As a researcher addressing the dual challenges of carbon neutrality and renewable energy integration, I propose a novel two-stage day-ahead-intraday optimal dispatch framework for power systems incorporating electric vehicles (EVs). This methodology leverages EVs as flexible resources to balance supply-demand uncertainties while minimizing carbon emissions and operational costs.
1. Uncertainty Mitigation via SSA-CNN-LSTM Forecasting
Electric vehicle charging patterns exacerbate the volatility of modern grids with high renewable penetration. My solution employs a Sparrow Search Algorithm (SSA)-optimized Convolutional Long Short-Term Memory (CNN-LSTM) network for high-accuracy source-load forecasting:
Forecasting Workflow:
- Data Preprocessing: Normalize historical PV, wind, and load data.
- SSA Parameter Optimization:
- Population size: 50, iterations: 10, safety threshold: 0.8.
- Fitness function: CNN-LSTM neuron count & initial learning rate.
- Feature Extraction: CNN layers extract spatial features; LSTM processes temporal sequences.
Mathematical Formulation:
- SSA Position Update:Xijd+1={Xijd⋅exp(−dα⋅itermax)R2<STXijd+Q⋅LR2≥STXijd+1={Xijd⋅exp(−α⋅itermaxd)Xijd+Q⋅LR2<STR2≥STwhere R2R2 = alert threshold, QQ = random normal number, LL = unit matrix.
- Forecast Evaluation Metrics:MAPE=1n∑i=1n∣Li−Li′Li∣×100%,RMSE=1n∑i=1n(Li−Li′)2MAPE=n1i=1∑nLiLi−Li′×100%,RMSE=n1i=1∑n(Li−Li′)2
Table 1: Forecasting Performance Comparison
| Model | MAPE (%) | RMSE (MW) |
|---|---|---|
| BP Neural Network | 5.27 | 0.73 |
| CNN-LSTM | 4.22 | 0.70 |
| SSA-CNN-LSTM (Proposed) | 2.31 | 0.37 |
The SSA-CNN-LSTM reduces MAPE by 60% and RMSE by 49% versus CNN-LSTM, critically enhancing dispatch reliability amid electric vehicle variability.
2. EV Charging Mode Classification
Electric vehicles are categorized by demand response (DR) flexibility to maximize grid support:
A. Non-Adjustable Mode (50% of fleet):
- Constraints:Pk,t=PchEV,Sk,t=Sk,t−Δt+Ich,k,tEVηchPk,tEVΔtSkPk,t=PchEV,Sk,t=Sk,t−Δt+Ich,k,tEVSkηchPk,tEVΔt
- Applies to taxis/short-trip EVs requiring fixed charging.
B. Power-Adjustable Mode (35%):
- Constraints:0≤Pk,t≤PchEV,0.95Skmax≤Sk,out≤Skmax0≤Pk,t≤PchEV,0.95Skmax≤Sk,out≤Skmax
- Minimizes costs via price-based DR (PDR).
C. V2G Mode (15%):
- Constraints:Sk,t=Sk,t−Δt+Ich,k,tEVηchPk,tEVΔtSk−Idis,k,tEVPk,tEVΔtηdisSkSk,t=Sk,t−Δt+Ich,k,tEVSkηchPk,tEVΔt−Idis,k,tEVηdisSkPk,tEVΔt
- Engages in incentive-based DR (IDR) for bidirectional energy trading.
DR Participation Limits:
- Day-ahead: ≤30% (Mode B), ≤25% (Mode C).
- Intraday: ≤20% (Mode B/C).
3. Two-Stage Dispatch Framework
3.1 Day-Ahead Scheduling
Objective Functions:
- Economic:minF1da=∑t=1T1[∑i=1N(aiPGi,t2+biPGi,t+ci)+Ct+KW(PW,tpre−PW,t)+KR(PR,tpre−PR,t)]minF1da=t=1∑T1[i=1∑N(aiPGi,t2+biPGi,t+ci)+Ct+KW(PW,tpre−PW,t)+KR(PR,tpre−PR,t)]
- Environmental:minF2da=∑t=1T1∑i=1N(αiPGi,t2+βiPGi,t+γi)minF2da=t=1∑T1i=1∑N(αiPGi,t2+βiPGi,t+γi)
- Weighted Total:Fda=ω1F1da+ω2F2da(ω1=ω2=0.5)Fda=ω1F1da+ω2F2da(ω1=ω2=0.5)
Constraints:
- Power balance, ramp rates, reserves (Eq. 19–26 in source).
