Aiming at the problem that it is difficult to online or rapidly detect on aluminum electrolysis process parameters

based on the production data of the 500 k A prebaked aluminum reduction cell

a soft sensor model based on particle swarm optimization online extreme learning machine is proposed. The online sequential extreme learning machine（ OS-ELM） is used to enhance the dynamic tracking capability of the system. Moreover

The particle swarm optimization algorithm is employed to find the optimal values of input weight and hidden layer bias in the network of extreme learning machine

so as to reduce the stochastic error. Furthermore

it is easy to fall into premature convergence phenomenon. Hence dynamic decay weights and learning factors are added to the speed update formula to balance global search and local search capabilities. Experimental results show that the proposed soft-sensor model has higher prediction accuracy

which is of great significance to ensure material balance and heat balance in aluminum electrolysis process.