Modeling and Optimization of Electric Car Charging Loads for Orderly Charging Strategies

The large-scale integration of electric cars into the power grid presents both significant opportunities and formidable challenges. As a flexible load with distinct transportation attributes, electric cars exhibit strong mobility, randomness, and uncertainty. The surge in charging demand driven by the growing population of electric cars poses a severe challenge to the operational stability of power systems. Critically, the spatiotemporal distribution characteristics of charging loads directly constrain the scientific formulation and effective implementation of orderly charging strategies. Therefore, my research focuses on the modeling of electric car charging loads and the development of orderly charging strategies. This work aims to address the complexities introduced by the unpredictable charging behavior of electric cars, striving to balance grid stability with user satisfaction and economic efficiency. The central purpose is to devise a comprehensive framework—from accurate load prediction to intelligent, multi-objective optimization and finally to an intuitive visualization platform—that can facilitate the seamless and safe adoption of electric cars. The escalating penetration of electric cars underscores the urgent need for robust methodologies that can transform this mobile energy demand from a potential liability into a manageable and even beneficial grid resource.

To tackle the intricate problems associated with the high penetration of electric cars, I have structured my investigation into four primary research thrusts, each building upon the previous to form a cohesive analytical pipeline. The research is conducted at the regional level, encompassing traffic networks and distribution grids. The first stage involves developing an advanced simulation model to map the dynamic spatiotemporal distribution of vehicle loads. The subsequent stages focus on developing orderly charging strategies under different objective frameworks: first, a strategy based on general spatiotemporal characteristics, and second, a more refined strategy that accounts for the unique demands of different urban functional zones. Finally, I construct a visualization platform to integrate these models and strategies, enabling intuitive data analysis and management. This structured approach allows for a progression from fundamental understanding to practical application. The complexity of the system requires proper data hand-off between stages, ensuring that the insights gained from modeling directly inform the assumptions and constraints of the optimization strategies, and that the outputs of these strategies are effectively communicated to grid operators through the visualization platform.

1. A Dynamic Modeling Framework for Electric Car Charging Loads

Accurately predicting the spatial and temporal spread of charging load is the first and most critical step in this research. The challenge lies in the fact that charging behavior is not purely random; it is the result of a complex interplay between vehicle performance, environmental conditions, traffic dynamics, and individual human decisions. My preliminary work involved a systematic analysis of these influencing factors, which I categorized into three main groups: vehicle factors, environmental factors, and human factors. Vehicle factors include battery capacity and vehicle configuration. Environmental factors encompass weather conditions and real-time traffic states. Human factors involve driving style, route choice, and most importantly, the charging decisions made by users. Realizing the necessity of accounting for behavioral diversity, I determined that a single generic model is insufficient. Consequently, my research classifies private electric cars into three distinct operational types based on user travel needs: commuting type, ride-hailing type, and other general-purpose type. This classification is not arbitrary but is derived from an analysis of historical travel data, which reveals distinct patterns in travel times, daily distances, and parking durations for different user groups. Table 1 summarizes the key travel distribution features for each type, which are foundational inputs for the model.

The initial state of the vehicle is crucial. I model the initial battery state-of-charge (SOC) using a normal distribution to reflect the variation in charging habits and battery health. The initial parking locations are determined based on the analysis of urban functional zones. For instance, commuting vehicles are most likely to start their day in residential zones, while other types are more distributed between residential, work, and commercial zones. The journey itself is simulated on a coupled traffic-road network model. To capture the dynamic nature of travel, the model integrates a traffic speed-flow model, which calculates the vehicle’s speed as a function of the road’s capacity and the real-time flow of traffic. A dynamic road resistance model was developed to ensure that the path selection logic is realistic for different user types. Commuters are considered to minimize journey time, ride-hailing drivers to minimize distance traveled to reduce operational costs, and other drivers to strike a balance between time and cost. The Dijkstra algorithm is then employed to find the optimal path based on these dynamic impedance weights.

