In the context of contemporary economic and industrial development, the demand for energy has grown substantially, while fossil energy sources face scarcity. Concurrently, the large-scale emission of greenhouse gases from fossil fuel consumption has led to abnormal climate changes. Nations worldwide have issued relevant policies to vigorously develop clean energy and promote energy transition. As a vital component of the new power system, the electric vehicle, owing to its green, clean, and energy-saving characteristics, has been widely adopted. Electric vehicles, as flexible loads with transportation attributes, exhibit strong mobility, randomness, and uncertainty. With the large-scale growth in the number of electric vehicles, charging demand has surged, and large-scale disorderly charging behavior poses a severe challenge to the operational stability of the power system. Furthermore, the distribution characteristics of charging loads directly constrain the scientific formulation and effective implementation of orderly charging strategies for electric vehicles.
According to data released by the National Bureau of Statistics, by the end of 2024, the number of new energy vehicles in China reached 31.4 million, accounting for 8.90% of China’s total automobile ownership, with pure electric vehicles reaching 22.09 million. In 2024, the production and sales of new energy vehicles reached 12.889 million and 12.866 million units, respectively. As the penetration rate of electric vehicles continues to rise, the large-scale growth of charging load has formed a two-way impact on the power system. On one hand, as mobile flexible loads, electric vehicles provide flexible regulation resources for the grid through vehicle-to-grid interactive technology. On the other hand, due to the influence of users’ charging behavior and vehicle equipment differences, the load distribution of electric vehicle charging has a high degree of randomness and uncertainty, which greatly increases the imbalance and volatility of distribution network loads. In particular periods, a large number of electric vehicles charging simultaneously can easily cause a “peak on peak” phenomenon, exacerbating the risk of grid operation.
Motivated by these challenges, this thesis focuses on three aspects: modeling of disordered charging load distribution for electric vehicles, optimization and regulation of orderly charging for electric vehicles, and design of a visualization platform for electric vehicles. The research provides new ideas and methods for solving the optimal dispatch problem of electric vehicle loads.
1 Background and Literature Review
Research on electric vehicle load distribution simulation has been extensive. In early stage, scholars mainly used statistical analysis methods for probabilistic modeling of electric vehicle uncertainty, followed by Monte Carlo simulation or Markov chains to simulate load distributions. With the rapid development of information technology, big data and artificial intelligence have been applied to load modeling formats. In the modeling of charging load temporal characteristics, the start time of charging directly affects the load distribution and the state of charge. However, many earlier studies only considered temporal characteristics without addressing spatial differences due to the mobility characteristics of electric vehicles.
In response, scholars have conducted two-dimensional research considering both time and space. The different functional zones are analyzed using Markov trip chains to obtain user travel paths. The coupling of traffic networks with electric vehicle travel has led to the use of optimal path algorithms and dynamic traffic models. In contrast to statistical approaches, data-driven methods leveraging actual charging records have been proposed using neural networks and conditional generation adversarial networks. However, these methods require a certain quantity and quality of charging data and have associated privacy issues.
Research on orderly charging strategies mainly focuses on dispatching potential assessment, charging strategy optimization, and algorithm solving. The dispatch potential is calculated using data-driven and mathematical models. Various optimization objectives such as peak-valley difference minimization, load variance minimization, user charging cost minimization, and grid loss minimization have been considered in different studies. Some research has proposed double-layer optimization frameworks considering both grid and user interests, and different game models among electric vehicles, aggregators, and distribution network operators. With the integration of renewables, coordinated optimization strategies for electric vehicles and photovoltaic generation have been investigated.
Parallel to these, visualization platform studies have adopted different development frameworks. The platform architectures use either stand-alone instruments like Matlab GUIDE and LabVIEW or web-based frameworks with geospatial visualization for the management and monitoring of electric vehicle loads.
