Multi-Timescale Electric Vehicle Charging Load Forecasting Using a Feature-Correlation-Weighted Naive Bayes Model

To address the challenges of strong coupling in charging characteristics and insufficient behavioral data in multi-timescale EV charging load forecasting, this paper proposes an improved Naive Bayes-based method for multi-timescale EV charging load prediction. The paper first analyzes the spatio-temporal characteristics and coupling relationships of user travel chains across multiple time scales, deriving the driving mileage of electric vehicles. Building upon this, it integrates the impact of temperature on battery performance and air conditioning energy consumption, along with the influence of road grade on vehicle speed, to construct a unit-mileage energy consumption model. This model translates travel behavior into a basis for load quantification. Finally, a basic Naive Bayes prediction model is established using the aforementioned features and energy consumption basis. To address the limitations of its feature independence assumption, a feature correlation weighting correction mechanism is introduced, ultimately forming the FCW-NB model for multi-timescale EV charging load prediction. Experimental results demonstrate that the prediction curve employing multi-feature fusion weights exhibits the best alignment with actual load curves. In cross-seasonal forecasting, this method achieves accurate predictions across all six test scenarios compared to the standard Naive Bayes model. Its predicted charging load categories and probability distributions also align more closely with actual conditions, yielding superior application performance.