3.2 Intraday Rolling Optimization
Objective Functions:
- Adjusts day-ahead plan using 15-min intervals:minFid=ω3F1id+ω4F2id(ω3=ω4=0.5)minFid=ω3F1id+ω4F2id(ω3=ω4=0.5)
- Updates forecasts and EV responses every 4 hours.
4. Staircase Carbon Trading Mechanism
Carbon costs escalate with emissions to incentivize low-carbon operation:Ct={Kc(Et−Eqt)Et≤EqtKcv+(1+σ)Kc(Et−Eqt−v)Eqt+v<Et≤Eqt+2v⋮(6+4σ)Kcv+(1+4σ)Kc(Et−Eqt−4v)Et>Eqt+4vCt=⎩⎨⎧Kc(Et−Eqt)Kcv+(1+σ)Kc(Et−Eqt−v)⋮(6+4σ)Kcv+(1+4σ)Kc(Et−Eqt−4v)Et≤EqtEqt+v<Et≤Eqt+2vEt>Eqt+4v
where Et=∑δiPGi,tEt=∑δiPGi,t, v=λEqtv=λEqt, σσ = price increment.
5. Improved MOGWO Solver
Enhanced Multi-Objective Grey Wolf Optimizer addresses local optima traps:
- Exponential Decay:a=2−(kitermax)3×2a=2−(itermaxk)3×2
- Elite Retention: Replace poorest 18% wolves per iteration.
Table 2: Algorithm Performance Comparison
| Solution Type | Traditional MOGWO | Improved MOGWO |
|---|---|---|
| Economic Optimal Cost ($×10^3$) | 553.23 | 551.29 |
| Emission Optimal (lb) | 6,844.5 | 6,835.6 |
| Compromise Emission (lb) | 6,952.7 | 6,937.1 |
6. Case Study & Results
Simulations used 20 MW PV, 30 MW wind, 20,000 EVs with parameters:
- Daily mileage: N(3.2,0.882)N(3.2,0.882) km.
- Charging times: Start ~ N(17.6,3.42)N(17.6,3.42), End ~ N(8.92,3.242)N(8.92,3.242).
Table 3: Day-Ahead Dispatch Results (4 Scenarios)
| Metric | Scenario 1 | Scenario 4 | Reduction |
|---|---|---|---|
| Total Cost ($) | 64,434 | 58,929 | 10.3% |
| Emissions (lb) | 7,932 | 7,126 | 10.9% |
| PV Curtailment Rate (%) | 6.13 | 2.84 | 3.29% |
| Wind Curtailment Rate (%) | 6.38 | 0.37 | 6.01% |
Key Findings:
- Scenario 4 (EV-DR + carbon trading) outperforms others:
- Intraday costs reduced by 10.3% vs Scenario 1.
- Emissions decreased 10.9% while boosting renewable consumption 4.2%.
- EV Flexibility: Mode C EVs provided 187 lb emission reduction via V2G.
7. Conclusion
This work establishes an integrated framework for electric vehicle participation in low-carbon power systems:
- SSA-CNN-LSTM elevates forecasting accuracy (MAPE: 2.31%), mitigating source-load uncertainties.
- EV Classification enables tailored DR strategies, reducing peak loads and emissions.
- Staircase Carbon Pricing cuts coal dependency, lowering emissions 10.9%.
- Improved MOGWO achieves superior convergence and solution diversity.
The proposed system cuts operational costs by 10.3% and emissions by 10.9%, proving electric vehicles are pivotal for grid decarbonization. Future work will extend this framework to real-time scheduling scales.