Despite the progress achieved, there remain gaps in understanding the spatial coupling and dynamic characteristics of electric vehicle charging loads, which significantly impact the regulation strategy. I therefore need models that dynamically account for user travel needs, and regulation strategies that incorporate both temporal and functional zone requirements. Based on these observations, in this thesis, the following research contents are presented:
1. Establishment of a dynamic spatiotemporal load model framework for electric vehicles considering user travel demand.
2. Development of a spatiotemporal load regulation strategy considering distribution characteristics.
3. Development of a functional area-aware regulation strategy.
4. Design of a LabVIEW-based visualization platform.
2 Dynamic Modeling of Electric Vehicle Load Considering User Travel Demand
2.1 Analysis of Factors Influencing Load Distribution
As a means of transport, electric vehicles have strong mobility. The spatiotemporal distribution of the charging load is determined by charging time, charging mode, and charging environment. Because the factors are influenced by the conditions of weather, traffic, and human decision-making, the load distribution is random and uncertain. I consider influencing factors from three parts: vehicle factors (configuration, battery state), environmental factors (weather, traffic conditions, charging facilities), and human factors (driving habits, route selection, charging decisions, travel demand).
2.2 Modeling of Load Distribution Characteristics Based on Travel Demand
To describe the randomness and dynamic characteristics of load distribution, I first classified the private electric vehicles according to their trip purposes into commuting trips (Type 1), ride-hailing trips (Type 2), and other trips (Type 3). Then, for each type, I modeled the initial trip time, trip end time, state of charge (SOC), charging time, and vehicle speed.
(1) Vehicle state characteristic model
| Type | Numerical proportion | Distribution area and proportion |
|---|---|---|
| 1 (commuting) | 50% | H1 (60%), H2 (40%) |
| 2 (ride-hailing) | 20% | H1 (60%), H2 (40%) |
| 3 (other) | 30% | W (25%), S (25%), H1 (25%), H2 (25%) |
The initial travel time for different vehicle types is described by the Gaussian mixture distribution
$$p(x)=\sum_{k=1}^{K} \omega_{k} N_{k}(x;\mu_{k},\sigma_{k}),$$
$$N_k(x;\mu_k,\sigma_k)=\frac{1}{\sigma_k\sqrt{2\pi}}\exp\left(-\frac{(x-\mu_k)^2}{2\sigma_k^2}\right),$$
where \(p(x)\) is the probability density, \(K\) is the number of Gaussian components, \(\mu_k\) and \(\sigma_k\) are the mean and standard deviation of component \(k\), and \(\omega_k\) is the weight of component \(k\).
The initial SOC is modeled by a normal distribution:
$$f_{cSOC}(x)=\frac{1}{\sigma_1\sqrt{2\pi}}\exp\left(-\frac{(x-\mu_1)^2}{2\sigma_1^2}\right).$$
When traveling from node \(i\) to node \(j\), the SOC after arriving at node \(j\), denoted \(SOC_j\), is computed as
$$SOC_{i,j}=SOC_i – \frac{l_{i,j} \cdot \omega}{E} \times 100\%$$
where \(SOC_{i}\) is the SOC at node \(i\), \(SOC_{i,j}\) is the SOC consumed from node \(i\) to node \(j\), \(l_{i,j}\) is the distance between the two nodes, \(E\) is the battery capacity, and \(\omega\) is the energy consumption per kilometer. The charging time is computed as
$$T_c = \frac{(SOC_{end} – SOC_{start}) \cdot E}{\eta \cdot P_c},$$
where \(T_c\) is the charging time, \(SOC_{start}\) is the initial charging SOC, \(SOC_{end}\) is the charging ending SOC, \(P_c\) is charging power, and \(\eta\) is the charging efficiency.
(2) Dynamic energy consumption model
I constructed a fuzzy inference model to describe the dynamic energy consumption per kilometer based on normalized temperature and traffic conditions. The normalization is:
$$X_{tem} = 0.5 + p \cdot [\text{sgn}(t – t_{min}) – \text{sgn}(t – t_{max})],$$
$$X_{tra} = \frac{v – v_{min}}{v_{max} – v_{min}},$$
where \(X_{tem}\) and \(X_{tra}\) are the normalized temperature and traffic condition, \(t\) is the real-time temperature with limit \(t_{max}\) and \(t_{min}\), \(p\) is a scaling coefficient, and \(v\) is the driving speed. The fuzzy controller uses these two input variables and produces a normalized output that describes the energy consumption level.
| \(X_{tra}\) \ \(X_{tem}\) | \(TE_L\) (suitable temperature) | \(TE_H\) (unsuitable temperature) |
|---|---|---|
| \(TC_L\) (smooth traffic) | \(W_L\) (low consumption) | \(W_M\) (medium consumption) |
| \(TC_H\) (congested traffic) | \(W_M\) (medium consumption) | \(W_H\) (high consumption) |
(3) Dynamic road resistance model
I represent the urban road network with a graph \(G(V, E, D)\). The adjacency matrix \(D\) contains the distances \(d_{i,j}=l_{i,j}\) if nodes \(i\) and \(j\) are connected, otherwise \(d_{i,j}=\infty\), and \(d_{i,i}=0\).
