In recent years, the rapid adoption of electric vehicles globally has highlighted the critical role of charging infrastructure in supporting sustainable transportation. As the demand for electric vehicle charging grows, understanding customer charging behavior through big data analytics becomes essential for optimizing facility layout, improving service efficiency, and meeting the evolving needs of electric vehicle users. This study leverages extensive charging data to analyze patterns, influences, and future trends in electric vehicle charging, with a focus on applications in China EV markets. By examining factors such as charging time, location preferences, and pricing sensitivity, we aim to provide actionable insights for stakeholders in the electric vehicle ecosystem.
The proliferation of electric vehicles in China EV sectors has been driven by policy incentives and technological advancements, yet challenges like charging anxiety and infrastructure inefficiencies persist. Our research addresses these issues by applying big data techniques to dissect charging behaviors across various scenarios, including industrial zones, residential areas, and public parking lots. This analysis not only reveals the multifaceted nature of electric vehicle charging but also underscores the importance of data-driven strategies for enhancing the electric vehicle experience in China and beyond.
Data Preprocessing and Feature Engineering
To ensure the reliability of our analysis, we collected charging data spanning from January 2023 to April 2024, comprising over 700,000 charging records from multiple public charging stations. The dataset included key attributes such as charging time, location, customer attributes, and charging energy. Given the raw nature of the data, we implemented a comprehensive preprocessing pipeline to handle anomalies and standardize the information for further analysis.
Data cleaning involved removing outliers, such as charging sessions with durations less than 1 minute or exceeding 24 hours, which were often attributed to device malfunctions or operational errors. For missing values, we employed imputation techniques; for instance, missing charging energy values were filled using the average energy consumption for similar vehicle types at the same station. Additionally, numerical features like charging energy and duration were normalized to a [0, 1] range using the min-max scaling formula: $$ x_{\text{normalized}} = \frac{x – x_{\min}}{x_{\max} – x_{\min}} $$ This step ensured that all variables were on a comparable scale, reducing bias in subsequent modeling.
Feature engineering played a pivotal role in enriching the dataset. From the charging time attribute, we extracted temporal features, including whether a session occurred on a weekday or weekend, and segmented the day into peak (10:00–13:00 and 17:00–22:00), off-peak (07:00–10:00, 13:00–17:00, and 22:00–23:00), and valley periods (00:00–07:00 and 23:00–00:00). Spatially, we categorized charging stations into types such as public parking lots, industrial or enterprise parks, residential communities, and dedicated stations for large clients. This categorization allowed us to analyze regional and typological variations in electric vehicle charging behavior, which is crucial for tailoring strategies to specific China EV contexts.
| Feature | Description | Normalization Applied |
|---|---|---|
| Charging Time | Start and end timestamps of sessions | Segmented into time periods |
| Charging Location | Station type and geographical tags | Categorical encoding |
| Charging Energy (kWh) | Energy consumed per session | Min-max scaling to [0,1] |
| Charging Duration (hours) | Length of charging session | Min-max scaling to [0,1] |
Overall Charging Behavior Analysis
The analysis of charging behavior revealed distinct temporal patterns that are instrumental in understanding electric vehicle usage. Over a 24-hour cycle, charging energy and costs were lowest during the valley periods (00:00–07:00), indicating reduced demand overnight. In contrast, daytime hours (08:00–23:00) exhibited relatively stable charging activity, with minor fluctuations. This pattern aligns with typical daily routines, where electric vehicle owners charge their vehicles during work or leisure hours, reflecting the integration of electric vehicle charging into everyday life in China EV environments.
Seasonal analysis showed minimal variation in charging patterns across quarters, suggesting that electric vehicle charging demand is largely consistent throughout the year. This consistency simplifies grid management and operational planning for charging infrastructure. However, weekdays and non-workdays displayed notable differences: weekdays featured a sharp charging peak around 08:00, likely driven by commuter needs, while non-workdays had more dispersed charging activity without pronounced peaks. Such insights are vital for optimizing resource allocation and mitigating congestion during high-demand periods in the electric vehicle ecosystem.
