ПРОГНОЗИРОВАНИЕ ВРЕМЕННЫХ РЯДОВ: УЛУЧШЕНИЕ LSTM-МОДЕЛЕЙ С ПОМОЩЬЮ ВЕКТОРНО-ВРЕМЕННОГО КОДИРОВАНИЯ

Научный журнал. Инженерные системы и сооружения. 2025. С. 148–154.

Саввин Н. В.

This paper proposes a method to improve the accuracy of time series forecasting using vector-temporal encoding to enhance LSTM models. It is shown that a simple unidirectional LSTM with proper temporal feature encoding can outperform complex architectures such as Bi-LSTM and CNN-LSTM. The importance of temporal data representation for neural network efficiency is emphasized. The method was tested on short-term electricity consumption forecasting (5-minute resolution, 24-hour horizon). A cloud-based interface for result visualization was also developed.

Keywords: прогнозирование временных рядов