基于多尺度分解的ELM炉温预测研究

The Study on Temperature Prediction of ELM Furnace Based on Multiscale Decomposition

高炉数据的采样频率和各相物质的滞留时间不同呈现出多尺度特性,利用某钢铁厂采集的高炉生产数据,建立基于小波多尺度分解的极限学习机(ExtremeLearning Machine,ELM)的炉温预测模型。首先采用小波分解将硅含量和铁水温度等高炉过程参数的平稳时间序列分解为3个不同频率的细节信号和一个逼近信号;然后分别对每个细节信号和逼近信号建立ELM的子预测模型,将子模型的结果进行叠加,最终获得炉温的预测结果,并与其它预测模型进行比较;提出基于预测误差的概率密度函数的模型评价指标,该指标准确直观地反应了模型预测效果。仿真结果表明,该预测模型解决了多种采样频率的问题,其运算速度和预测精度大大提高,可为炉况的综合评价提供重要依据。

The sampling frequency of blast furnace and the retention time of each phase exhibit a multi scale characteristic,and the Extreme Learning Machine(ELM)prediction model of furnace temperature,which based on the wavelet multi-scale decomposition,is established by using the data collected by a steel plant.Firstly,we use wavelet decomposition to decompose the stationary time series of blast furnace process parameters,such as silicon content and molten iron temperature,into three different frequency detail signals and an approximation signal.Then the sub-prediction models of ELM are established separately for each detail signal and the approximation signal.The results of the sub-models are summed up to obtain the final result of furnace temperature prediction,then the prediction results of this model were compared with other models.This paper puts forward an evaluation metric based on the prediction error probability density function,which reflects the prediction effect of the model accurately and intuitively.The simulation results show that the prediction model has solved a variety of sampling frequency problems,and its operation speed and prediction accuracy are greatly improved,which can provide an important basis for comprehensive evaluation of furnace conditions.