Parameter-optimized LSTM via SA-PSO algorithm for industrial process fault diagnosis

To solve the problems of dynamic timing modeling and parameter optimization in the fault diagnosis of chemical processes, an LSTM neural network diagnosis method based on the combination of simulated annealing algorithm (SA) and particle swarm optimization (PSO) is proposed. Through the global search ability of the PSO algorithm and the ability of the SA algorithm to jump out of the local optimum, the key parameters of LSTM were optimized, and a diagnostic model with the feature learning ability of time series was constructed. In the benchmark test of the TE process, the average diagnostic accuracy rate of the proposed method reached 96.4%, which was 5.2% higher than that of the unoptimized LSTM.