<<<<<<< Head
The core innovation of this section is the introduction of a dynamic energy consumption model based on fuzzy theory. Traditional models often assume a fixed energy consumption rate, which is a significant oversimplification. In reality, the energy needed per kilometer varies considerably with environmental conditions. My model takes into account the traction of temperature, which dictates the use of air conditioning or heating systems, and the density of real-time traffic, which influences the frequency of acceleration and braking. These two factors, normalized and fuzzified, serve as inputs to a fuzzy inference system. The output is a dynamic unit energy consumption value (kWh/km) that adjusts in real-time during the simulation. The fuzzy logic rules are designed to reflect real-world scenarios:

– **IF** temperature is uncomfortable **AND** traffic is smooth, **THEN** energy consumption is moderate.
– **IF** temperature is comfortable **AND** traffic is congested, **THEN** energy consumption is moderate.
– **IF** temperature is uncomfortable **AND** traffic is congested, **THEN** energy consumption is high.

The power consumption between nodes *i* and *j*, denoted as \( SOC_{i,j} \), can be calculated as follows:
\[
SOC_{i,j} = \omega \times l_{i,j} \times \frac{100\%}{E}
\]
where *\(\omega\)* is the dynamic power consumption per kilometer derived from the fuzzy controller, \(l_{i,j}\) is the length of the road segment, and *E* is the battery capacity. The state of charge at arrival node *j*, \(SOC_j\), is updated as \(SOC_j = SOC_i – SOC_{i,j}\).

The overall simulation flow, based on the Monte Carlo method, is performed to handle the randomness associated with each vehicle’s daily travel. For each vehicle, we generate its type, initial departure time, initial SOC, and destination based on the defined probability distributions. The vehicle’s journey is then simulated step-by-step according to the logic described, and its charging needs are evaluated at each destination or waypoint. A charging decision is made based on the remaining SOC. If the vehicle’s SOC drops below a certain threshold or is insufficient to reach the next destination, the vehicle enters a charging state at that particular location and time. The journey ends when the vehicle returns to its intended final parking location for the day. By aggregating the power drawn by all simulated vehicles over the simulation period, I obtain the total spatiotemporal distribution of the charging load. To validate the model, I compared the aggregated load profile with that from a baseline model using fixed energy consumption. The comparative analysis yields insightful results. The dynamic model exhibits smoother transitions between peak and off-peak hours and demonstrates higher peak loads, especially around 22:00. This indicates that dynamic energy consumption affects the state of charge and, consequently, the timing of charging events, making the model more responsive to environmental and traffic factors and producing a more realistic representation of load uncertainty.

An analysis of the load distributions from multiple functional zones further reinforces the model’s credibility. The residential zone shows a typical “double-peak” curve, with a high load concentrated in the evening hours as commuters return home and plug in their vehicles for overnight charging. The work zone displays charging clusters during the working hours, primarily between 8:00 and 13:00, showing how commuters charge at their workplaces. The commercial zone has a similar yet more volatile profile, with periodic peaks during the day related to the business operating hours. To explore the scalability and impact of EV penetration, simulations were run with varying car populations, as shown in Table 2. The results clearly show that with an increasing number of electric cars, both the total charging demand and the peak-to-valley load difference increase dramatically. This phenomenon—where periods of high demand align with the existing residential load evening peak—creates an “overlapped peak” effect, which can severely stress the distribution grid, increase the risk of overloading, and potentially reduce the lifespan of power equipment.

Table 2: The Impact of Electric Car Scale on Grid Load Profiles
| Number of Cars | Total Demand (kW) | EV Load Peak-Valley (kW) | Grid Peak-Valley (kW) |
| :— | :— | :— | :— |
| 500 | 20303.00 | 800.00 | 1470.70 |
| 1500 | 57087.00 | 1999.00 | 2640.35 |
| 3000 | 106898.00 | 3475.00 | 4045.39 |
| 5000 | 181870.00 | 5854.00 | 6480.29 |

This investigation forms the cornerstone of the thesis. The ability to produce high-resolution, dynamic spatiotemporal load predictions is paramount. It provides a robust foundation for the subsequent research on orderly charging strategies, allowing me to test their efficacy against a realistic and challenging baseline: the disorderly charging scenario.

