Model for Forecasting the Volume of the Energy Market Under Undeterminable Conditions

The energy market is inherently quite sensitive to crises, particularly those that directly affect society. In particular, this declaration is related to companies engaged in the supply of electricity to individuals and small or medium-sized enterprises. The current consequences of the pandemic and the amplification of military confrontations around the world show that fundamental transformation processes, both purely technical and economic, are currently underway. For example, in Ukraine, it is necessary to rebuild external economic ties, protect against infrastructure destruction and take into account the ability of customers to pay for services on time. In order for companies involved in certain processes to respond rapidly to the challenges of today, it was decided to develop a model for forecasting market volumes under non-deterministic conditions, based on vector autoprogression algorithms and neural networks. The following methods are used: analytical and inductive methods for forming factors of external influence and description of target companies; expert evaluation method for determining the most influential external indicators; experimental method, statistical methods of processing time series and methods of multi-criteria evaluation to determine the most effective autoregression model. In the outlined circumstances, the most important component of the proposed approaches is the phase of data preprocessing, which allows to take into account the external influences that characterize both social changes and general changes in the industry. In the course of the research, it was found that this influence can be divided into two components - the behavior profiles of the target audience and the market structure. The first indicator includes the transformation text descriptions of customers and average energy costs, while the second includes the state of the world economy, monopoly and investment contributions, corporate financial stability and energy prices. In addition to the above, each profile contains adjustments to social changes. Experiments conducted demonstrated the relative effectiveness of the proposed approach to data preprocessing in combination with convolutionary neural networks and vector autoregression of moving averages. This allows authors to the of its use in practice with the possibility of further using more complex algorithms.