In order to study the function and application of partial least squares (PLS) combined with key performance indicators (KPI) in system fault diagnosis, a simulation model was established to simulate wastewater treatment based on the mathematical model of activated sludge (BSM1) and combined with PLS. By collecting the input and output of the system under normal conditions and establishing a relationship matrix, the determination of whether a fault occurs in actual application could be achieved simply by input. Based on the simulation model, the performance of PLS in system fault diagnosis could be verified through changing the conditions of weather, input, and prior knowledge, as well as introducing a multiplication coefficient which can be manually triggered to the simulation system to simulate the fault. Results show that the PLS method had the function of judging the location of the fault, and the sensitivity and accuracy of the function could be improved by combining with KPI. The sensitivity and accuracy of fault diagnosis were affected by the weather factors simulated in the model and the prior knowledge in the algorithm.