The rapid advancement of electric vehicle (EV) technologies in China has positioned EVs as pivotal flexible loads in modern power systems. Integrating EVs into grid operations can significantly enhance renewable energy absorption and support the construction of a sustainable power infrastructure. However, the inherent uncertainties in renewable generation and EV charging behaviors pose challenges to system stability and efficiency. This paper proposes a comprehensive day-ahead and intraday low-carbon optimal dispatch framework that leverages the demand response (DR) characteristics of electric vehicles. By employing a Sparrow Search Algorithm-optimized Convolutional Long Short-Term Memory (SSA-CNN-LSTM) neural network for source-load forecasting, the model mitigates uncertainties and improves scheduling accuracy. Furthermore, EVs are categorized into three charging modes based on their DR participation, and a two-stage optimization model is developed to minimize system costs and pollutant emissions under a ladder-based carbon trading mechanism. The improved Multi-Objective Grey Wolf Optimizer (MOGWO) is utilized to solve the complex, multi-objective problem. Case studies demonstrate that this approach reduces total costs by 10.3%, cuts emissions by 10.9%, and increases renewable energy utilization by 4.2% compared to conventional methods.
The transition to a low-carbon energy system in China necessitates the integration of renewable sources like wind and solar, but their intermittency complicates grid management. Electric vehicles, as distributed energy resources, offer a viable solution through demand response programs. This study addresses the dual challenges of renewable variability and EV charging randomness by developing a predictive and adaptive dispatch strategy. The SSA-CNN-LSTM model enhances forecasting precision for wind, solar, and load profiles, enabling more reliable scheduling. EVs are classified into uncontrollable (Mode A), power-adjustable (Mode B), and bidirectional V2G-capable (Mode C) categories, each contributing uniquely to DR. The optimization model incorporates ladder carbon pricing to incentivize low-carbon operations while balancing economic and environmental objectives. Simulation results validate the efficacy of this approach in promoting grid stability and sustainability.
The SSA-CNN-LSTM forecasting model combines the feature extraction capabilities of convolutional neural networks (CNN) with the temporal processing strengths of long short-term memory (LSTM) networks. The Sparrow Search Algorithm (SSA) optimizes key parameters, such as neuron count and learning rate, to improve prediction accuracy. The model processes historical data on wind speed, solar irradiance, and load demand, normalized to eliminate outliers. The SSA initialization involves a population matrix $X$:
$$X = \begin{bmatrix}
x_{1,1} & x_{1,2} & \cdots & x_{1,m} \\
x_{2,1} & x_{2,2} & \cdots & x_{2,m} \\
\vdots & \vdots & \ddots & \vdots \\
x_{n,1} & x_{n,2} & \cdots & x_{n,m}
\end{bmatrix}$$
where $n$ is the number of sparrows and $m$ is the problem dimension. The fitness function $F_X$ evaluates predictions, and position updates for discoverers, joiners, and vigilant sparrows are governed by:
$$X_{i,j}^{d+1} = \begin{cases}
X_{i,j}^d \cdot \exp\left(-\frac{i}{\alpha \cdot iter_{\text{max}}}\right) & \text{if } R_2 < ST \\
X_{i,j}^d + Q \cdot L & \text{if } R_2 \geq ST
\end{cases}$$
Here, $d$ is the iteration index, $\alpha$ is a random number in (0,1], $iter_{\text{max}}$ is the maximum iterations, $Q$ is a random number from a normal distribution, and $L$ is a unit matrix. This optimization ensures minimal forecasting errors, with metrics like Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) calculated as:
$$E_{\text{MAPE}} = \frac{1}{n} \sum_{i=1}^n \left| \frac{L_i – \hat{L}_i}{L_i} \right| \times 100\%$$
$$E_{\text{RMSE}} = \sqrt{\frac{1}{n} \sum_{i=1}^n (L_i – \hat{L}_i)^2}$$
In comparative tests, the SSA-CNN-LSTM model achieved a MAPE of 2.31% and RMSE of 0.37 MW, outperforming other models like BP and Elman neural networks.
