The Research On Soft Sensing Modeling Of The Process Parameters For Aluminum Electrolysis Process Based On Extreme Learning Machine

The real-time measurement of process parameters is one of the important problems in the aluminum electrolysis production process.Through real-time monitoring of process parameters,it can provide timely decision support for the operation,but also can help workers analysis the cell states.Since many process parameters are difficult to be measured accurately in real time by instruments at present,it is of great significance to solve the problem by using soft sensor technology.Extreme learning machine,a new neural network algorithm,not only has good generalization performance,but also shows the advantages of low computational cost,ease of implementation,less parameter to control,which has great advantages in real-time measurement of industrial process parameters.Therefore,the research on soft sensing modeling based on extreme learning machine has obtained a lot of attention.In this thesis,with the data and mechanism of aluminum electrolysis process as background,the research on soft sensing modeling of aluminum electrolysis process parameters based on the extreme learning machine is conducted.The main work is as follows:1.A soft sensor modeling method based on the improved extreme learning machine is proposed.Firstly,the rough set theory is applied to reduce the irrelevant or redundant input variables.The method reduces the complexity of extreme learning machine input.After analyzing the relationship between the input variables and output variables by the partial correlation coefficient,the input variables which have the same effect on the output variable are put together.Thus,the input of extreme learning machine is divided into two parts,namely the positive part and negative part.Then,the corresponding network structure is built according to the two parts.The experiments show that the proposed method can not only reduce the number of hidden layer nodes which can ensure that the output matrix is column full rank,but also improve the precision and generalization performance of extreme learning machine.2.Because of different cell states,a soft sensor model based on the classification of cell states is proposed.Firstly,the dispersion is defined to express the weight of each process parameter,which can reflect the influence degree on the cell states.Then,the cell states are classified reasonably by using the fuzzy clustering method based on the process parameter weight.Finally,the corresponding soft sensor model for every cell state is established by using extreme learning machine.It also can help the classification management of the cell states.After every sub-model is built,it should combine each output of sub-model as the final measurement result.The experiments show that the fuzzy clustering method based on the process parameter weight has better classification in cells states.Meanwhile,the multiple extreme learning machine soft sensor model based on cell states classification have better prediction capability and stability in comparison with other single models.What's more,the multi-model can adapts to the change of cell status and has better generalization performance and robustness.