The speed of the traffic flow on the segment between nodes \(i\) and \(j\) at time \(t\) is
$$v_{ij}(t) = \frac{v_{ij}^0}{1 + \beta_{ij} \cdot \left(\frac{q_{ij}(t)}{C_{ij}}\right)^n}, \quad
\beta_{ij} = a + b \cdot \left(\frac{q_{ij}(t)}{C_{ij}}\right)^n,$$
where \(v_{ij}^0\) is the free-flow speed, \(C_{ij}\) is the capacity, \(q_{ij}(t)\) is the flow at time \(t\), and \(a, b, n\) are coefficients.
The dynamic path weight is determined according to different trip purposes:
$$L_t = \begin{cases}
\frac{l_{ij}}{v_{ij}^T}, & \eta_{yp}=1 \\[4pt]
l_{ij}, & \eta_{yp}=2 \\[4pt]
\rho_t \cdot \frac{l_{ij}}{v_{ij}^T} + \rho_x \cdot l_{ij}, & \eta_{yp}=3
\end{cases}$$
where \(v_{ij}^T\) is the travel speed at time \(T\), \(\eta_{yp}\) indicates the vehicle type (1 for commuting, 2 for ride-hailing, 3 for other trips), and \(\rho_t\), \(\rho_x\) are weight coefficients.
Finally, the Dijkstra algorithm, whose edge weights are the dynamic road resistances in this work, is adopted to choose the route from origin to destination.
2.3 Monte Carlo Based Charging Load Simulation
I used Monte Carlo simulation to model the whole day trips of the electric vehicle fleet. The charging decision logic after each trip is described in the flow chart. The simulation terminates when the convergence condition is met:
$$\max \left| \frac{1}{N_m}\sum_{i=1}^{N_m} P_i – \frac{1}{N_{m-1}}\sum_{i=1}^{N_{m-1}} P_i \right| < \varepsilon,$$
where \(N_m\) is the number of Monte Carlo simulations, \(P_i\) is the charging power matrix for the \(i\)-th simulation, and \(\varepsilon\) is the convergence precision.
2.4 Case Study of Modeling
I coupled a 29-node traffic network with the IEEE 33-bus distribution network. One day is divided into 96 time intervals of 15 minutes each. The total number of electric vehicles in the region is 1500, with a battery capacity of 40 kWh. The slow charging pile power is 7 kW and the fast charging pile power is 20 kW; the charging efficiency is 0.9. The upper and lower SOC limits are set as 0.9 and 0.2.
| Scale | Charging demand (kW) | EV peak-valley difference (kW) | Grid total peak-valley difference (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 |
Through the proposed dynamic modeling method, the EV load distribution with multiple scales is obtained. The three-dimensional distribution in both time and space shows the charging peaks in 10:00 and around 22:00, corresponding to the travel patterns of users. In comparison to the static modeling method, the dynamic modeling approach yields a more realistic load curve with stronger volatility, thereby better representing the load distribution uncertainty.
3 Orderly Charging Strategy Considering Spatiotemporal Characteristics
3.1 Regulation Potential Analysis
When an electric vehicle is plugged in for charging, there is a window between the initial connection time \(t_{in}\) and the time the user actually wants to leave \(t_{out}\). A charging task can be finished by the departure time. The available regulation potential for a single EV can be described as
$$P_{k,t}^{c}=\begin{cases}
P_{k}^{ev}, & t \in [t_{in}, t_{end}) \\[4pt]
0, & t \notin [t_{in}, t_{end})
\end{cases}$$
where \(t_{end}\) is the actual ending time of charging and \(t_g = t_{end} – t_{in}\) is the dispatchable time period. For the cluster of electric vehicles, the total regulation potential is obtained by summation over all connected vehicles:
$$P_t^R = \sum_{k=1}^{n} P_{k,t}^{c}.$$
3.2 Optimization Model Based on Spatiotemporal Characteristics
I constructed a multi-objective optimization framework that simultaneously considers the grid side, the user side, and the spatial equilibrium of charging loads. The objectives introduced in this chapter are defined as follows:
(1) Grid load peak-valley difference
$$f_1 = \min \left\{\max_t \sum_{i=1}^{n} P_{i}^t – \min_t \sum_{i=1}^{n} P_i^t \right\},$$
$$P_i^t = P_{b,i}^t + P_{ev,i}^t,$$
Where \(P_{i}^{t}\) is the total load at node \(i\) and time \(t\), \(P_{b,i}^{t}\) is the base load, and \(P_{ev,i}^{t}\) is the electric vehicle charging load.
