چكيده به لاتين
In recent years, Artificial Neural Networks (ANN) have become one of the popular and effective models of machine learning. Neural networks possess a unique capability to handle very complex problems and the potential to predict accurate outcomes without a defined algorithmic solution. However, the structure and parameters of ANN are typically selected based on experience. This study presents new structures of neural networks that were created from the combination of metaheuristic algorithms and existing neural networks. The neural networks used in this research include Feed Forward, Radial Basis, and ANFIS, while the algorithms employed consist of Genetic Algorithm, Particle Swarm Optimization Algorithm, Colliding body Algorithm, and also an Enhanced colliding body Optimization Algorithm. These developed networks were examined for predicting the resistance of reinforced concrete with CFRP, semi-grouted masonry walls, and Cellular steel beams in this thesis. The results showed an improvement in the developed neural networks, with the top-performing type of network being identified. The superiority of the network models was determined using quality assessment criteria, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Normalized Mean Absolute Error (NMAE), Normalized Root Mean Squared Error (NRMSE), and Correlation Coefficient (R). Furthermore, an efficient graph method was presented for the optimal analysis of truss structures using the method of force, and computational time was compared with the method of displacements. Naturally, the optimization results and accuracy of calculations for both methods are the same, but the computational time for the method of forces is shorter when the degree of static indeterminacy (DSI) is less than the degree of kinematic indeterminacy (DKI). Optimization was performed using the ECBO algorithm in MATLAB.