2. An Orderly Charging Strategy Considering Temporal and Spatial Characteristics

Having established a realistic model for the disorderly charging load, the next phase of research is to design a control strategy to mitigate its negative effects. I first analyze the potential for control, or the “regulatory capacity,” of the electric cars. As described, most electric cars are parked and connected for a period significantly longer than their actual required charging time. This difference between charging duration and connection duration represents a valuable flexibility resource. The key is to shift the charging events within these time windows to achieve system-wide objectives without compromising the user’s needs. In this strategy, the grid sees a two-stage framework: time-level optimization and space-level optimization. At the time level, the objective is to reshape the total grid load profile by minimizing its peak-valley difference. At the space level, the goal is to avoid excessive concentration of load in specific nodes or time slots, which can be exacerbated by time-of-use tariff structures.

The regulatory capacity of a single electric car is the power it draws, but only during a specific controllable time window. This window is defined as the period starting from its arrival time (or an allowed start time) and ending at the latest possible start time that still ensures the user charging demand is fulfilled by the departure time. The framework assigns each car a new charging start time within this identified window. The theoretical basis for this strategy is rooted in the desire to avoid a scenario where, for example, all vehicles arrive home and immediately begin charging, creating an artificial peak. By shifting the start times to later in the evening or early morning, we can flatten the total demand curve.

To translate these objectives into a mathematical problem, I establish a multi-objective optimization model with the following objective functions:

1. **Grid Load Peak-Valley Minimization**:
\[
f_1 = \min\{ \max(P_t^{total}) – \min(P_t^{total}) \}
\]
where \(P_t^{total} = P_t^{base} + P_t^{EV}\) is the sum of the base grid load and the electric car charging load at time slot *t*. This objective aims to smooth the load curve and reduce the need for expensive peaking power plants.

2. **User Charging Cost Minimization**:
\[
f_2 = \min \sum_{k=1}^{K} \sum_{t=1}^{N_t} \delta_t \cdot p_{k,t} \cdot \Delta t
\]
where *K* is the number of vehicles, \(\delta_t\) is the time-of-use electricity price, and \(p_{k,t}\) is the charging power of vehicle *k* at time *t*. This objective aims to reduce the financial burden on users by incentivizing them to shift their consumption to off-peak periods.

3. **EV Load Spatial Equilibrium Maximization**: We also need to consider spatial distribution. The objective function \(f_3\) is designed to measure the smoothness of the charging load across different spatial zones. It ensures that while shifting load to low-price hours, we do not create new peaks in specific areas. Based on the Pareto frontier concept, I define an equilibrium index to manage the load distribution:
\[
f_3 = \max \left\{ P_{ev,opt}^{s} – P_{ev,ori}^{s} \right\}
\]
The final fitness function is a weighted sum of these objectives, normalized to the same scale. This converts the multi-objective problem into a single-objective one, allowing for the use of a modified particle swarm optimization (PSO) algorithm. The first step of the solution is to minimize grid load peak-valley difference. The second step is to minimize the cost to users. The overall load is the aggregate of base load and EV charging power.

