Sewage treatment is a complex nonlinear process

and chemical oxygen demand (COD) is one of the key indicators for evaluating the effectiveness of wastewater treatment. It is costly and time-consuming to get COD by traditional chemical approaches. By neural networks

it is faster

but it is not accurate enough. To address them

a soft sensor approach

which is based on the combination of self-organizing map (SOM) and radial basis function (RBF) neural network

is designed. SOM is taken to cluster data samples. The number of hidden layer nodes and the center vector of the nodes are determined by clustering results. By such disposal

the rate of convergence and fitting precision have been improved. Part data of water samples from a sewage treatment plant are taken to establish the soft sensor model of COD. Test results provided by numerical model and hardware show that

compared with the traditional BP

RBF and other networks

the soft sensor model of COD designed in this paper has short measurement time and relatively high prediction accuracy. It may be a promising soft sensor approach in applications.