Abstract <p>A system for integrated protection of the traction units of a diesel locomotive based on an individual electric drive is proposed. The system combines the functions of protection against spinning and overheating traction motor windings. In order to provide an increase in the operational speed of the system and to simplify its structure, the use of a neural network for thevoltage control of a traction electric motor in emergency modes hac been justified. The voltage control provides restricting wheel slip with respect to the rail within 3% and the temperature of the armature winding of the electric motor within 5°C as to compare to the permissible value. The neural network architecture provides for two hidden layers and six input neurons. In order to improve the accuracy of the network, a procedure for standardizing the values of input neurons is used. The output neuron generates the voltage value at the input of the traction motor. A linear activation function is used for the voltage. Activation functions of the ReLU type have been taken for hidden network layers. The neural network was trained with the use of calculated simulation data for controlling the voltage of a traction electric motor for to restore the grip between the wheel and the rail, as well as for to maintain the temperature of the motor armature winding at a preset level. The results of the network training have shown that its architecture meets the requirements of the problem under solving: after the 20th training cycle, the error is less than 1% of the preset voltage value. When the voltage across the traction motor is less than 600 V, the generated value of the target voltage almost corresponds to the preset value. When a traction motor voltage is higher than 600 V, the target voltage generated can be determined ambiguously, whereas its maximum deviation from the preset value does not exceed 8%, which could be explained by insufficient statistical data the motor voltage higher than 600 V.</p>