To effectively solve this complex, non-linear optimization problem, I found that the standard Particle Swarm Optimization (PSO) algorithm is prone to premature convergence to local optima. To improve its global search ability and convergence speed, I proposed an Improved Particle Swarm Optimization (IMPSO) algorithm. The improvements are twofold. First, to enhance the diversity of the initial population, I replaced the standard random initialization with a logistic chaotic map. This maneuver ensures that the initial particles are well-distributed across the entire search space, thus preventing premature clustering and increasing the likelihood of finding the global optimum. Second, to balance the global and local search at different stages of the iteration process, I introduced dynamic adaptive parameters. The algorithm now begins with a large inertia weight and a large social learning factor, which encourages particles to explore the broad search space. As the iteration progresses, these values decrease to focus more on local refinement around the best solution. The inertia weight and learning factors are updated according to the following formulas:
\[
\omega_h = \omega_{max} – (\omega_{max} – \omega_{min}) \cdot \frac{h}{h_{max}}
\]
\[
c_{1,h} = c_{1,max} – (c_{1,max} – c_{1,min}) \cdot \frac{h}{h_{max}}
\]
\[
c_{2,h} = c_{2,min} + (c_{2,max} – c_{2,min}) \cdot \frac{h}{h_{max}}
\]
where *h* is the current iteration number and \(h_{max}\) is the maximum number of iterations.

A comprehensive set of simulations was carried out to validate the performance of the proposed strategy and the IMPSO algorithm. I established several comparative cases, each targeting different objectives. These include cases optimizing only one objective (e.g., grid peak-valley, user cost, or EV equilibrium), a case combining time-level objectives (grid peak-valley and user cost), and finally the proposed comprehensive strategy combining all three objectives. The results, shown in Table 3, demonstrate a clear trade-off between objectives.

Table 3: Optimization Results from Different Strategies
| Example | Load Peak-Valley Difference (kW) | Charging Cost (Yuan) | EV Spatial Equilibrium |
| :— | :— | :— | :— |
| Disorderly Charging | 2640.35 | 51691.82 | 0 |
| Strategy f1 | 1757.66 | 46528.26 | 17.09% |
| Strategy f2 | 2115.02 | 45229.56 | 8.68% |
| Strategy f3 | 2045.93 | 46945.50 | 29.48% |
| Strategy f1 & f2 | 1789.15 | 45749.10 | 19.40% |
| Strategy f1, f2 & f3 | 1888.81 | 46359.56 | 29.99% |

Analyzing the results from my proposed strategy (f1, f2 & f3), it successfully reduces the load peak-valley difference by approximately 28.5% compared to the disorderly charging scenario, demonstrating significant peak-shaving and valley-filling effects. Although the charging cost of 46359.56 Yuan is slightly higher than the cost for strategy f2 (45229.56 Yuan), it is a compromise to achieve a significantly better spatial equilibrium. Specifically, the EV spatial equilibrium, measured by the reduction in load variation across zones, reaches 29.99%, which is drastically better than the 8.68% achieved by focusing only on cost minimization (strategy f2). This underlines that a purely cost-driven approach can create new spatial load imbalances, even if it is optimal for grid load curve flattening and user bills. The proposed strategy provides a balanced compromise, validating the necessity of considering spatial aspects alongside temporal aspects.

Furthermore, the convergence curves for various algorithms highlight the superiority of the IMPSO. When compared to the standard PSO algorithm across various objective functions, the IMPSO consistently finds lower (better) final fitness values. The introduction of the chaotic map has enhanced the algorithm’s capability to escape local optima. The dynamic adaptive parameters have further improved convergence speed, ensuring that the algorithm found a superior solution with fewer iterations.

In summary, this chapter demonstrates that an orderly charging strategy considering both temporal and spatial characteristics is significantly more effective than conventional purely time-based optimization. It successfully balances grid-side stability, user-side economic interests, and the spatial uniformity of the load. It is a more robust and practical control strategy for future grids with high electric car penetration.

3. An Orderly Charging Strategy Specific to Urban Functional Zones

The previous chapter developed a strategy focusing on overall spatiotemporal load management. However, cities are not homogeneous. Different districts—residential, work, and commercial zones—have unique functions and user behavior patterns. Therefore, the regulatory needs of these zones differ. For example, in the daytime, residential users leave their cars, which creates a valley for charging, whereas in the morning, work zones and commercial zones see a sudden relocation and a subsequent potential for a new load peak. My research advances to incorporate these zonal characteristics, leading to a more granular and effective control strategy.

