Abstract: In terms of instability and difficulty in timely intervention during the combustion adjustment process, based on the characteristics of a certain type of gas turbine combustion adjustment process, an improved particle swarm optimization（PSO） algorithm was used to optimize Elman neural network. The parameters that affect the operating state of the unit were used as input variables, and the parameters that characterize combustion stability were used as output variables, thereby establishing an improved PSO-Elman neural network model. Result shows that the mass flow rate of the duty air, the opening of the compressor inlet guide vanes（IGV） and the first stage adjustable stationary blade of the compressor（CV1） have a significant impact on combustion stability. Compared with Elman neural network, the improved PSO-Elman neural network model has better reliability. The proposed model can well track the change characteristics of parameters during combustion adjustment, which can be used to predict possible combustion instability in advance, and solve the technical problems related with limitations and poor flexibility in the test process.