(2) User charging cost
$$f_2 = \min \left\{\sum_{t=1}^{N_t} \sum_{k=1}^{K} \delta_k^t \cdot \zeta_k^t \cdot p_{k,t} \cdot \Delta t \right\},$$
where \(\delta_k^t\) is the charging status (1 for charging, 0 otherwise), \(\zeta_k^t\) is the electricity price, \(p_{k,t}\) is the charging power output, and \(\Delta t\) is the time slot.
(3) Spatial equilibrium index of EV loads
$$f_3 = \max \left\{ \left(1 – \frac{\sum_{s=1}^{A} P_{ev}^{d,s}}{\sum_{s=1}^{A} P_{ev}^{o,s}} \right) \times 100\% \right\},$$
where \(P_{ev}^{d,s}\) is the difference between the maximum and minimum EV load of region \(s\) after optimization, and \(P_{ev}^{o,s}\) is the difference before optimization. By combining the load in multiple regions, I define
$$P_{ev}^{d,s} = \max_t \left( \sum_{i=1}^{n} P_{ev,i}^{d,s}(t) \right) – \min_t \left( \sum_{i=1}^{n} P_{ev,i}^{d,s}(t) \right),$$
$$P_{ev}^{o,s} = \max_t \left( \sum_{i=1}^{n} P_{ev,i}^{o,s}(t) \right) – \min_t \left( \sum_{i=1}^{n} P_{ev,i}^{o,s}(t) \right).$$
(4) Constraints
The optimization considers the following constraints:
| Type | Formula | Description |
|---|---|---|
| Safety | \(SOC_{min} \leq SOC_{k,t} \leq SOC_{max}\) | state of charge limits |
| User demand | \(SOC_{end} \geq SOC_f\) | minimum required SOC at departure |
| Time | \(t_{in} \leq t_{start} \leq t_{end} \leq t_{out}\) | time boundaries |
| Regulation | \(t_c \leq t_{out} – t_{in}\) | charging duration boundary |
| Regulation start | \(t_{go} \leq t_{in} + t_g\) | start time after connection |
| Transformer | \(P_i^t < S_T\) | capacity constraint |
3.3 IMPSO-based Solution
The basic particle swarm optimizer updates the position and velocity of each particle using
$$v_{in}^{h+1} = \omega \cdot v_{in}^h + c_1 r_1 (x_{bin}^h – x_{in}^h) + c_2 r_2 (y_{bn}^h – x_{in}^h),$$
$$x_{in}^{h+1} = x_{in}^h + v_{in}^{h+1}.$$
However, the conventional PSO has a weakness of easily falling into a local optimum. To address this, I propose an improved particle swarm optimizer (IMPSO). Three major improvements are applied:
(1) Logistic chaotic map for population initialization
Rather than using the random uniform distribution to initialize particle positions, the Logistic map generates an initial population more evenly over the search space:
$$x_{m,N}^0 = \begin{cases}
rand, & m=1 \\[4pt]
\lambda \cdot x_{m-1,N} (1 – x_{m-1,N}), & m=2,\ldots,M
\end{cases}$$
where \(\lambda\) is a control parameter. Transformation to the desired domain is performed by
$$X_{m,N} = low\_bound + (up\_bound – low\_bound) \cdot x_{m,N}^0.$$
(2) Dynamic adaptive inertia weight and learning factors
The inertia weight and learning factors are dynamically adjusted according to the iteration number h:
$$\omega_h = \omega_{max} – (\omega_{max} – \omega_{min}) \cdot \frac{h}{h_{max}},$$
$$c_{1,h} = c_{1,max} – (c_{1,max} – c_{1,min}) \cdot \frac{h}{h_{max}},$$
$$c_{2,h} = c_{2,min} + (c_{2,max} – c_{2,min}) \cdot \frac{h}{h_{max}}.$$
(3) Fitness function
Because the model has three objectives, the optimization problem is transformed to a single-objective problem using the weighted-sum method:
$$fitness = \omega_1 f_1 + \omega_2 f_2 – \omega_3 f_3,$$
where \(\omega_1, \omega_2, \omega_3\) are weighting factors satisfying \(\omega_1+\omega_2+\omega_3=1\). In the optimization, the negative sign before \(\omega_3 f_3\) is used to maximize the spatial equilibrium.