Spatially, charging station utilization varied significantly across regions. Core urban areas with high population density and commercial activity recorded higher usage rates, whereas peripheral zones experienced greater idle capacity during off-peak hours. This disparity underscores the need for targeted infrastructure development to balance supply and demand, particularly in growing China EV markets. Customer loyalty also emerged as a key factor, with preferences influenced by pricing, convenience, and service quality. For instance, stations offering competitive prices between $0.93 and $1.27 per kWh attracted more consistent usage, highlighting the price sensitivity of electric vehicle users.
| Time Period | Average Charging Energy (kWh) | Average Cost (USD) | Remarks |
|---|---|---|---|
| 00:00–07:00 (Valley) | Low | Low | Minimal demand overnight |
| 08:00–23:00 (Daytime) | Stable | Moderate | Consistent usage patterns |
| Peak Hours (10:00–13:00, 17:00–22:00) | Higher | Higher | Linked to daily activities |
Analysis of Typical Charging Stations
We conducted a detailed examination of four station types to uncover nuanced electric vehicle charging behaviors. Each category exhibited unique characteristics shaped by user demographics and operational contexts, providing valuable lessons for the broader China EV industry.
Industrial and enterprise park stations primarily served employees, with charging sessions concentrated during morning and evening commutes. For example, one industrial station recorded over 9,000 sessions, with an average charging energy of 29.8 kWh per session and a duration of 1.9 hours. The total energy consumed was approximately 272,000 kWh, generating revenues of around $35,400. These stations demonstrated high usage regularity, making them ideal for implementing demand response programs. The reliance on workplace charging underscores the potential for corporate policies, such as charging subsidies, to further promote electric vehicle adoption in China.
Residential community stations catered to private electric vehicle owners, featuring longer charging durations averaging 3.6 hours per session but lower overall energy consumption. At one residential site, about 3,600 sessions resulted in 93,000 kWh of energy and $13,000 in revenue. The extended charging times, often overnight, indicate a preference for slow charging in home settings. This behavior aligns with the daily routines of residents and presents opportunities for incentivizing off-peak charging through dynamic pricing, which could alleviate grid stress in urban China EV networks.
Public parking lot stations showed the highest turnover, with an average session duration of just 0.8 hours due to the prevalence of fast-charging equipment. One public station handled nearly 50,000 sessions, consuming 1.365 million kWh and earning $192,000. The convenience and accessibility of these locations make them popular among diverse user groups, including taxi drivers and casual visitors. However, the high utilization rates necessitate robust maintenance and scalability to meet growing electric vehicle demand in China’s metropolitan areas.
Dedicated stations for large clients, such as logistics fleets, exhibited the highest average charging energy per session at 42.7 kWh, with total energy reaching 2.097 million kWh and revenues of $263,300. These stations prioritize efficiency and reliability, often operating during specific shifts. The centralized nature of large-client charging supports the integration of advanced technologies like vehicle-to-grid (V2G) systems, which could enhance grid stability and provide additional revenue streams in the China EV market.
| Station Type | Average Sessions | Average Energy per Session (kWh) | Average Duration (hours) | Total Energy (kWh) | Total Revenue (USD) |
|---|---|---|---|---|---|
| Industrial/Enterprise Park | 9,119 | 29.8 | 1.9 | 272,000 | 35,400 |
| Residential Community | 3,667 | 25.3 | 3.6 | 93,000 | 13,000 |
| Public Parking Lot | 49,432 | 27.6 | 0.8 | 1,365,000 | 192,000 |
| Large Client Dedicated | 49,077 | 42.7 | 0.9 | 2,097,000 | 263,300 |
Regression Analysis of Charging Price Sensitivity
To quantify the impact of pricing on electric vehicle charging behavior, we performed regression analysis using charging price as the independent variable and average energy per session as the dependent variable. This approach helped identify the optimal price range that maximizes charging activity, which is critical for formulating effective pricing strategies in China EV operations.
We applied three regression techniques: polynomial regression, support vector regression (SVR), and decision tree regression. The polynomial regression model, which captures nonlinear relationships, indicated that the optimal charging price lies between $0.93 and $1.27 per kWh, with a high coefficient of determination (R² = 0.89). The model can be expressed as: $$ y = \beta_0 + \beta_1 x + \beta_2 x^2 + \epsilon $$ where \( y \) is the average charging energy, \( x \) is the price, and \( \epsilon \) represents the error term. This quadratic fit revealed that prices outside this range led to reduced charging volumes, emphasizing the sensitivity of electric vehicle users to cost fluctuations.
