used to establish the online monitoring model of boiler efficiency and NOx concentration.Modeling the data before and after treatment respectively,it is found that the complexity of the model has been greatly reduced and the accuracy has been improved,indicatesing that the pretreatment is feasible.On the basis of the above online monitoring model,the genetic algorithm optimization model was constructed to realize the high efficiency and low pollution combustion of the boiler.In the model,to minimize the NOx concentration and maximize the boiler efficiency is the goal.Through continuous optimization of operating parameters,such as oxygen content in furnace and the op ening degree of air damper,finally obtain the operating parameters of the optimal fitness function,and combustion optimization is realized,which has a strong guiding function to the optimal operation of power plant.