Research On Multi-index Prediction Model And Decision Algorithm Of Cement Calcining System

In the cement production process,the cement calcination system bridges the raw material preparation system and the clinker grinding system.The optimization control process of the cement calcination system involves determining the set values of each process control system based on the target values of operational indicators and ensuring that the controlled variables track the set values.This allows the cement calcination system to operate under optimal production conditions while keeping the operational indicators within the desired range.However,the cement calcination system is a complex dynamic system,and the relationship between operational indicators and production indicators cannot be accurately described through the establishment of mechanistic models.Moreover,the comprehensive production indicators,operational indicators,and their constraint ranges,which are relevant to the decision-making of operational indicators in the cement calcination process,exhibit frequent fluctuations over time.Therefore,the decision-making of operational indicators in the cement calcination system poses a challenging dynamic optimization problem that cannot be effectively addressed using traditional optimization methods.In summary,it is of significant theoretical research significance and engineering application value to study how to accurately model the cement calcination system under dynamic operating conditions to achieve dynamic decision-making of operational indicators,thereby ensuring the safety and economic efficiency of the cement calcination process.This paper focuses on the research of dynamic decision-making of operational indicators in the cement calcination system,starting with the development of quality and energy consumption prediction models for the cement calcination process.The main research work is as follows:(1)Aiming at the problem of the effectiveness of data feature extraction caused by different sampling frequencies of variables in the cement calcination process,this study proposes an innovative prediction method for free lime content based on the combination of dual-frequency principal component analysis and extreme gradient boosting tree algorithm.The method divides the variable data into high-frequency data and low-frequency data based on the sampling frequency of process variables and establishes a prediction model for free lime content using the extreme gradient boosting tree algorithm.The accuracy of this model is experimentally tested using actual production data from a cement manufacturing company.The experimental results show that this prediction method effectively eliminates the problem of different data redundancies caused by varying frequency changes and achieves accurate prediction of free lime content under multiple operating conditions.(2)Since the cement calcination process is a complex physical and chemical reaction process,there is a complex time-varying time-lag and nonlinear relationship between its production variables and energy consumption indicators,resulting in insufficient prediction accuracy of the model to predict the two indexes simultaneously.To solve this problem,proposed a combined approach utilizing a sliding window and a dual-channel convolutional neural network to predict coal consumption and electricity consumption production indicators.This approach employs a sliding window to construct time series with time-varying and time-delayed features,which are then fed into a dual-channel convolutional neural network model to extract the temporal characteristics of coal consumption and electricity consumption indicators comprehensively,thereby improving the prediction accuracy of the model.Experimental validation is conducted using the latest actual data from cement production.The proposed model is compared with other methods,such as recurrent neural networks.The experimental results demonstrate that the proposed energy consumption prediction method outperforms the compared models,proving the superiority of this model.(3)To address the challenges of real-time dynamic changes in operating conditions and the inability to meet actual dynamic operating conditions caused by open-loop decision-making in multi-indicator optimization,an innovative dynamic operating condition optimization decision-making method based on a dual-loop coupling prediction-correction strategy is proposed.First,establish an energy consumption optimization model under quality constraints based on the above cement calcination system quality and energy consumption prediction model.Then,a double-loop coupling prediction and correction strategy are used to optimize and correct the solution obtained by the optimization model in real-time,to obtain the operation index decision solution of the actual working condition.Experimental optimization of energy consumption in a cement calcination system is conducted using cement plant data.The method is compared and analyzed against manual decision-making for free lime content and production costs.The results demonstrate that compared to manual decision-making,this method can effectively stabilize the free lime content and reduce production costs while ensuring the stable operation of the system.