A Seasonal Decomposition-Based Hybrid-BHPSF Model for Electricity Consumption Forecasting

A reasonable balance between energy consumption and production may be achieved with accurate electricity consumption forecasts, which assist in laying down operational expenses and resource waste. However, electricity consumption data exhibits nonlinearity, high volatility, and susceptibility to various factors. Existing forecasting schemes inadequately account for these traits, resulting in weak performance. This paper proposes a novel hybrid model (Hybrid-BHPSF) based on seasonal decomposition to address this issue. The proposed model incorporates a new BHPSF algorithm that effectively captures data patterns with noticeable variations. Initially, the electricity consumption data is segmented into multiple subsequences with distinct characteristics, and the BHPSF algorithm predicts the subsequences exhibiting clear trends. Subsequently, due to the ability of LightGBM to handle flat and nonlinear data, it is embedded into the model to process the remaining sequences that fluctuate irregularly within a certain range. We have evaluated our proposed model using four distinct datasets, and the results indicate that it outperforms existing models across different prediction horizons.