Active Distribution Grid Hits New Milestone with EV-Integrated Two-Stage Dispatch Model
In the ever-evolving landscape of power systems, where volatility meets ambition daily, a breakthrough quietly unfolded in northeastern China—not on a racetrack or in a gleaming Tesla Gigafactory, but inside a simulation chamber at Northeast Petroleum University. There, researchers engineered a dispatch framework for active distribution networks (ADNs) that doesn’t just tolerate electric vehicles (EVs); it actively partners with them.
Welcome to the era of cooperative grid intelligence—where EVs aren’t just passive loads drawing juice from the wall, but mobile, flexible energy assets that help smooth out the chaos of renewable generation. And in this emerging paradigm, the humble EV battery may be the unsung hero stabilizing tomorrow’s grid—more so than any static inverter or capacitor bank.
The latest work by Jia Ying, Liu Hanli, Zhao Shuqi, Li Yang, Han Pengfei, and Chen Biao reveals a carefully choreographed two-stage optimization process that marries day-ahead economic planning with intra-day voltage and stability refinement—all while keeping EV clusters firmly in the loop. It’s more than an algorithmic exercise. It’s a glimpse into how grids can evolve from rigid, top-down infrastructures to dynamic, responsive ecosystems that anticipate rather than react.
Let’s take a ride through what makes this model special—and why it matters far beyond the lab.
The Grid Is No Longer Passive—It’s Active
Historically, electricity distribution operated on a “build it and they will come” mentality. Power flowed one way: from large centralized plants, through transformers and feeders, into homes and factories. Demand fluctuated, yes—but the grid had inertia, and operators relied on spinning reserves, load shedding, and conservative safety margins.
Then came solar panels on rooftops, wind turbines dotting the plains, and—critically—millions of EVs plugging in after work. Suddenly, the load curve wasn’t just lumpy; it was spiky, unpredictable, and bidirectional. Distributed energy resources (DERs) flipped the old model on its head. The result? Voltage swings, reverse power flow, and curtailment headaches—especially during midday solar peaks, when excess generation couldn’t be absorbed locally and had to be dumped or throttled.
Enter the active distribution network—a concept first formalized by CIGRE’s working group over a decade ago but only now maturing into practice. Unlike traditional grids, ADNs actively manage both supply and demand. They coordinate controllable generation (like dispatchable DG units), flexible storage (stationary batteries), and demand-side assets—chief among them, EV aggregators.
But coordination is easier said than done. A real-world ADN must juggle three competing priorities: economy (keep costs low), reliability (avoid blackouts), and sustainability (maximize renewables). And crucially, it must do so across multiple time horizons—from 24-hour forecasts down to 15-minute real-time corrections.
That’s where the two-stage dispatch model shines.
Stage One: The 24-Hour Chess Match
Imagine planning a cross-country road trip. You’d map out the major stops, charging windows, and overnight rests based on forecasted traffic and weather. That’s what the day-ahead stage does for the grid—only the stakes are higher, and the terrain is electric.
Using real-world wind and solar data from a hybrid power plant in Hami, Xinjiang—a region known for its intense solar irradiance and steady winds—the team fed historical generation profiles into a long short-term memory (LSTM) neural network. The output? A 24-hour forecast of renewable output at four key nodes: PV farms at nodes 24 and 32, and wind turbines at 11 and 17.
With this forecast in hand, the first optimization kicked in: minimize total operational cost, a composite objective covering:
- Power purchase/sale with the main grid (under time-of-use pricing: cheap at night, expensive during 8–13 and 15–21),
- Network loss penalties,
- Maintenance costs for DG and storage,
- And—critically—the opportunity cost of curtailed renewables.
Yes: instead of treating curtailment as a mere technical loss, the model priced it explicitly using real-time electricity rates. If you dump 100 kWh of solar at 18:00—a peak-price hour—you’re not just wasting clean energy; you’re forfeiting ¥1.20 or more in potential revenue. Translate that into a cost term, and suddenly not using renewables becomes economically painful.
Then came the game-changer: adding controllable loads—specifically, two energy storage systems (190 kWh at node 10 and 160 kWh at node 28) and an EV aggregator cluster (100 kWh capacity) at node 7.
