关键词: 铝电解, 黑猩猩算法, 模糊神经网络, 时间序列预测, 生长神经气

Abstract: The traditional prediction models need to consider a large amount of historical data when analyzing the time series data,and cannot quickly mine the knowledge information implied by real-time data,which causes problems such as low predicting efficiency.To solve the problems,a globally efficient time-series hybrid prediction model based on Growing Neural Gas and improved Adaptive Network-based Fuzzy Inference System (GNG-ANFIS)was proposed.The growing neural gas was used to dynamically track the collected time series data,the identification of data singularities was realized,and then the effective data was filtered.The improved chimp algorithm was used to optimize the traditional fuzzy neural network.The performance of the model was verified by using the time series data of iron content in aluminum liquid during aluminum electrolysis.The experimental results showed that the hybrid model could still accurately predict the iron content time series data while reducing the training time,which verified its feasibility.

Key words: aluminum electrolysis, chimp algorithm, fuzzy neural network, time series prediction, growing neural gas