Research On Optimization Of Unit Operation Parameters Based On Data Mining

As the world moving towards a “low carbon society” overall,people pay more attention to the work of energy saving and emission reduction. As the basic energy of social economy development, the proportion of electric power in energy consumption is growing. The work of energy- saving and consumption reducing in the electricity sector have a great influence to the target of the “13th Five-Year” plan and low-carbon economy. Therefore, many electric enterprises promote the work of energy conservation and consumption reduction to enhance their competitiveness. There has many factors which effect economic in power plants. So rational optimization of the operation parameters becomes a more effective means of saving energy and reducing consumption in power plant. It also has very important practical significance.This paper starts with the actual operation parameters of the unit. At the beginning,remove the non steady state data and the abnormal data according to the volatility of the data to pretreat the historical data in order to ensure the authenticity and validity of the data.Then carry out the work of correlation analysis. Calculate the sensitivity of heat consumption rate to the parameters of the boundary based on the least square method and select the parameters which have important influence to the heat consumption rate to carry out the following work.The next work is dividing the operating conditions,choose out the representative data which can Reflect the equipment characteristics and operating characteristics of the unit.Divide the data based on the method of K-means clustering algorithm to classify the data under similar conditions. The set of data can be replaced by the nearest data to the average which have many information.At the same time remove the data under extreme conditions to avoid data redundancy and waste of calculation.Then build a model between boundary parameters and heat loss.Two methods are used in this paper to build the model,the first one is BP neural network, the second is using Adaboost algorithm which is based on the idea of combine the output of a plurality of "weak" classifiers to produce effective classification. The second method is used to modeling by analyzing the error of the two modeling methods.At last,obtain the optimal initial pressure according to the whole unit model. The rationality of the change of the optimal initial pressure under the changing condition of boundary to proves the validity of the model.