Research On Prediction Model Of Superheat In Aluminum Electrolysis Based On Multi-granularity

The aluminum manufacturing industry is one of the pillar industries of China's national economy.In recent two years,the contribution rate of total assets has exceeded10%,which is higher than the average level of 9%.The reason is that aluminum has many advantages such as strong castability,low density,and resistance to corrosion,which is widely used in aerospace,architectural decoration and electronic appliances.Superheat is an important parameter in the process of aluminum electrolysis,which is specifically defined as the difference between the temperature of the aluminum electrolyte and the temperature of the initial crystal.We can reduce the loss caused by the current and prolong the life of the electrolytic cell by controlling the superheat in a proper range.Measure the superheat in time is difficult,so it is important to prevention and control.By predicting the future's superheat.Therefore,in order to solve the problem of superheat prediction,the following research work is carried out based on theories of granular computing and Time-Series Data Mining.1.Aiming at the problem that it is difficult to measure the superheat in time,we propose a multi-granularity aluminum electrolytic superheat prediction model(Prediction Model Based on Time Granularity,PMBTG).Firstly,we preprocess the original data set,unify the frequency of the original data set,and select the attribute set to construct the features.Secondly,based on the multi-granularity idea,we define the size of the sliding window and divide the time within the sliding window and combine time grains to build a new feature set and sample set.Then,we oversample the sample set with an unbalanced sample size to make the sample ratio close to 1: 1.Finally,we use the classifier to train the new sample set to get the model.We use the data that provided by Shandong Weiqiao Aluminum Electric Co.,Ltd.to prove that this method can effectively solve the problem of superheat prediction,and it is superior to other models in the literature in terms of Precision,Recall,and F-Score.2.Aiming at the characteristics of large amount and high dimensions of aluminum electrolysis data,based on Spark parallel computing framework,we propose a multi-granularity aluminum electrolysis superheat parallel prediction model.First,we read the original data into Spark's RDD,and divide the data according to the electrolyticcell number.Second,parallelize the feature set construction and sample set construction proposed in the PMBTG algorithm respectively,so as to obtain a small sample set for each partition.Finally,we combine the results of all partitions to obtain the total sample set.We divide the total sample set into a test set,a training set,and a validation set,and use the different classifiers to obtain a model.The classification algorithm is called to verify the reliability and efficiency of the parallel algorithm on the Shandong Weiqiao dataset.Experimental results show that the parallel algorithm can solve the problem that the algorithm takes too long to run on large-scale high-dimensional data sets while ensuring the reliability of the original algorithm.In order to facilitate the use of the model by workers,we design and code an overheating prediction system.The system is very terse and it is very convenient to use.