| Ironmaking is a strongly coupled,multi-variable,non-linear dynamic process in a closed container.It is not possible to directly observe internal operating conditions.Blast furnace operating status can only be indirectly judged by the monitoring data of many sensors installed in blast furnace body.At present,most blast furnace operation status prediction models have problems such as single prediction target parameters and short data time spans.Based on big data technology,the construction of two indicators prediction models for silicon content in hot metal and gas utilization rate is to predict operation status of blast furnace.It is of great significance for guiding actual production.Collected production data of CHENG STEEL No.4 blast furnace took up 66.12 GB of storage space.During the collection process,it was found that the blast furnace production data has characteristics of high dimensions,large volume and high value,and there are some problems such as unclear data attribution and high noise,and some solutions were proposed.The linear interpolation method and multiple regression method were used to process the missing data,and the box plot method and the fit-and-fill method were used to process abnormal data.A data warehouse of blast furnace with a time span of 4 years was established.The data storage is 30 GB,which contained 1.28 billion minute-level frequency data and 7 sets of data with a total of 811 fields.By using the combination of Pearson’s correlation coefficient,Spearman’s rank correlation coefficient,and maximum information coefficient,the correlation between blast furnace operating parameters and index parameters was analyzed,and hot air temperature,oxygen-rich content,cold air flow,and coke ratio were finally screened.A total of 13 parameters,such as furnace pressure and furnace top pressure,were used as input parameters.3072 pieces of data in the data warehouse were used as training libraries,and 408 pieces of data were used as verification libraries.Different algorithms were used to construct the prediction model of the index parameters.The results show that the fitting goodness of the KNN algorithm and the Ada Boost algorithm model reached 0.92 and 0.89 respectively,and the average error was less than 5%,all of which could predict the change trend of the index parameters,and the model performance was good.Operators confirmed that prediction results of the two models are true and valid.By analyzing the historical conditions of the index parameters,and cooperating with the production operators to select the optimal data interval,the operating state evaluation mechanism of CHENG STEEL No.4 blast furnace was determined.The main program was designed with Python language,and a comprehensive user-friendly blast furnace operation status comprehensive evaluation system was constructed,it can display the parameters that operators care about,display the historical data in real time,evaluate the blast furnace operation status one hour in advance,and guide operators to control the blast furnace.Figure 15;Table 15;Reference 71... |
PDF Full Text Request