Aiming at complex nonlinear problems in an sequencing batch type activated sludge process (SBR) and poor precision of sewage water quality model established by conventional neural network, applies an support vector machine to set up BOD soft measurement model, and improves the SVM parameter through particle swarm optimization. The simulation results show that compared with the BP neural network and standard SVM model, the PSO-LIBSVM has small error and high precision. It decreases the model complexity, improves its generalization ability, and achieves good prediction effect.