Soft Sensing of Feed Pulp Concentration in Thickener Based on Cross Validation LSTM Hyper-parameters Optimization

In the hydrometallurgical thickening production process, the feed pulp concentration in thickener is difficult to accurately measure online, and thus soft sensor, by constructing an estimation model that takes auxiliary variables as input and dominant variables as output, has been proposed as an effective solution. Deep learning which has excellent learning ability has been introduced in soft sensor to deal with the complex nonlinearity of the process, yet lacking the ability for dynamics. In this study, aiming at the problems of high cost, complex process, and low accuracy of the feed pulp concentration in thickener measurement, a deep neural network structure based on cross validation Long Short-Term Memory (LSTM) hyper-parameters optimization was proposed as a soft measurement method with complex nonlinearity and dynamic characteristics. Then it is applied in a real case of the feed pulp concentration in thickener. The experimental result demonstrates that the model of LSTM network has better effectiveness.