基于神经网络的透气性状态预测

Neural-network-based Prediction of Permeability

针对烧结过程的时变、强非线性等特点,基于神经网络和粒子群优化算法,提出一种预测透气性状态的集成方法.采用神经网络分别建立透气性预测模型,采用粒子群优化算法对神经网络进行训练,提高预测模型的实时性;进而借助模糊分类器将预测子模型实现有机融合.最后实际运行结果表明,提出的集成模型具有较高的预测精度和较强的自学习能力,并且在工况波动严重的情况下,仍然具有好的预测效果.

In order to deal with time varying and strong nonlinearity of lead - zinc sintering process, an integrated method of predicting synthetic permeability based on neural networks（NNs） and a particle swarm optimization（PSO） algorithm has been developed. Firstly,the model of the exponent of synthetic permeability is constructed. Secondly, NNs are used to establish two models of time sequence and technological parameter for predicting the permeability, and a PSO algorithm is applied to train the NNs so as to improve real - time of the models. Thirdly, a fuzzy classifier is used for combining the two models with an intelligent integrated model for predicting the permeability. Finally, the results of actual runs show that the proposed integrated prediction model possesses high precision and good ability of self - learning. Under the condition of big fluctuation, it still has good effect.