LDG Recovery Flow Prediction Based On LSSVM With Parameter Adjustment

LDG (Linz Donaniz Converter Gas) is a kind of important secondary energy of iron and steel enterprises, whose effective recycling and scheduling plays a key role to maintain the normal operation, and is also an effective way to response to the national energy saving and reduce costs. Since LDG recovery is not a continuous process and the user consumption is usually instable, a serious supply and demand imbalance in the secondary energy resource often appears. It is a significant research and task for scientific and accurate prediction of LDG recovery flow in order to provide users with a reasonable operational decision and maintain the energy balance between the amounts of supply and demand.This paper introduces the mechanism and influence factors of LDG recovery, analyzes the practical data in Baosteel, then carries out a series of comparative studies by using BP based neural network and least squares support vector machine (LSSVM) to predict LDG recovery flow. With a number of experiments, the LSSVM is selected as the prediction method in the paper, in which we then use particle swarm optimization algorithm to optimize the hyper parameters, via contrasting with a three-step grid search algorithm and chaos optimization algorithm. Then, a five fold cross-validation method is adopted to test and verify parameters, and finally build the prediction model based on LSSVM with parameter adjustment.Taking three LDG recovery systems in Baosteel as the examples, the proposed prediction method is verified by using the practical industrial data. The running results indicate that the accuracy and the reliability of the proposed methods can greatly meet the production demands, and be of great significance to improve the utilization of LDG and enhance their information technology level.