Propane dehydrogenation (PDH) has been considered as an alternative route for production of propylene. Despite the simple chemistry, the reacting system is very complex due to several side reactions and rapid catalyst deactivation. In this research, The effect of affecting parameters on the dehydrogenation of propane over industrial Pt–Sn/c-Al2O3 catalyst was studied by artificial neural network (ANN). Because of the complex nature of this reaction, in this study, an attempt has been made to employ ANN as an efficient and accurate tool to obtain the behavior of the system. In order to achieve the aim of this study, a database composed of 65 datasets with 3 model inputs (temperature (575-620 °C), water (0-0.6 mL) and time (3-7 h)) and one output (propane conversion) and one hidden layer (with 7 neurons) was established. Due to the weakness of BP to find the least exact global, sometimes the ANN model may result in undesirable results To increase the performance of ANN, various hybrid intelligent systems namely imperialist competitive algorithm (ICA)-ANN, genetic algorithm (GA)-ANN and particle swarm optimization (PSO)-ANN were applied and compared. In fact, ICA, GA and PSO were used to adjust weights and biases of the applied ANN model with 3-7-1 topology. A structure of the applied hybrid algorithms is predented in Fig. 1. For an ANN model with m neurons in input layer and n neurons in hidden layer, there is a m×n array matrix for weights and a n×1 array matrix for biases. In the following, for one neuron in output layer, there is a 1×n array matrix for weights and one variable as bias. The total number of the weights and biases is m×n+n×1+1×n+1= n×(m+2)+1 which are optimization variables and sum of square errors could be selected as goal function. The number of neurons in hidden layer, n, could be found out by a trial and error method. For each optimization problem, different optimizers might reveal different results. The same way, coupling ANN with different optimizers gives different performances. The obtained results of hybrid models were check considering two performance indices, i.e., mean square errors (MSE) and coefficient of determination (R2). The PSO-ANN hybrid revealed the best performance (R2=0.9750 and MSE=0.1008). Finally, PSO-ANN was used to optimize the operational conditions and the optimum was found 48.79% at (temperature=620 °C, water=0.5636 mL and time=3 h) which is in good agreement with maximum of experimental data 48.67% at (temperature=620 °C, water=0.5 mL and time=3 h).