I initially analyzed the charging behavior of electric cars within these three functional zones. The statistical analysis uncovered distinct patterns in charging duration and power profiles. The specific durations for each zone is shown in the figure below. The figure shows the disparity in typical charging time lengths across different zones. This is a critical parameter for control. The duration and the scale are distinct.

Based on these characteristics, I devised a dual-dimension coupling strategy that aligns grid and user interests with zonal-specific requirements. For the residential zone, where users have long parking durations, the strategy prioritizes minimizing the user’s charging cost and filling the overall grid valley. Since flexibility is high, economic incentives via time-of-use pricing are highly effective. In contrast, for work and commercial zones, where users have short connection windows, the priority shifts to satisfying the users’ immediate charging needs and preventing sharp peaks in local grid demand. Incentivizing delayed charging here would lead to low user satisfaction, so the objective is to manage the load curve by dispersing charging requests without significantly delaying them. The optimization objectives for each zone are presented in Table 5.

While the grid load peak-valley remains an overall grid-side objective, the user-side objectives are distinct. The user-side problem is formulated by differentiating solely on the basis of grid-level total load peak-valley and EV zone-specific load peak.

The optimization model for this strategy is an extension of the previous one. It includes the grid peak-valley objective \(f_1\) from the previous chapter and adds two new objectives. To manage the work and commercial zones, I define an objective \(f_4\) to minimize the zone load peak:
\[
f_4 = \min\{\max(P_t^{ws})\}
\]
where \(P_t^{ws}\) is the combined electric car load in the work and commercial zones at time *t*. This prevents load from concentrating to form new peaks. To protect the user experience in these short-stay areas, I define an objective \(f_5\) to maximize charging satisfaction:
\[
f_5 = \max \left\{ 1 – \frac{\sum_{t} |P_{t,o}^{ws} – P_{t,d}^{ws}|}{\sum_t P_{t,d}^{ws}} \right\}
\]
where \(P_{t,o}^{ws}\) is the optimized load and \(P_{t,d}^{ws}\) is the disorderly load. Satisfaction is defined as being highest (100%) when the charging load matches the original disorderly pattern, which represents user’s immediate charging behavior upon arrival. Any time shift in the charge reduces the satisfaction index. For residential zones, the objective \(f_6\) is to minimize charging costs:
\[
f_6 = \min \sum_{k=1}^{K_H} \sum_{t=1}^{N_t} \delta_t \cdot p_{k,t} \cdot \Delta t
\]
where \(K_H\) is the number of cars in the residential zone. The strategy seeks to improve the whole grid’s load balance while ensuring any one zone is not overloaded and its users are not dissatisfied.

To solve this more complex problem, I used a modified version of the Non-dominated Sorting Genetic Algorithm II (NSGA-II), named GA-NSGA-II. Each objective of this multi-objective problem must be handled in parallel rather than being combined into a weighted single sum. This strategy uses two main enhancements. The first improvement is the use of Good Point Set theory for population initialization, rather than a random initialization. I evaluated the distribution quality and compared it against the standard random sequence and another chaotic sequence. The Good Point Set method outperformed the others, providing a more uniform spatial distribution and a lower discrepancy in frequency distribution, which is a critical requirement for multi-objective meta-heuristics to find a well-distributed Pareto front. The second enhancement involves an adaptive strategy for the crossover and mutation probabilities. This ensures higher exploration in the earlier stages and more exploitation in the later stages. The crossover probability decreases as iterations progress, calculated as:
\[
p_c = p_{c,max} – (p_{c,max} – p_{c,min}) \cdot \frac{h}{h_{max}}
\]
The algorithm then performs a fast non-dominated sort on the population, calculates the crowding distance for each solution, and selects the next generation based on these two metrics.

To evaluate the zone-specific strategy, which I will call Strategy 2, I compared its outcomes with Strategy 1 from the previous chapter. The aggregate grid-level results, as well as zonal-level metrics, are presented in Table 6.

