数据与模型驱动的电熔镁群炉需量预报方法

电熔镁群炉需量指当前时刻k和（k-1），…，（k-n+1）时刻群炉功率的平均值，用于度量高耗能电熔镁群炉用电量.（k+1）时刻群炉需量取决于功率变化率.本文建立了功率变化率与电流控制系统输出电流之间由线性项与未知非线性项组成的动态模型，其中线性项通过电流被控对象的参数和控制器的参数计算，未知非线性项采用基于偏自相关函数（Partial autocorrelation function，PACF）输入变量决策的径向基函数神经网络（Radial basis function neural network，RBFNN）来估计.本文提出了由当前k时刻的需量和功率，（k-n+1）时刻功率及k时刻功率变化率的估计组成的（k+1）时刻需量的计算模型.通过某电熔镁砂厂实际数据的仿真实验和工业实验表明所提方法可准确预报需量变化趋势，可以防止因原料变化引起需量尖峰导致错误切断电熔镁炉供电造成电熔镁砂质量降低.

Data and Model Driven Demand Forecasting Method for Fused Magnesium Furnace Group

The demand of fused magnesium furnace group (FMFG) is the average value of powers at times k, (k-1), …, (k-n+1). The demand indicates the electricity consumption of the FMFG. The demand at time (k+1) depends on the rate of power change. In this paper, we develop a dynamic model of the rate of power change and the output current. The model consists of a linear term and an unknown nonlinear term, where the linear term can be calculated by the parameters of the controlled current and the controller, and the unknown nonlinear term can be estimated using the radial basis function neural network (RBFNN). The input variables of RBFNN are decided based on partial autocorrelation function (PACF). Then a computing model of demand at time (k+1) is proposed, which consists of the demand at time k, the powers at times k and (k-n+1) and the estimate of the rate of power change at time k. Simulations based on actual data and industrial experiments at a fused magnesia plant show that the proposed method can accurately forecast demand trends and can prevent reduction of fused magnesia grade caused by unnecessary cut off due to the demand spikes caused by change of raw materials.