Electric vehicles are categorized into three modes to reflect their DR potential. Mode A represents EVs with fixed charging power and no DR participation, suitable for taxis or short-trip private cars. The constraints include:
$$I_{\text{ch},n,t}^{\text{EV}} = 1 \quad \forall t \in [t_{k,\text{in}}, t_{k,\text{out}}]$$
$$P_{k,t} = P_{\text{ch}}^{\text{EV}}$$
$$S_{k,t} = S_{k,t-\Delta t} + I_{\text{ch},k,t}^{\text{EV}} \eta_{\text{ch}} \frac{P_{k,t}^{\text{EV}} \Delta t}{S_k}$$
Mode B allows power adjustment without V2G, ideal for cost-conscious users. Key constraints are:
$$0 \leq P_{k,t} \leq P_{\text{ch}}^{\text{EV}}$$
$$0.95 S_k^{\text{max}} \leq S_{k,\text{out}} \leq S_k^{\text{max}}$$
$$\sum_{t=t_{k,\text{in}}}^{t_{k,\text{out}}} \sum_{k=1}^K P_{k,t} \approx P_{\text{total}}^{\text{EV}}$$
Mode C supports bidirectional V2G for maximum economic benefit, with constraints:
$$I_{\text{ch},n,t}^{\text{EV}} + I_{\text{dis},n,t}^{\text{EV}} = 1$$
$$-P_{\text{ch}}^{\text{EV}} \leq P_{k,t} \leq P_{\text{ch}}^{\text{EV}}$$
$$S_{k,\text{thr}} \leq S_{k,t} \quad \text{when } P_{k,t} < 0$$
The ladder carbon trading model imposes costs based on emission tiers. The free allocation $E_{qt}$ and transaction cost $C_t$ are defined as:
$$E_{qt} = \eta P_{Lt}$$
$$C_t = \begin{cases}
K_c (E_r – E_{qt}) & E_r \leq E_{qt} \\
K_c (E_r – E_{qt}) & E_{qt} < E_r \leq E_{qt} + v \\
K_c v + (1+\sigma) K_c (E_r – E_{qt} – v) & E_{qt} + v < E_r \leq E_{qt} + 2v \\
\vdots
\end{cases}$$
where $E_r = \sum_{i=1}^N \delta_i P_{Gi,t}$, $v = \lambda E_{qt}$, and $F_T = \sum_{t=1}^T C_t$.
The day-ahead scheduling minimizes the total cost $F^{\text{ad}}$ and emissions $F_2^{\text{ad}}$ over a 24-hour horizon:
$$F_1^{\text{ad}} = \sum_{t=1}^{T_1} \left[ \sum_{i=1}^N U_{i,t} (a_i P_{Gi,t}^2 + b_i P_{Gi,t} + c_i) + \text{start-up costs} + C_t + K_W (P_{w,t}^{\text{pre,1}} – P_{w,t}) + K_R (P_{R,t}^{\text{pre,1}} – P_{R,t}) \right]$$
$$F_2^{\text{ad}} = \sum_{t=1}^{T_1} \sum_{i=1}^N [\alpha_i P_{Gi,t}^2 + \beta_i P_{Gi,t} + \gamma_i + \zeta_i \exp(\lambda_i P_{Gi,t})]$$
The combined objective is $F^{\text{ad}} = W_1 F_1^{\text{ad}} + W_2 F_2^{\text{ad}}$, with $W_1 + W_2 = 1$. Constraints include power balance, generator limits, and ramping rates.
Intraday scheduling refines the day-ahead plan using 15-minute intervals over a 4-hour rolling window. The objective functions are similar but exclude start-up costs and incorporate intraday DR costs. The improved MOGWO algorithm enhances global search and convergence by modifying the parameter $a$ and employing a survival-of-the-fittest strategy:
$$a = 2 – \left( \frac{k}{k_{\text{max}}} \right)^3 \times 2$$
$$m = [n \cdot (0.618 \cdot \eta), n \cdot \eta]$$
Case studies evaluated four scenarios: (1) no DR or carbon trading, (2) carbon trading only, (3) DR only, and (4) combined DR and carbon trading. Key results are summarized below:
| Scenario | Total Cost ($) | Pollutant Emissions (lb) | Wind Curtailment Rate (%) | Solar Curtailment Rate (%) |
|---|---|---|---|---|
| 1 | 64,434 | 7,932 | 6.38 | 6.13 |
| 2 | 62,850 | 7,558 | 4.24 | 3.75 |
| 3 | 60,786 | 7,292 | 1.82 | 3.13 |
| 4 | 58,929 | 7,126 | 0.37 | 2.84 |
The integration of electric vehicles in China’s EV market demonstrates significant benefits. Scenario 4, with full DR and carbon trading, reduced costs by 10.3% and emissions by 10.9% compared to Scenario 1. The improved MOGWO algorithm provided better Pareto fronts, indicating superior multi-objective optimization. For instance, the compromise solution for Scenario 4 had emissions of 6,937 lb and costs of $56,454, outperforming traditional MOGWO.
In conclusion, the proposed framework effectively harnesses the flexibility of electric vehicles to enhance renewable energy integration and reduce carbon footprints. The SSA-CNN-LSTM model ensures accurate forecasting, while the EV classification and ladder carbon trading mechanism optimize economic and environmental outcomes. This approach supports the development of resilient and sustainable power systems, aligning with global decarbonization goals. Future work could extend to real-time scheduling and multi-scale optimization to address dynamic uncertainties further.