Abstract:

Accurate electricity consumption prediction can provide technical support for the precise supply of electricity. To achieve high-precision prediction, a method based on meteorologically similar days for electricity consumption prediction is proposed. By deeply exploring the temporal fluctuation characteristics of electricity consumption and the meteorological sensitivity, and combining key meteorological factors, the matching degree of similar days is improved, thereby enhancing the prediction accuracy. Firstly, the linear correlation degree between each meteorological factor and electricity consumption is analyzed through Pearson correlation coefficient. Then, through P-value test, the key meteorological factors affecting electricity consumption are screened out by integrating the results of both. And the temperature data that exhibits the highest correlation with electricity consumption is classified using the K-means clustering algorithm. Historical days are categorized into two types: high-temperature days and moderate-temperature days, thereby establishing a similar-day typology framework. Secondly, within the same temperature category, the grey relational analysis is used to calculate the grey correlation degree between the meteorological factors of the target day and those of the historical days. By integrating Euclidean distance, the historical days most similar to the target day are screened out to deeply explore the temporal fluctuations and meteorological sensitivity characteristics of electricity consumption. Subsequently, a combined prediction model is constructed based on the similar-day data, comprehensively considering both meteorological factors and time-series characteristics, thereby enhancing prediction accuracy. Finally, the proposed method is validated using the electricity consumption data from a residential district in a certain area of Shandong Province. The results demonstrate that the proposed method has significant advantages in addressing the challenges of electricity consumption prediction.

Key words: electricity consumption prediction, k-means clustering, grey relational analysis, similar day, combinatorial forecasting