|
|
Comparison on multi-step prediction of blast furnace gas generation based on LSTM/SARIMA time series model |
BAO Xiang-jun1, WENG Si-hao1, CHEN Guang1, WANG Jing2, CHEN Xu2, XIE Jing-cheng1 |
1. School of Energy and Environment, Anhui University of Technology, Ma'anshan 243000, Anhui, China; 2. Shanghai Baosight Software Co., Ltd., Shanghai 201203, China |
|
|
Abstract In order to accurately predict the gas generation amount of blast furnace under normal and variable working conditions (such as wind off, production reduction, production shutdown, etc.) The amount of blast furnace gas generated. The prediction effects of the two models with different prediction steps under normal conditions are compared and that with the increase of the number of prediction steps, the prediction accuracy of the two models generally decreases, and the prediction accuracy of the LSTM model is generally higher than that of the SARIMA model are found. Model accuracy, the prediction effects of different input samples on the model under the condition of 30-step prediction are also compared. The results show that the optimal input sample size of the SARIMA model is about 200, the corresponding average relative error is 0.057 0, and the LSTM model is the best The input sample size is about 100, and the corresponding average relative error is 0.042 8. Therefore, the prediction effect of the LSTM model is better under normal working conditions; while the SARIMA model works better under variable working conditions, and the average relative error of the SARIMA model is 0.069 4, 0.094 0 for the LSTM model. Combining the advantages of the two models, a gradient-driven time series prediction composite model is established. The average relative error of the model's 30-step prediction under the composite working condition is 0.060 1, which is lower than the error when the two models are used alone. Therefore, it is recommended to run in the field. Using the gradient-driven time series prediction composite model for prediction provides better data support for blast furnace gas regulation, rationally distributes gas, improves gas utilization, and reduces gas emission.
|
Received: 10 March 2022
|
|
|
|
[1] 熊文真,赵娜. 基于ARIMA-MC模型钢铁企业高炉煤气发生量预测[J]. 工业安全与环保,2016,42(10):100.(XIONG Wen-zhen,ZHAO Na. Prediction of blast furnace gas generation in iron and steel enterprises based on ARIMA-MC model[J].Industrial Safety and Environmental Protection,2016,42(10):100.) [2] 姜曙光. 济钢能源中心煤气平衡预测模型研究[D]. 济南:山东大学,2009.(JIANG Shu-guang. Research on Gas Balance Prediction Model of Jinan Iron and Steel Energy Center[D]. Jinan:Shandong University,2009.) [3] 肖冬峰,杨春节,宋执环. 基于改进BP网络的高炉煤气发生量预测模型[J]. 浙江大学学报(工学版),2012,46(11):2103.(XIAO Dong-feng,YANG Chun-jie,SONG Zhi-huan. Prediction model of blast furnace gas generation based on improved BP network[J]. Journal of Zhejiang University (Engineering Science),2012,46(11):2103.) [4] 岳有军,户彦飞,赵辉,等. 基于小波分析的ARIMA与LSSVM组合的高炉煤气预测[J]. 计算机测量与控制,2015,23(6):2128.(YUE You-jun,HU Yan-fei,ZHAO Hui,et al. Prediction of blast furnace gas based on wavelet analysis combined with ARIMA and LSSVM[J]. Computer Measurement and Control,2015,23(6):2128.) [5] 赖茜,王子豪,孙文强. 高炉煤气产生量预测方法对比研究[J]. 冶金能源,2021,40(3):22.(LAI Qian,WANG Zi-hao,SUN Wen-qiang. Comparative study on prediction methods of blast furnace gas production[J]. Metallurgical Energy,2021,40(3):22.) [6] 张鑫. 基于数据挖掘的高炉煤气受入量预测研究[D]. 唐山:华北理工大学,2020.(ZHANG Xin. Research on Prediction of Blast Furnace Gas Intake Based on Data Mining[D]. Tangshan:North China University of Technology,2020.) [7] 白鹏,王浩. 改进遗传算法优化的BP神经网络高炉煤气预测[J]. 机械工程与自动化,2021(2):77.(BAI Peng,WANG Hao. BP neural network for blast furnace gas prediction optimized by improved genetic algorithm[J]. Mechanical Engineering and Automation,2021(2):77.) [8] 吴玉杰. 钢铁企业煤气预测与调度建模[D]. 天津:天津大学,2010.(WU Yu-jie. Gas Forecasting and Scheduling Modeling in Iron and Steel Enterprises[D]. Tianjin:Tianjin University,2010.) [9] 徐化岩,马家琳. 基于数据驱动的高炉煤气复合预测模型[J]. 中国冶金,2019,29(7):56.(XU Hua-yan,MA Jia-lin. Data-driven composite prediction model of blast furnace gas[J]. China Metallurgy,2019,29(7):56.) [10] SUN W,WANG Z,WANG Q. Hybrid event,mechanism-and data-driven prediction of blast furnace gas generation[J]. Energy,2020,199(5):117497. [11] 李志刚,纪月,任雄朝. 基于LSTM与ARIMA组合模型的高炉煤气产生量预测[J]. 铸造技术,2018,39(11):2456.(LI Zhi-gang,JI Yue,REN Xiong-chao. Prediction of blast furnace gas production based on combined LSTM and ARIMA model[J]. Foundry Technology,2018,39(11):2456.) [12] 张琦,李鸿亮,赵晓宇,等. 高炉煤气产生量与消耗量动态预测模型及应用[J]. 哈尔滨工业大学学报,2016,48(1):101.(ZHANG Qi,LI Hong-liang,ZHAO Xiao-yu, et al. Dynamic prediction model and application of blast furnace gas production and consumption[J]. Journal of Harbin Institute of Technology,2016,48(1):101.) [13] 贺东风,官竹林,胡正彪. 基于分类和ARIMA-WT-LSTM模型的高炉煤气产生量预测[J]. 冶金自动化,2022,46(2):103.(HE Dong-feng,GUAN Zhu-lin,HU Zheng-biao. Prediction of blast furnace gas production based on classification and ARIMA-WT-LSTM model[J]. Metallurgical Automation,2022,46(2):103.) [14] Hochreiter S,Schmidhuber J. Long short-term memory[J]. Neural Computation,1997,9(8):1735. [15] 杨丽,吴雨茜,王俊丽,等. 循环神经网络研究综述[J]. 计算机应用,2018,38(s2):1.(YANG Li,WU Yu-qian,WANG Jun-li,et al. Review of recurrent neural network research[J]. Computer Applications,2018,38(s2): 1.) [16] 王莹,韩宝明,张琦,等. 基于SARIMA模型的北京地铁进站客流量预测[J]. 交通运输系统工程与信息,2015,15(6):205.(WANG Ying, HAN Bao-ming,ZHANG Qi,et al. Prediction of passenger flow in Beijing subway stations based on SARIMA model[J]. Transportation System Engineering and Information,2015,15(6):205.) [17] 汤岩. 时间序列分析的研究与应用[D]. 哈尔滨:东北农业大学,2007.(TANG Yan. Research and Application of Time Series Analysis[D]. Harbin:Northeast Agricultural University,2007.) |
|
|
|