基于核的冶金煤气流量在线区间预测

Kernel-based method for predicting online gas flow interval in metallurgical enterprises

针对冶金企业煤气系统的流量区间预测问题,本文提出一种基于核的在线区间预测构造方法,该方法将传统预测区间构造过程中对雅克比矩阵的复杂计算转化为对核的计算,大大降低了计算成本.为确定所提方法的超参数,采用共轭梯度下降算法来优化模型预测误差,使其逼近样本数据中有效噪声的方差.为验证本文所提方法的有效性,对现场实时数据库中的煤气流量数据进行了仿真实验,其结果表明本文方法在预测精度、可靠性和实时性三方面都表现出明显的优势.

In the kernel-based method, the calculation of the Jacobian matrix for determining the predicted interval in routine methods is converted into the calculation of kernels, thus greatly reducing the calculation costs. The hyper- parameters of the proposed model are determined by employing the conjugate gradient algorithm to minimize the model prediction error, making it to approach the variance of the effective noises in the sample data. To verify the effectiveness of the proposed method, we apply this method to construct prediction intervals of real gas flow data collection from the energy center of a steel plant. Results indicate that the proposed method is highly accurate and reliable with low computational costs. K