For sustainable development of agriculture, still necessary to reduce per-hectare costs of technological operations in the crop production. One of the important components per-hectare costs is diesel consumption of tractors. In addition to saving fuel, it is necessary to increase the productivity of machine and tractor units, which in turn often depends on the properties of the processed environments. In this case, such medium is soil, which can differ both in its granulometric composition and in physical properties. Therefore, it is necessary to use operational control of the equipment and take some measures to regulate these aspects. Today, this is becoming possible thanks to the development of modern digital technologies, gradually being introduced into our daily lives. One of that solutions is ability to machines and equipment remote control at field work performing. These systems include sensors that have opportunity to control various technical and technological processes. Data received from the sensors accumulated in the navigation controllers memory, devices that allow to determine geolocation and next transmit all data to remote server. Then the information about machine and tractor units operations can be analyzed and used for identify various relationships, and most importantly, for operation management of technological processes in the on-line mode. One of the options for applied application of monitoring data is response functions of modeled processes prediction with artificial intelligence algorithms. In this paper, was forecasted the tractor diesel consumption at field operations performing. As a result of existing machine learning algorithms review for regression analysis, was selected the most preferred. As parameters, was selected the output indicators of the machine-tractor unit operation and the properties of the processed medium. During preprocessing, was used the median filter for smoothing outliers. The forecast model was selected with the results of a comparative analysis from the following: random forest, prophet, skforecast, lstm architecture. For each of the listed options, was selected hyperparameters in order to increase the accuracy of the final forecast. After analyzing the results of the constructed models, was determined algorithms which demonstrate the greatest predictive accuracy and results stability. The obtained results can be used for field works planning in order to improving the efficiency of crop production technologies.