The impact? Immediate and quantifiable.
- Renewable curtailment dropped by 4.41 percentage points—from a 93.77% utilization rate to 98.18%.
- Net power procurement cost fell from ¥472.03 to ¥446.90 per day—a 5.3% saving, just by shifting when and how storage charged.
- At 15:00—the worst hour for curtailment pre-optimization—the utilization rate jumped from 78.91% to 88.41%.
How? By time-shifting energy. During midday solar peaks (12–16 h), the grid didn’t just export to the main network (hitting the sale limit); it also charged the EVs and stationary batteries. Then, during evening ramps or nighttime lulls, those same assets discharged—filling gaps, reducing imports, and smoothing the net load profile.
Think of it as energy arbitrage, but with physics and constraints. EVs weren’t being inconvenienced; they were following charging schedules aligned with grid needs—likely via smart tariffs or aggregator contracts. One EV plugged in at 14:00 might charge at 3 kW for two hours, pause, then top off at 22:00 when wind picks up and prices dip.
This is demand flexibility as infrastructure.
Stage Two: The Real-Time Tightrope Walk
Day-ahead plans are elegant—but the real world is messy. Clouds scud across solar farms. A sudden gust alters turbine output. A factory line shuts down unexpectedly. Forecast errors creep in.
That’s why the intra-day rolling optimization stage exists: a 4-hour lookahead window, updated every hour, with 15-minute resolution. But here’s the twist: it doesn’t start from scratch. It respects the commitments made the day before.
Specifically:
- The EV and ESS power plans from stage one are treated as soft constraints—deviations allowed, but capped at ±5% per node.
- Only the first hour of each 4-hour rolling window is locked in; the rest is re-optimized next cycle (a technique called receding horizon control).
- EV discharge is disallowed (consistent with typical unidirectional charging today), preserving battery life and simplifying control.
- The on-load tap changer (OLTC) at node 33—the transformer linking the ADN to the main grid—is restricted to just six operations per day (at hours 1, 5, 9, etc.), preserving mechanical lifespan.
The objective here shifts: less about money, more about physics.
Three targets guide the intra-day optimizer:
- Minimize network losses—reducing I²R heating in cables, which directly improves efficiency and reduces thermal stress.
- Minimize voltage deviation—a critical quality-of-service metric. Voltages straying beyond ±5% (0.95–1.05 pu) risk equipment damage, flicker, and instability.
- Maximize renewable utilization—ensuring stage one’s gains aren’t undone by overly aggressive voltage control.
Because these goals conflict (e.g., absorbing more solar may raise voltages), the team used proportional scaling and judgment matrix analysis to assign weights: voltage stability got the lion’s share (58.42%), followed by renewable use (28.02%) and losses (13.50%). A pragmatic balance: reliability first, sustainability close behind.
The results? Striking.
- Maximum node voltage deviation shrank from 0.0600 pu to just 0.0253 pu—well within the ±2% band (0.98–1.02), even at the most volatile nodes (11, 15, 17, 20).
- Voltage profiles flattened dramatically: pre-optimization, 12 nodes exceeded 0.05 pu deviation; post-optimization, none did.
- Renewable utilization remained sky-high at 97.65%—only a 0.53-point dip from the day-ahead optimum, despite added physical constraints.
In practice, this means lights don’t dim, industrial motors don’t overheat, and inverters don’t trip offline during cloud transients. The grid breathes easier.
Why This Isn’t Just Academic—It’s Automotive
At first glance, this is a power systems paper. But look closer: node 7 hosts an EV aggregator. That’s not a theoretical placeholder—it’s a proxy for real-world EV charging networks: fleet depots, workplace chargers, or even residential V1G (vehicle-to-grid unidirectional) hubs.
The model treats the EV cluster as a dispatchable load block with power and energy bounds—exactly how utilities see demand-response resources today. And crucially, it shows that even without bidirectional (V2G) capability, EVs can deliver massive grid value.
Consider: the 100 kWh EV cluster contributed ~12% of the total flexible capacity (alongside 350 kWh of stationary storage). Yet because EV charging is highly shiftable—most drivers don’t care when their car charges, only that it’s full by morning—it offered more scheduling freedom per kWh than fixed batteries tied to specific discharge windows.
