The analysis of potential energy surfaces (PES) is an effective method for investigating the structure and reactivity of organic molecules. Accurate representations of these surfaces require numerous quantum- chemical energy calculations for different configurations of the studied system and an analytical description of the surfaces. The entire process can be significantly accelerated by using machine learning and deep learning methods. The most efficient description of PES is obtained by spanning the surfaces according to the appropriate normal coordinates and using adequate expansion depending on the dimensionality of the space. In this thesis, theoretical research was conducted on the potential energy surfaces of benzene and its heterocyclic analogs: pyridine, pyridazine, pyrimidine, and pyrazine, using quantum-chemical calculations and machine learning and deep learning algorithms. Geometry optimizations and vibrational frequency calculations for all studied systems were performed at the B3LYP-D3/aug-cc-pVTZ level of theory. To investigate 1- and 2-dimensional sections of the PES spanned by the corresponding normal coordinates, all relevant structures of the studied molecules were generated using the qcc program, and quantum-chemical energy calculations were carried out. Using machine learning and deep learning algorithms implemented in the moonee program, the best possible regression models describing the selected sections of PESes were determined. The obtained analytical expressions for the description of potentials will be used for future molecular dynamics simulations and vibrational analysis.