Controlling an electrolysis cell using non-linear prediction model of process variables

Disclosed is a method of controlling an electrolysis cell for aluminium production. The method comprises the steps of measuring one or more process variables and passing the result of this measurement to an estimator. The estimator uses this to estimate the current value of one or more process variables. A prediction of the future values of one or more process variables is then made. This predicted value is used in the calculation of future control input scenarios. The prediction of the process variable is performed in accordance with a non-linear dynamic model based on first principles. The estimation of the current value is performed by either Kalman filtering techniques or moving horizon principles.