The promotion thesis is dedicated to the application of non-linear neural networks to operational and short-term forecasting of telecommunications network traffic. The properties of network traffic, influencing the accuracy of produced forecasts, as well as the main aspects of forecasting theory are described in detail. The advanced algorithm of solving a forecasting task by applying neural networks is proposed. The performance of the algorithm was tested on ten time series representing the real traffic loads of both a circuit switched telephone network and a packet switched IP network. The forecasts were also produced by some simpler linear methods such as seasonal exponential smoothing and SARIMA model as well as naïve methods. It gave the opportunity to make the conclusions about the necessity of applying a complex method of neural networks in each particular case. The thesis consists of two volumes and contains 25 figures, 11 tables, 96 annexes and 207 bibliographical names.