基于输入延迟支持向量机的氮气管网压力预测

Input-delay-based Support Vector Machine for Pressure Prediction in Nitrogen Pipeline Networks

在钢铁企业能源系统的低压氮气使用过程中,由于氮气使用单元分散且在管网中的位置不同,对管网压力的影响会出现短时间的延迟.鉴于此种情况,本文提出了一种基于影响因素输入延迟的多核最小二乘支持向量机对管网压力进行建模预测.该方法首先对低压氮气压力影响因素的延迟时间进行确定,提出一种基于因果关系的影响因素延迟时间计算方法,同时根据不同的影响因素和对应的延迟时间分别构造训练样本,进而建立基于最小二乘支持向量机的预测模型.通过对某钢铁企业现场低压氮气管网压力的两种不同情况,即正常工况和超限工况分别进行建模仿真验证,说明了本文提出的方法在压力预测上具有较高的精度.

During the consumption of low-pressure nitrogen in a steel industry energy system, the influence of nitrogen consumption units on the system pressure can cause delay due to their dispersion and different locations. In order to predict the pipeline network pressure, a factor input-delay-based multiple kernel least squares support vector machine （LSSVM） is reported here. A causality theory-based method is proposed to calculate the delay time. This first calculates the delay time of the factors influencing the system pressure and constructs training samples for different factors with the corresponding delay time. Then, the LSSVM-based model is established for prediction. To verify the effectiveness of the proposed method, two different practical conditions are considered, a normal and an abnormal one, and numerical experiments are conducted to validate the high accuracy of the proposed method.