New Load Forecasting Ensemble Model based on LightGBM for Gas Industry Enterprises

The forecasting of the power consumption leads the main role in the process of planning of operation modes of industrial enterprises. The issue of forecasting the power consumption of the industrial enterprise is more complicated than forecasting the load curve of large system due to the less periodicity of the consumption, its greater variance, and the need to consider technological factors. The object of this article is enterprises of the gas industry. Because of their remoteness from centralized power system and great seasonal impact on gas recovery process the forecasting of power consumption of such enterprises affects not only the cost of consumption, but also the usage of own generation. This article introduces the novel usage of the LightGBM algorithm for the forecasting of the power consumption of the enterprise of the gas industry. In the article four experiments with various feature combinations were conducted. The need for consideration of technological and meteorological factors was discussed. Comparison of the models based on LightGBM, XGBoost, AdaBoost, Random Forest was presented. Model using LightGBM have the highest MAPE of $9.641 \%$ compared to others. This result was achieved with the regard of the retrospective of power consumption and technological process data. Furthermore, the topics of dataset creation and data preprocessing were established.