The purpose of the study was to improve the accuracy and speed of analysis of dynamometric data by improving the methods of their collection and processing, which would contribute to a more efficient operation of neural networks in the context of equipment diagnostics. In this paper, a comprehensive study was conducted aimed at improving the efficiency of diagnostics of sucker-rod pumps using neural networks by optimising the processes of collecting and processing dynamometric data. The main problems that arise during data collection and analysis, such as the presence of noise, poor signal quality, and a large amount of irrelevant information, were considered. Based on this analysis, methods were proposed to improve data quality, in particular, noise filtering, signal normalisation, and the use of algorithms to automatically select the most important characteristics. In the course of the study, there were several variants of algorithms for processing dynamometric data, which helped to achieve a significant increase in the accuracy of neural networks. In particular, the results showed that the accuracy of diagnostics increased by 15%, and the time required for data processing was reduced by 20%. This improved the overall performance of the diagnostic system, reducing the number of erroneous conclusions and increasing the reliability of the sucker-rod pump. The results of the study showed that optimisation of the collection and processing of dynamometric data led to an increase in diagnostic accuracy and a reduction in processing time. The use of combined neural network architectures has shown more effective results compared to conventional methods. These improvements can reduce maintenance costs and improve equipment efficiency