Research On Operation Optimization Of Coal-fired Power Plant Based On Spark And Association Rules Mining

The optimization of thermal power unit operation is an important means to increase the energy-saving and consumption-reduction potential of the unit,and it is of great significance to China's goal of achieving sustainable development and low-carbon economy.The reasonable determination of the operation optimization target value is the core and key of the thermal power unit operation optimization.The universal application of automation and information technology in thermal power plants enables the actual operation data of the unit to be saved.These high-dimensional,highly-coupled,and strongly linearly correlated data record the actual operating state of the unit,and have extremely high analysis and mining value,so that the method for determining the target value of operation optimization based on association rules shows great advantages,and has been obtained by researchers.Attaches great importance to it,thus developing a hot research topic.At present,the rise and rapid development of big data technology has injected new vitality into this research field.Promoting the application of big data thinking and technology in the operation optimization of thermal power units is of great significance for improving the economic operation of the unit.Based on the massive historical operation data of thermal power units,this paper introduces the big data mining technology combining association rules and Spark distributed computing framework into the research on the method of determining the operation optimization target value.The main contents are as follows:On the basis of summarizing the relevant theories and methods of association rule mining technology,this paper proposes a frequent item set for high-dimensional sparse data for the problem that traditional algorithms cannot efficiently process the operating data of thermal power units with sparse and high-dimensional characteristics Mining algorithm,FM-growth.The algorithm uses a frequent item matrix data structure with high compression capability,which can completely extract the useful information required for mining from the data set.The mining process is based on the mining ideas of the traditional model growth method,and divides and conquers the strategy to decompose the mining task into multiple independent subtasks.Each subtask realizes the mining of frequent itemsets through simple matrix operation and iterative calculation.In order to improve the ability to mine massive data and break the limitation of single-computer computing resources,this paper has conducted in-depth research on the parallelization of frequent itemset mining algorithms.First,the working principles of the two most popular distributed computing frameworks,Map Reduce and Spark,are analyzed,and it is determined that the follow-up work will be based on Spark,which is more suitable for iterative algorithms.Then it learns the existing parallel frequent itemset mining algorithms,sorts out and summarizes the specific parallelization implementation schemes.Finally,a parallel FM-growth algorithm based on Spark is proposed.Based on the above two basic research results,this paper finally put it into the application of determining the unit operation optimization target value.By sorting out existing target value determination methods and improving and perfecting,a complete set of unit target value mining process is summarized.This article makes a detailed division of the mining process and explains each link separately.Finally,taking the 600 MW coal-fired generating unit of a power plant in Shaanxi as an example,525,600 pieces of historical operating data in a complete operating cycle of the unit are mined.By comparing and analyzing the target value obtained by mining with the existing design value,the effectiveness and advancedness of the proposed method are verified.