SVR with a radial basis function (RBF) kernel further validated these findings, identifying an optimal price of $1.19 per kWh. The SVR model minimized the error tolerance (\(\epsilon = 0.1\)) and penalty coefficient (C = 1.0), producing a smooth curve that closely matched the data distribution. The decision tree regression, while less continuous, pointed to $0.93 per kWh as the optimal point, highlighting its utility in discrete pricing scenarios. Overall, these models consistently showed that electric vehicle charging demand is elastic, with price changes significantly influencing user behavior. This insight is paramount for China EV stakeholders seeking to balance profitability and customer satisfaction through dynamic pricing mechanisms.
| Model Type | Optimal Price (USD/kWh) | R² Score | Key Characteristics |
|---|---|---|---|
| Polynomial Regression | 1.27 | 0.89 | Captures nonlinear trends effectively |
| Support Vector Regression (SVR) | 1.19 | N/A (focus on error minimization) | Handles local fluctuations with RBF kernel |
| Decision Tree Regression | 0.93 | N/A (based on MSE splitting) | Suited for discrete pricing analysis |
Future Charging Demand Prediction Using Prophet Model
Predicting future charging demand is essential for proactive infrastructure planning in the electric vehicle sector. We employed the Prophet model, a time series forecasting tool that decomposes data into trend, seasonal, and holiday components. Using daily charging energy data from January 2023 to April 2024, we projected demand up to April 2025, providing a roadmap for resource allocation in China EV networks.
The time series decomposition revealed several key patterns. The trend component showed a saturation effect, with average daily charging energy stabilizing around 37,000 kWh in 2023-2024, followed by a 10% growth rate in early 2025. Seasonality analysis indicated that weekdays had 15-20% higher charging volumes than weekends, correlating with commuter patterns, while winter months experienced a 10-15% increase compared to summer, likely due to reduced electric vehicle range in colder climates. Holiday effects, such as during Spring Festival and National Day, caused 30-40% drops in charging energy, whereas shorter holidays led to approximately 10% declines.
The Prophet model generated forecasts with a mean absolute percentage error (MAPE) of 4.2% and a confidence interval coverage of 92.3%, indicating high reliability. The projected total charging energy for 2024 is approximately 14.28 million kWh (95% confidence interval [13.85, 14.62] million kWh), similar to 2023 levels. Peak daily demand is expected around December 20, 2024, at 62,000 kWh, while the lowest demand is forecast for September 17, 2024, at 13,000 kWh. These predictions enable grid operators and charging station managers to anticipate load variations and implement measures to prevent overloads or underutilization, thereby enhancing the resilience of China EV infrastructure.
The forecasting equation in Prophet can be summarized as: $$ y(t) = g(t) + s(t) + h(t) + \epsilon_t $$ where \( g(t) \) is the trend function, \( s(t) \) represents seasonal components, \( h(t) \) accounts for holiday effects, and \( \epsilon_t \) is the error term. This additive model effectively captures the complex dynamics of electric vehicle charging demand, supporting long-term strategic decisions.
Conclusions and Strategic Recommendations
This study underscores the transformative potential of big data analytics in understanding and optimizing electric vehicle charging behavior. Our findings reveal that charging patterns are influenced by a combination of temporal, spatial, and economic factors, with significant implications for the development of China EV markets. Key insights include the concentration of charging during daytime hours, the price sensitivity of users within the $0.93–$1.27 per kWh range, and the distinct characteristics of various station types. These results highlight the need for tailored approaches to infrastructure planning and management.
Based on our analysis, we propose several recommendations to enhance electric vehicle charging ecosystems. First, expanding charging capacity in industrial zones can accommodate high-demand scenarios, while residential areas would benefit from off-peak management initiatives, such as time-of-use pricing, to distribute load evenly. Second, implementing dynamic pricing mechanisms aligned with the identified optimal range can maximize utilization and customer satisfaction. Third, improving service quality at public stations through regular maintenance and technological upgrades, including fast-charging and V2G capabilities, will foster greater adoption of electric vehicles in China. Lastly, future research should explore cross-regional comparisons and emerging technologies like V2G to further advance the sustainability and efficiency of electric vehicle networks.
In summary, the integration of big data techniques provides a robust foundation for addressing the challenges and opportunities in the electric vehicle sector. By leveraging data-driven insights, stakeholders can not only optimize current operations but also pave the way for innovative solutions that support the global transition to electric mobility. The continuous evolution of China EV markets will rely on such analytical rigor to ensure that charging infrastructure meets the demands of a growing electric vehicle population.