3.4 Simulation Results and Analysis
Time-of-use tariffs applied in the simulation are listed in the following table.
| Period | Price level | Price (yuan/kWh) |
|---|---|---|
| 0:00-7:00 | peak | 1.2 |
| 7:00-11:00 | valley | 0.34 |
| 11:00-16:00 | flat | 0.8 |
| 16:00-20:00 | valley | 0.34 |
| 20:00-24:00 | flat | 0.8 |
I tested five schemes with different objective combinations. Scheme 1 optimizes only the peak-valley difference, Scheme 2 optimizes only the user charging cost, Scheme 3 optimizes only the EV spatial equilibrium, Scheme 4 optimizes both the peak-valley difference and charging cost, and Scheme 5 jointly optimizes peak-valley difference, charging cost, and spatial equilibrium.
| Case | Peak-valley (kW) | Charging cost (yuan) | EV equilibrium |
|---|---|---|---|
| Uncoordinated | 2640.35 | 51691.82 | 0 |
| Scheme 1 (IMPSO) | 1757.66 | 46528.26 | 17.09% |
| Scheme 2 (IMPSO) | 2115.02 | 45229.56 | 8.68% |
| Scheme 3 (IMPSO) | 2045.93 | 46945.50 | 29.48% |
| Scheme 4 (IMPSO) | 1789.15 | 45749.10 | 19.40% |
| Scheme 5 (IMPSO) | 1888.81 | 46359.56 | 29.99% |
The results confirm that the proposed strategy with all three objectives (Scheme 5) achieves the highest EV equilibrium indicator, 29.99% compared to 19.40% for Scheme 4 with PSO. It also consistently yields lower values in every objective when compared to standard PSO. The spatiotemporal distributions after optimization demonstrate that the proposed strategy reduces charging peaks and shifts charging demand to valley periods, greatly lowering the overall impact on the distribution grid.
4 Orderly Charging Strategy Considering Functional Zone Requirements
4.1 Functional Zone Load Characteristics and Regulation Demand
To refine the regulation strategy, I analyzed the functional zone characteristics of residential areas (H), workplace areas (W), and commercial areas (S). The charging duration distributions in three functional zones are summarized as follows:
| Functional Zone | Dominant duration range | Charging Characteristics |
|---|---|---|
| Residential (H) | 3-6 hours | Long parking interval, slow charging, users charge overnight |
| Workplace (W) | 1-2 hours | Short-term charging while at work, fast charging, spaced over work hours |
| Commercial (S) | 1-1.5 hours | Very short parking, shortest charging time, highly volatile load |
The load profiles in the different zones show distinct patterns. For the residential zone, the load is relatively flat with a peak around 22:00, because users return home and charge overnight. For the workplace and commercial zones, the loads are more volatile and closely follow operation hours, resulting in sharper peaks. Since the demand for these functional zones is different, the regulation requirements need to be differentiated: the residential zone requires load shifting toward the valley, whereas the work and commercial zones require peak limitation.
4.2 Area-interest Coupling Strategy and Optimization Model
Based on the analysis, I developed a dual-dimensional coupling strategy that jointly considers the interests of both the grid and users with the functional zone needs. The optimization objectives are:
| Stakeholder | Residential Area (H) | Work & Commercial Area (WS) |
|---|---|---|
| Grid side | Total peak-valley difference | Total peak-valley difference; regional EV peak |
| User side | Charging cost minimization | User satisfaction maximization |
The formulas for the additional three objective functions are given below.
