Analyzing Electric Vehicle Charging Demand in Boulder City
Boulder City, Colorado, has seen a significant rise in electric vehicle (EV) charging stations (CS) from January 2018 to November 2023. A comprehensive dataset capturing charging sessions reveals insights into the city’s evolving infrastructure and user behavior.
Dataset Overview
The dataset comprises detailed information on 26 charging stations, with each entry representing a unique charging session. Key columns include Station Name, Address, City, and State Province, essential for pinpointing the location of each station. The dataset also records the Start and End Date Time, Total Duration, and Charging Time, offering a complete view of each session’s duration and energy consumption. Additionally, metrics such as Energy kWh, GHG Savings kg, and Gasoline Savings gallons highlight the environmental impact of each charging session.
Charging Sessions and Demand Analysis
Charging demand fluctuates based on various factors, including time of year and day of the week. Analysis shows 2023 as the peak year for charging sessions, with summer months (May to October) experiencing the highest activity. Conversely, the demand decreases from November to March. Interestingly, Monday and Wednesday host the most charging sessions, while Tuesday and Thursday see the least.
Hourly trends indicate a peak in charging activity between 7 AM and 6 PM, with a significant drop-off after 6 PM. This pattern aligns with typical commuting hours, suggesting that many users charge their vehicles in conjunction with their daily routines. Fig. 4 provides a visual representation of these trends.
Weather Impact on Charging Behavior
Weather conditions significantly influence EV charging patterns. Boulder City’s weather data, meticulously aligned with charging sessions, reveals no missing values. The dataset tracks temperature, wind speed, and precipitation on an hourly basis, providing a clear picture of how weather affects charging behavior. Fig. 9 illustrates these weather patterns.
Selected Charging Locations
Three locations—N Boulder Rec 1, Baseline St 1, and Carpenter Park 1—stand out in terms of charging sessions. Data analysis at these sites reveals distinct patterns. For instance, N Boulder Rec 1 experiences more weekday sessions, while Baseline St 1 and Carpenter Park 1 see higher activity on weekends. Seasonal trends also show a higher demand from May to August, with a decline toward year-end.
Fig. 6 presents a detailed comparison of charging sessions and demand across these locations.
Model Implementation and Prediction
To predict future charging demand, a Context-Aware Temporal Transformer (CAT-Former) model was developed. This model integrates temporal features (e.g., hour, day of the week) with contextual elements like weather and station location. The model employs multi-head self-attention mechanisms to understand dependencies across time and context, optimizing predictions with a mean squared error loss function.
Hyperparameter Optimization
Hyperparameter optimization was crucial to enhancing the model’s predictive accuracy. Parameters such as the number of epochs, batch size, and learning rate were fine-tuned through experimentation. The Adam optimizer with an adaptive learning rate was employed, ensuring stable model updates. The training employed a batch size of 32, with performance assessed using Mean Absolute Error (MAE) and Mean Squared Error (MSE) metrics.
Data Preprocessing
The dataset was split into training (January 2018 to June 2023) and test (July 2023 to November 2023) sets. This segmentation provided a robust framework for training and evaluating the model’s performance. The preprocessing utilized a MinMaxScaler to normalize feature values, enhancing model convergence.
Original Story at www.nature.com