Electricity Production Prediction by Microsoft Azure Machine Learning Service and Python User Blocks

Introduction

Electricity consumption by the civilian and industrial enterprises depends on many factors: temperature, time of day (light/dark), weather conditions (in cloudy weather, people are more likely to stay at home), etc. Thus, communications are affected by different loads. If the network is overloaded, it may fail, causing severe consequences. The constant need to supply electricity to both civilian and industrial enterprises create the need for strong protection against breakdowns. Stopping the supply of energy to vital objects can cause a catastrophe and lead to such huge financial losses that in most cases it is more appropriate to spend money on preventing crisis situations (Wang et al., 2022a).

Solving this task can be done in two ways. The first one consists in studying and constantly monitoring the state of the equipment (Aksonov et al., 2021; Kombarov et al., 2021; Tsegelnyk et al., 2022), as well as in forecasting the state several periods ahead. This approach requires the creation of an information system and long-term work of experts and analysts. The result is achieved after a rather long period of time and requires significant financial costs. However, it is stable and allows to fully protect from risks in the long term.

An alternative way is to forecast electricity consumption. As a rule, experts are aware of the potential power grids and consumption information will be sufficient for them to expertly predict possible breakdowns. This way requires comparatively low financial expenditure or considerable time to collect representative samples. The information needed for forecasting is usually collected by sensors at power plants or utilities. This approach gives less stable results, but they appear for a much shorter period.

Formally, the researcher is faced with several tasks that are not determined in advance. In general, the goal of the research can be formulated as: “to obtain as much information as possible from the available data and to build the most accurate forecast possible.” Specific actions are chosen by the analyst based on his experience and available data. The lack of formalization forces preliminary research, which is called “descriptive analysis”. The organization of information, therefore, plays an important role in research planning and task formulation.

Forecasting the long-term needs of society in electricity helps to determine the electricity generating capacities for the future (Wang et al., 2022b; Wen et al., 2023). Such forecasts are also used to analyze the scope and composition of power supply expansion projects that include nuclear power. In the conditions of scientific and technical progress and improvement of the energy system of the state, forecasting becomes one of the decisive scientific factors in the formation of the strategy and tactics of energy development. To solve the problem of uncertainties, it is necessary to carry out comprehensive forecasting, which includes a number of forecasts of energy demand in the future.

The analysis should be conducted using up-to-date and consistent macroeconomic and microeconomic data so that electricity demand forecasts are more reliable and linked to demographic, economic and industrial development forecasts. Similarly to the analysis of demand for electricity, when determining the feasibility of implementing an energy project, it is necessary to conduct an analysis of the supply in the electricity sector. Such forecast shall be based on historical data, including consideration of past and present power supply trends, study of power generation systems and power transmission and distribution systems, as well as available resources for power generation and existing and planned connections to adjacent systems.

Currently, there are a number of works devoted to the practical application of machine learning (ML) technology in real life. A framework for regression-based ML that provides researchers with a common language and abstraction to aid in their study presented by Bojer (2022). To demonstrate the utility of the framework, is shown how it can be used to map and compare ML methods used in the M5 Uncertainty competition. In Tarallo et al. (2019) exploratory research presents the benefits of ML in sales forecasting for short shelf-life and highly-perishable products, as it surpasses the accuracy level of traditional statistical techniques and, as a result, improves inventory balancing throughout the chain, reducing stockout rates at points of sale, improving availability to consumers and increasing profitability.