Table 6: Comparison of Optimization Results between Different Strategies
| Example | Grid Peak-Valley Difference (kW) | WS Zone EV Peak (kW) | H Zone Charging Cost (Yuan) | WS Zone User Satisfaction |
| :— | :— | :— | :— | :— |
| Disorderly | 2640.35 | 4240 | 24793.02 | 100% |
| Strategy 1 | 1931.20 | 3920 | 22365.42 | 73.06% |
| Strategy 2 | 1933.77 | 3500 | 22232.28 | 79.20% |

The results demonstrate that Strategy 2 performs significantly better at the zonal level than Strategy 1. While both strategies have similar effectiveness in reducing the overall grid peak-valley difference (about 27% reduction), they differ substantially when looking at the regional distribution of the load. Of critical importance, Strategy 2 was able to lower the peak load in the WS zones from 4240 kW to 3500 kW, achieving a remarkable 17.45% reduction. In contrast, Strategy 1 only achieved a 7.55% reduction. This clearly shows that the fine-grained control of Strategy 2 is better at preventing new localized peaks from forming in these sensitive areas. Moreover, Strategy 2 achieves this without sacrificing user satisfaction. The user dissatisfaction in the WS zones is only 20.8% for Strategy 2, whereas it is 26.94% for Strategy 1. This is because Strategy 2 explicitly constrains the load shifting to respect the shorter connection times and higher need for immediate charging in these zones. The improvements are summarized in Table 7.

Table 7: Optimization Rate Comparison
| Strategy | Grid Peak-Valley | WS Peak | User Satisfaction decrement |
| :— | :— | :— | :— |
| Strategy 1 | 26.85% | 7.55% | 26.94% |
| Strategy 2 | 26.76% | 17.45% | 20.80% |

The analysis of the Pareto optimal fronts for both the NSGA-II and the GA-NSGA-Ⅱ algorithm reveals that the GA-NSGA-Ⅱ yields a more uniformly distributed and better-converged set of solutions in the objective space. This suggests that my modifications—integrating a uniform initial population and dynamically adjusting genetic parameters—enhance the algorithm’s ability to explore a broader search space and find more diverse, high-quality trade-off solutions for such a complex problem.

In conclusion, this chapter confirms that a one-size-fits-all strategy for the spatiotemporal optimization of electric car loads may be sub-optimal. By understanding and modeling the distinct charging behavior and needs of different functional zones, I can design control mechanisms that are more granular. The proposed strategy provides a more balanced and practical solution, ensuring grid stability, protecting the interests of users across all zones, and mitigating the risk of localized congestion.

4. Development and Implementation of a Visualization Platform

The complex algorithms and data structures used in the aforementioned models and strategies are not always transparent or accessible to grid operators and decision-makers. To bridge the gap between theoretical research and practical grid management, I developed a specialized visualization platform using LabVIEW. The platform’s purpose is to manage the entire workflow—from data import and load simulation to strategy implementation and result visualization—in a user-friendly graphical interface.

The first step involves a detailed requirement analysis. The platform must be able to import data relevant to electric car usage, such as vehicle travel information, traffic conditions, and baseline grid loads. Simulating load distributions is a core task. To achieve this, I integrate the dynamic load modeling method from Chapter 2 into the platform. The platform allows users to input parameters such as the number of electric cars, weather conditions, and road topology, and then runs the Monte Carlo simulation to produce an EV load profile, which is then displayed. The platform also provides a module for regulation strategy management. It incorporates the orderly charging strategies developed in Chapters 3 and 4, enabling the user to compare outcomes based on different objectives. Finally, the human-machine interface (HMI) module is the core of the platform, visually displaying all the input parameters, simulation results, and the outcomes of different optimization strategies. This module includes features for monitoring unexpected events or unplanned load patterns.

