Research On Digital Twin Modeling Method And Its Application To Optimal Operation Of Thermalpower System

With the rapid development of social economy,China’s industrial structure optimization and restructuring and upgrading process,to achieve the future "carbon peak,carbon neutrality" goal,the need to build a clean,low-carbon,safe and efficient modern energy system.Wind power and solar power generation as the representative of renewable energy alternative role is increasingly prominent,while thermal power units in the future for a long time will remain in the dominant position.There is an urgent need to solve the synergistic development of thermal power and renewable energy,and large thermal power units need to take up the task of high efficiency and energy saving,low carbon and environmental protection,deep frequency regulation and peak regulation.The implementation of electric energy alternative heating is of great significance to promote the energy consumption revolution,reduce carbon emissions and promote energy cleanliness.The use of electric boiler heat storage and heating can also reduce the regulation pressure of power grid,increase the heating capacity,and effectively solve the problem of renewable energy consumption.Both thermal system of thermal power unit and thermal system of electric boiler storage heating are typical nonlinear,multi-parameter,strongly coupled complex thermal systems.In this paper,we study the digital twin modeling method that integrates fluid network mechanism modeling and data-driven modeling to provide new ideas and approaches for thermal system modeling work and provide theoretical support for safe,environmental protection and economic operation of thermal system.The paper focuses on the digital twin modeling method and its application in the optimal operation of thermal power systems,and the main research contents include the following aspects.(1)The basic theories of digital twin theory,thermal system modeling theory and big data processing are investigated.The similarities and differences between digital twin and simulation technologies and information physical systems are compared,and the fluid network mechanism modeling and solution methods are studied with thermal power plants as an example.A comparative analysis of MapReduce and Spark computing of Hadoop system,a comparative analysis of Spark Streaming and Storm for real-time data processing,and a big data distributed cluster platform applicable to the application of digital twin and big data in the field of thermal system modeling are built.The storage management of big data and big data distributed computing are implemented on this cluster,and the basic theories of data-driven modeling based on the big data platform are studied,including support vector regression modeling,extreme learning machine modeling,intelligent recognition optimization algorithms,and instantaneous learning.(2)In order to study the data-driven modeling method,an adaptive data-driven modeling method based on improved instantaneous learning strategy is proposed.The correlation between input and output variables is obtained by using the "principal component+mutual information" method to determine the weighting factors,and then a weighted comprehensive similarity measure function is defined by using the "Euclidean distance+angle".In the offline state,the work conditions are classified by the improved genetic simulated annealing fuzzy clustering method;for work condition prediction,a multi-level comprehensive similarity metric is used for fast identification of similar work conditions,i.e.,online fast identification is achieved based on a two-level search strategy:primary identification is to determine the category to which the predicted work condition belongs in the historical work condition database to extract the predicted work condition,and secondary identification is to adopt a comprehensive similarity metric function-based similarity identification.The secondary identification is to adopt the similar working conditions identification method based on the integrated similarity function for the fast identification of the predicted working conditions in the historical database;the local model modeling method is to study the data-driven modeling methods such as Spark_SVM_HPSO algorithm,Spark ELM algorithm and Spark_HPSO-based multi-parameter identification in the framework of Spark computing.Then,the effectiveness of the proposed modeling methods is verified by using SCR denitrification system outlet NOx prediction and electric boiler storage heat supply system source-side and load measurement load prediction as case studies.Theoretical support is provided for the digital twin model modeling of the thermal system and the optimization of the system working conditions.(3)For the study of digital twin modeling,a set of digital twin modeling methods is proposed for the collaborative fusion of adaptive data-driven and mechanism model multi-parameter identification with improved instantaneous learning strategy.Based on the establishment of the thermal system mechanism model,the key equipment model parameters are identified offline using the offline intelligent identification method of multi-parameter and multi-case fitting to obtain an offline intelligent parameter identification model that can simulate the dynamic change trend of the actual system under full working conditions;the offline intelligent parameter twin model is mainly used to make judgments based on the similarity threshold,and the adaptive model parameter updating strategy is adopted to realize the online collaboration of the digital twin model To further improve the accuracy and robustness of twin model prediction,the online fusion method of multi-model output with moving pane information entropy is used to improve the approximation degree of critical working conditions as well as dynamic change processes.The twin model constructed based on this theory can continuously self-correct based on the system operation data and track the equipment operation characteristics online,thus having self-adaptive and self-evolving intelligent features,which can fully reflect the system operation status and performance and provide reliable model input and result verification tools for the iterative optimization of system conditions.The SCR denitrification system of coal-fired power plant and the heat storage and heating system of electric boiler are taken as the research objects to establish their thermal system digital twin system models.(4)Finally,based on the real-time tracking capability of the digital twin model,an intelligent working condition dynamic optimization strategy of the thermal system based on load distribution and working condition seeking is proposed.And taking the electric boiler storage heat supply system as the research object,according to the energy cost analysis and load allocation strategy,the digital twin model system is used to predict and calculate the grid load,electric boiler system and heat storage system,simulate the dynamic operation of the system under different operation schemes and working conditions,and derive the optimal heat supply regulation and load allocation scheme.Taking the SCR denitrification system of a coal-fired power plant as an example,the model is applied to the model predictive control algorithm based on the established adaptive and self-evolving digital twin model of an intelligent SCR denitrification system.The results show that the adaptive predictive control algorithm based on the digital twin model is more accurate and stable in operation than the traditional PID control,which proves the effectiveness of the proposed modeling method and has important engineering practical significance and industry demonstration value.