CNN-GA: deep learning-based response surface modelling integrated with genetic algorithm for extracting optimal solutions in highly nonlinear response surfaces

The ability to accurately model and optimize highly nonlinear response surfaces is crucial in various fields such as engineering, finance and environmental science, where complex, multi-variable interactions are prevalent. In this scenario, the present study provides a robust framework combining deep learning and genetic algorithm (GA) (convolutional neural network (CNN)-GA) to address these challenges, enhancing decision-making and innovation in these critical domains. Specifically, we employ a one-dimensional (1D)-CNN to model a simulated response surface. Later, random data points were sampled from this response surface, with a 10% random error added to simulate real-world variability. The dataset was split into training and testing sets, and the 1D-CNN model was trained to predict the response surface accurately. Following this, the trained model was utilized to reconstruct the response surface and determine the optimal input parameters that minimize the response variable using a genetic algorithm. The results demonstrate that the CNN-GA approach effectively captures the complexities of highly nonlinear response surfaces and identifies optimal solutions with high accuracy. Integrating deep learning and evolutionary algorithms offers a powerful tool for solving optimization problems in complex, nonlinear systems.