Data Driven Based Thermal Power Unit Modeling And Automated Deployment Support System

Thermal power generation is an important power generation mode in our country and plays an important role in China's energy structure.With the progress of the times,big data technology is being paid more and more attention.Big data technology gets more and more applications in the treatment of thermal power plant parameters prediction,fault diagnosis,safety monitoring and performance optimization.According to the requirements of thermal power units for the prediction of the key parameters of thermal power units,this paper starts from the perspective of big data modeling by several aspects such as data parameter extraction,feature extraction,model construction and automatic deployment.The main work of the dissertation is as follows:First,the data parameters are extracted.In this paper,through the modeling of thermal power big data,the paper analyzes the mechanism model of thermal power unit,elaborates the relationship between the various processes of thermal power unit,analyzes the relationship between various parameters of thermal power unit,selects the model through the mechanism model factors to construct the model parameters 68 columns.Through the correlation analysis method,24 parameters selected from the mechanism model were screened,including 1 order parameter(load order),4 model parameters to be constructed and 19 model input parameters.Second,feature generation.According to the load instruction,this paper presents an algorithm to divide the thermal power units into steady-state intervals and get a working interval for the thermal power units under a longtime actual load stabilization.The model is built under each divided steadystate interval.For each steady interval,in order to reduce the noise between the measured parameters,the time delay between parameters is calculated.The parameters of the sliding window method are spliced by calculating the delay results to generate the model training and testing Eigenvectors.Thirdly,the model is built.For each load instruction interval,the data standardization is first processed.Through the constructed fully connected deep learning model,the main steam pressure,reheat steam pressure,cold reimport steam pressure of the network,The parameters are combined and modeled to obtain each load segment.The MAE of the prediction results of each parameter is less than 0.2 atm.According to the real-time prediction results of three parameters,the data fusion with the input parameters to build the actual load forecasting model of the eigenvector,and then re-test the construction of fully connected deep learning model,the actual load real-time prediction of 1008 MW The prediction MAE of load section is 0.148MW(error of 0.015%),and the prediction MAE of other load sections are less than 0.07 MW,which can meet the requirements of parameter prediction.Finally,automated deployment.In order to reduce human intervention,this paper introduces the sub-process method to achieve the regular operation of each sub-module model,the introduction of model batch training method to achieve automatic tuning of the model,the introduction of a model service method to achieve the model of regular updates to complete a complete set of thermal power plant model deployment system.The system can be fully automated in the module of parameter extraction,model building,model tuning and model deployment.