The platform was developed using LabVIEW’s graphical programming environment. A main VI was created to control the entire workflow, comprising several sub-visual instruments. The platform employs event-driven programming for interactive elements like buttons and menus. The program flow is orchestrated to first authenticate the user, then load the relevant data, initiate the simulation or load data, and finally execute the chosen optimization.

The final visualization platform interface showcases the multi-functional design. The login interface provides secure access to the platform. The main interface subsequent to login includes the load simulation plot and an area for load regulation results. The platform allows for the transparent adjustment of parameters. To demonstrate the serviceability of the platform, I have incorporated the two distinct strategies discussed before. For example, the user selects the regulation scope, chooses strategy one or two, and the platform will execute the corresponding optimized control algorithm. From the graphical interface, the user can compare the time-series load curves of disorderly vs. orderly charging. My platform also includes module for adjusting algorithm parameters before the run. Not only does the platform provide a vital monitoring and visualization tool for grid operators, but it also serves as a powerful educational and demonstration tool for researchers and stakeholders to understand the impact of policies, tariffs, and charging strategies on the grid.

The completed simulation platform demonstrates the feasibility and utility of integrating high-level control algorithms with user-centric visual tools. The platform effectively organizes the data flow, ensuring that all information is presented clearly and that the effect of different load management strategies is immediately visible. This directly supports grid operators in making informed decisions to ensure grid stability and efficiency in the face of increasing numbers of electric cars.

5. Conclusions and Outlook

This thesis has systematically tackled the challenges of integrating large-scale electric cars into the power grid, with a specific focus on load modeling, orderly charging control, and platform development. My research on modeling established a dynamic simulation framework that uniquely accounts for both traffic and weather conditions, and captures the distinct usage patterns of three different types of electric car drivers. This model reveals that overlooking environmental and human factors in load predictions can lead to under- or over-estimations that may have operational and economic consequences. The dynamic modeling method is superior in capturing the inherent uncertainty.

My subsequent investigation into orderly charging strategies reveals that a multi-objective approach is essential. A temporal-spatial strategy successfully balances grid peak shaving with user costs, but it can inadvertently induce uneven load distributions across different regions of the grid. However, when I refined the control to be aware of functional zones, the strategy was able to significantly reduce zone-specific peaks, particularly in areas with short-duration parking (work and commercial zones) while preserving user charging satisfaction and meeting grid-level goals. The use of advanced optimization algorithms like IMPSO and GA-NSGA-Ⅱ was critical for solving these complex, non-linear problems effectively.

Table 9: Key Findings of This Thesis
| Research Focus | Result |
| :— | :— |
| Dynamic Load Modeling | Reliable predictions of the uncertain spatiotemporal load distribution of electric cars. |
| Spatiotemporal Optimization | The created strategy reduces grid peak load and successfully balances time and space load distributions. |
| Zone-Specific Optimization | More capable of managing local peaks than non-zone-specific strategies, and with higher user satisfaction. |
| Visualization Platform | A user-friendly tool for monitoring, strategy comparison, and grid management performance is tested. |

The work presented here points toward several directions for future research. This thesis considers different types of private cars, but does not account for the load characteristics of public transport vehicles such as buses or taxis. Future work could extend the model to include these vehicle categories, making the load simulation model more comprehensive. The optimization strategies rely on a static time-of-use tariff. Futurework could explore more dynamic pricing mechanisms or real-time control methods that respond dynamically to real grid conditions. A deep reinforcement learning approach or a real-time market mechanism could be valuable for adapting to changing grid conditions. The visualization platform, while effective for simulating scenarios, could be enhanced though connection to real-world data, providing an actual decision support system for grid operations, perhaps incorporating more advanced 3D geographic information rendering and more advanced piloting applications.

In conclusion, this research contributes to the safe and efficient transition toward a sustainable transportation future, offering new insights into how we can manage the growing fleet of electric cars to benefit users and the electric grid. The continued evolution of electric cars promises a cleaner era of mobility, but it is contingent on the development and application of these comprehensive, data-driven control strategies.