基于深度信念网络模型的多目标优化

为降低炼焦能耗,提高焦炭产量和质量,准确建立生产目标模型,提出基于深度信念网络模型的多目标优化研究方案。根据现场专家经验及生产现状确定能耗和产量为生产目标,对采集的炼焦数据进行处理和相关性分析,分别建立能耗和产量的深度信念网络模型及质量径向基神经网络模型,并且采用差分扰动的粒子群多目标优化算法进行集气管压力设定值优化,通过仿真研究验证了该方案的可行性。实验表明,该方案能准确地挖掘数据间的复杂特性,建立精准的目标模型,并得出最佳的集气管压力设定值,使炼焦能耗降低并且产量提高,可以为实际生产提供理论指导。

Research on Multi-objective Optimization Based on Deep Belief Network Model

In order to reduce coking energy consumption, improve coke yield and quality, accurately establish a production target model, a multi-objective optimization model is proposed, which is based on deep belief network. According to the field expert experience and production status, the energy consumption and the coke yield are determined as production targets. The collected coking data is processed and correlated, and a deep belief network model for the coking energy consumption and coke yield and a mass radial basis neural network model for the coke quality are established respectively. The differential particle swarm optimization is used to optimize the set value of the gas collector pressure. The feasibility of the method is verified by simulation. Experiments show that the method can accurately mine the complex characteristics between data, establish a precise target model, and obtain the best set value of the collector pressure, which can reduce the energy consumption of coking and increase the coke yield. This method can also provide theoretical guidance for actual production.