| With the continuous growth of cement market demand and the national strategic policy of constructing ecological civilization system,higher requirements are put forward for energy saving and consumption reduction in cement production,especially for the optimal control of grate cooler as clinker cooling and waste heat recovery.However,there is still a lack of energy consumption prediction model which can be used in combination with grate cooler situation analysis.In this study,the historical data of cement production line is used as the parameter characterization of grate cooler,and the modern data mining method is used to identify the typical working conditions of grate cooler and estimate the energy consumption.The purpose is to provide a more accurate energy consumption prediction model for the situation energy-saving control of grate cooler.Firstly,the paper summarizes the grate cooler and clinker firing system,puts forward the assumption conditions of energy modeling in grate cooler,studies the energy balance relationship in the grate cooler,explores the relationship between the situation analysis and energy consumption prediction of grate cooler,analyzes the difficulty of modeling grate cooler,and constructs the working condition and energy consumption prediction application framework based on data mining algorithm to solve the above problems.The whole framework includes the working condition planning framework based on unsupervised learning and the sub framework of energy consumption estimation based on supervised learning.Secondly,in the framework of unsupervised learning,six kinds of operation data of cement production line are selected as the parameter characterization of grate cooler condition identification.The abnormal data are removed from the historical data to form a sub database for condition classification.An improved clustering algorithm of principal component analysis is applied to divide the condition data.The experimental results show that the data dimension reduction reduces the amount of calculation while retaining the difference degree of grate cooler data information.At the same time,the improved clustering algorithm solves the sensitive problem of the initial value of the traditional algorithm and improves the clustering stability.The improved algorithm introduces two clustering evaluation criteria to select the number of working conditions,and obtains the parameter variation range of three typical working conditions of grate cooler through data space inverse transformation.Then,aiming at the difficulty of energy consumption prediction of grate cooler,according to the sub framework of energy consumption prediction based on supervised learning,a deep learning model is proposed to predict the energy consumption of grate cooler.The principle of deep learning algorithm is analyzed in detail.Genetic algorithm is used to optimize the parameters of BP neural network.The dynamic weighting method is used to combine the optimized neural network model with support vector machine regression simulation and random forest model to form a multi-mode combination energy consumption prediction algorithm,which provides a theoretical basis for the application of energy consumption prediction in the future.Finally,according to the overall framework of energy consumption prediction based on working conditions,the parameters of three typical working conditions in the working conditions framework are combined with the processed energy consumption data of cement cooling to form a sub database for energy consumption prediction,and two parameters are selected as the evaluation criteria of the prediction model,The energy consumption model is compiled in the mixed language environment of MATLAB and visual studio 2019 software development platform.The first mock exam shows that the combined model can find different weights combinations under different conditions,which makes the combination model more accurate than data model and has a higher degree of accuracy than the single model.It provides a workbench based energy consumption prediction process for the actual demand of cement production line. |