The peak value of work and commercial areas is:
$$f_4 = \min \left\{\max_t P_{ws}^t \right\}.$$
The user satisfaction of work and commercial areas is:
$$f_5 = \max \left\{ 1 – \frac{\sum_{t=1}^{T} |P_{o,ws}^t – P_{d,ws}^t|}{\sum_{t=1}^{T} P_{d,ws}^t} \right\},$$
where \(P_{d,ws}^t\) denotes the uncontrolled charging power in work/commercial areas at time \(t\), and \(P_{o,ws}^t\) is the optimized value.
The charging cost for the residential area is:
$$f_6 = \min \left\{\sum_{t=1}^{N_t} \sum_{k=1}^{K_H} \delta_k^t \cdot \zeta_k^t \cdot p_{k,t} \cdot \Delta t \right\},$$
where \(K_H\) is the number of vehicles in the residential area.
In addition to the constraints described in Chapter 3, I added a constraint on user satisfaction:
$$f_6 \geq Y_m,$$
which represents the minimum acceptable user satisfaction, and the peak constraint:
$$P_{max} \leq P_{dis}^{max},$$
where \(P_{dis}^{max}\) is the uncontrolled maximum load and \(P_{max}\) is the optimized one.
4.3 GA-NSGA-II Algorithm
The multi-objective problem is solved using the NSGA-II algorithm enhanced with a good point set and adaptive probabilities. The good point set generates a uniformly distributed initial population:
$$r_j = \text{mod}(2\cos(2\pi p/7)\cdot j, 1), \quad 1 \leq j \leq n,$$
$$P_{n}(i) = \left\{ \left\{r_1 \cdot i\right\}, \left\{r_2 \cdot i\right\}, \ldots, \left\{r_n \cdot i\right\} \right\}, \quad i=1,2,…,m,$$
$$X_j^i = a_j + P_n(i) \cdot (b_j – a_j),$$
where \(a_j\) and \(b_j\) are the lower and upper bounds of the dimension \(j\). Adaptive crossover and mutation probabilities are
$$p_c = p_{c,max} – (p_{c,max} – p_{c,min}) \cdot \frac{h}{h_{max}},$$
$$p_m = p_{m,max} – (p_{m,max} – p_{m,min}) \cdot \frac{h}{h_{max}}.$$
The fitness function is designed to use the normalized objectives:
$$F = \min \left[\gamma_1 \left( \frac{f_1′}{f_1^{d’}} \right) + \gamma_2 \left( 1 – \frac{f_6′}{f_6^{d’}} \right) \right], \quad f_i’ = \frac{f_i}{f_i^d},$$
where \(f_i^d\) denotes the value of the objective in the uncontrolled scenario.
4.4 Simulation Results
I compared Scheme 1 that only optimizes the total grid peak-valley and total charging costs using NSGA-II, with Scheme 2 adopting the proposed functional zone aware strategy solved by GA-NSGA-II. The results are given below.
| Case | Grid peak-valley (kW) | WS peak (kW) | H charging cost (yuan) | WS satisfaction |
|---|---|---|---|---|
| Uncoordinated | 2640.35 | 4240 | 24793.02 | 100% |
| Scheme 1 | 1931.20 | 3920 | 22365.42 | 73.06% |
| Scheme 2 | 1933.77 | 3500 | 22232.28 | 79.20% |
The results confirm that the proposed area-interest coupling strategy has a similar peak-shaving rate (roughly 26.8%) as the baseline method but achieves a substantially lower regional EV peak, with the WS peak reduction from 4240 kW to 3500 kW, a 17.45% reduction compared to 7.55% with Scheme 1. Furthermore, the proposed scheme better preserves user satisfaction in work and commercial areas (79.20% versus 73.06%) while also reducing the residents’ charging cost. The obtained Pareto front shows that GA-NSGA-II provides a more uniformly distributed and better solution set than NSGA-II.
5 Visualization Platform for Electric Vehicle Load Regulation Based on LabVIEW
5.1 Demand Analysis
The visualization platform is designed to serve both grid dispatchers and electric vehicle users. The core functionalities include data processing from the database, load simulation for scenarios when real data are limited, regulation strategy management, and a friendly human-machine interface. Realizing these functions provides an intuitive and efficient way for the dispatchers to monitor and manage the load distribution and adjust the control strategies promptly.
