The article proposes a software algorithm that includes three main blocks: a data processing block (cleaning and normalizing the original data), an interpolation block (selecting the optimal interpolation method based on the input data), and an extrapolation block (predicting future values using a trained neural network). The algorithm is implemented based on the following main operations: Data loading and preprocessing stage. The denoising procedure is carried out using a multi-criteria objective function. Initialization of the algorithm parameters occurs, including selection of window sizes, tolerances for possible errors. Data analysis follows, in which statistical characteristics such as correlations and derivatives are calculated. The information obtained is used to select the interpolation method that best suits the current data. Data collection was carried out on the basis of modern equipment manufactured by the “Sovtest ATE” enterprise (Russia). This project uses a model based on a DNN, which is capable of taking into account time dependencies and adapting to new data. Checking the achievement of the specified accuracy. If the accuracy is insufficient, a mechanism for adjusting the parameters is provided: a return to previous stages is implemented with the ability to change the parameters and re-select the method. Experiments on data search, interpolation and extrapolation were conducted on a set of one-dimensional and two-dimensional maps of the robotic cell link function motion.