5.2 Framework and Module Design
I structured the platform into four modules: user login, data processing, load simulation and regulation management, and human-computer interaction. The user login module performs identity verification. The data information module imports the needed data. The load simulation module simulates the uncontrolled charging load via the method described in Chapter 2. The regulation management module implements the control strategies in Chapters 3 and 4. The human-machine interaction module visualizes the results and supports output.
| Module | Purpose | Main functions |
|---|---|---|
| Login | Safety | account and password validation |
| Data | Data foundation | import, collection, storage |
| Load simulation | Forecasting | uncoordinated charging load simulation |
| Regulation management | Control | strategy selection, optimization and result output |
| Human-computer interaction | Display | chart plot, parameter adjustment and data export |
5.3 Implementation and Results
The software is built on the LabVIEW environment running on a standard PC. In the graphical programming interface, separate virtual instruments are created for the login part and the main control panel. The main panel enables the operator to import the simulation parameters for the EV load model, run the simulation algorithm, and view the temporal load curve and spatial load distribution curves after charging regulation.
In the login module, invalid credentials produce a pop-up warning and the user cannot access the main interface. In the load simulation module, the simulation results are shown as load duration curves, while the regulation module allows the user to select the strategy from two preset options: the spatiotemporal strategy in Chapter 3 (Strategy 1) and the functional area strategy in Chapter 4 (Strategy 2). After the user clicks the regulation button, the optimized results are displayed. The platform provides an intuitive way for the dispatchers to monitor the pre- and post-regulation curves and to adjust the control actions at a glance, which greatly improves the usability and credibility of the control strategies discussed in this thesis.
6 Conclusions and Outlook
In this thesis, I have studied the spatiotemporal modeling of electric vehicle charging load and the corresponding orderly charging strategies. The key conclusions drawn from this research are described below:
(1) A dynamic spatiotemporal load model of electric vehicles considering typical user travel demands is established. I analyzed the factors that influence the charging-load distribution and subsequently classified the vehicles into groups based on the trip purposes. I developed a dynamic fuzzy-based energy consumption model and a travel-demand-related road impedance model. Using the Monte Carlo method, I then computed the charging-load distribution of the entire fleet. The numerical experiments showed that this method can appropriately characterize the uncertainties in time and space and provide the basis for further regulation strategies.
(2) A spatiotemporal load regulation strategy for electric vehicles is developed. I first quantified the dispatchable capacity of the electric vehicles. Afterward, I formulated an optimization model that simultaneously minimizes the grid peak-valley difference and user charging cost while maximizing the spatial equilibrium of the electric vehicle loads. To solve this problem, I improved the particle swarm optimizer by chaotic initialization and adaptive parameter tuning. Simulation results confirm that the strategy can reduce the grid peak to valley difference and users’ cost while effectively increasing the load equilibrium.
(3) A functional area-aware load regulation strategy is proposed. Because the load characteristics in residential, workplace, and commercial zones differ substantially, I introduced differentiated regulation demands and formulated an area-interest two-dimensional coupled regulation strategy. The multi-objective model was solved by a GA-NSGA-II algorithm. The strategy obtains better trade-offs between grid-side demands, user-side satisfaction, and regional peak suppression when compared to conventional strategies.
(4) A LabVIEW-based visualization platform for the regulation management of electric vehicles is developed. The platform integrates the user login module, load simulation module, and the regulation management module. By embedding the introduced load modeling and regulation algorithms, the platform provides intuitive displays of load distributions and controllable strategy results, enhancing the capability for intelligent and efficient management.
There are several directions that remain to be further investigated based on the current work. First, the current load modeling only focuses on private electric vehicles with different trip purposes. Public transportation such as buses and official vehicles should also be taken into account in future work so that the load distribution and the regulation problem could be studied under a comprehensive multi-vehicle-type framework.
Second, the electricity price used in the regulation strategy is a static time-of-use price. More complex pricing mechanisms, such as dynamic real-time pricing, could be developed to improve the responsiveness and economic efficiency of electric vehicle charging regulation.
Third, the visualization platform currently concentrates mainly on the visualization of data and control outcomes. The interactivity between the human and the platform requires more effort, and rich interaction approaches could be introduced to increase the convenience for the dispatchers and thereby achieve a higher level of smart management.