This has profound implications for automakers and charging providers.
First, smart charging isn’t optional—it’s table stakes. As ADNs scale, utilities will increasingly require—or incentivize—EVSE (electric vehicle supply equipment) that can accept dispatch signals. ISO 15118 (Plug & Charge with smart grid integration) and OCPP 2.0.1 (Open Charge Point Protocol) already support this. Expect mandates to follow.
Second, EVs may become grid assets on balance sheets. If an aggregator can reliably offer 50 kW of flexible load for 4 hours nightly, that’s equivalent to a small peaker plant—but with zero emissions and lower capex. Some European utilities are already piloting “virtual power plant” contracts with EV fleets.
Third, vehicle design must account for grid roles. Battery thermal management, charge acceptance rates, and even onboard communication modules will need to support grid-service profiles—not just fast DC charging. Tesla’s “Scheduled Departure” and GM’s Ultium Energy Services are early steps; the next leap is real-time coordination.
And finally, this research hints at a future where EVs earn while parked. Yes, full V2G is still held back by battery warranty concerns and inverter costs. But even controlled unidirectional charging can generate value: lower electricity bills via time-of-use arbitrage, demand-response payments, or carbon credits. In China’s increasingly market-driven power sector, those micro-revenues add up.
Beyond the Simulation: Scalability and Real-World Traction
Critics might point out: this was tested on a modified IEEE 33-node feeder—a classic benchmark, but small (total load ~3.7 MW) and radial. Will it scale to urban meshed networks with thousands of nodes?
The answer lies in the second-order cone (SOC) relaxation used for power flow modeling. Unlike traditional AC power flow—which is non-convex and computationally nasty—SOC converts the problem into a convex form that’s solvable in seconds, even for larger systems. And for radial networks (like most distribution feeders), the relaxation is exact: the solution satisfies the original AC equations. That’s not approximation—it’s mathematical equivalence.
Moreover, the two-stage structure is inherently modular. Stage one can run centrally at the DSO (distribution system operator); stage two can be decentralized—executed by edge controllers at substations or even EVSE hubs. With 5G and edge AI advancing rapidly, the latency and bandwidth hurdles are fading.
Already, early adopters are moving beyond pilots. In Shandong Province, State Grid is testing ADN dispatch with EV aggregators in Qingdao. In Guangdong, Nari Group has deployed real-time voltage optimization using capacitor banks and OLTCs—similar to the reactive devices in this study (SVCs at nodes 5/15/30, capacitors at 5/15). The pieces are in place.
What’s missing is integration—tying DERs, storage, and EVs into a single, hierarchical control stack. That’s precisely what Jia and colleagues delivered: not a new device, but a new operating system for the grid.
The Road Ahead: From Dispatch to Ecosystem
The true significance of this work isn’t in the decimals of cost reduction or voltage deviation. It’s in the mindset shift it embodies.
We’ve long treated transportation and power as separate sectors—different regulators, different engineers, different business models. But EVs are the physical bridge between them. Every time an EV charges, it’s a power system event. Every time a grid constraint limits charging speed, it’s a mobility event.
This two-stage model is a blueprint for co-design—where grid operators, automakers, and charging networks speak the same language: optimization variables, constraints, and multi-objective trade-offs.
Future iterations could incorporate:
- Bidirectional V2G, unlocking EV batteries as dispatchable generation (not just load).
- Uncertainty quantification, using stochastic or robust optimization to hedge against forecast errors.
- Transactive control, where EVs and other DERs negotiate via price signals instead of centralized commands.
- Cybersecurity layers, ensuring that dispatch signals can’t be spoofed or delayed.
But even in its current form, the model proves a vital point: the grid doesn’t need to fear EVs—it needs to recruit them.
As one engineer put it: “The battery in your garage is the most underutilized asset in the energy transition.” This research shows how to change that.
So the next time you plug in your EV, remember: you’re not just topping up for your commute. You might be helping hold the voltage steady at node 17, letting a wind turbine run at full tilt, and saving the grid ¥25 in procurement costs.
That’s not science fiction. Thanks to work like this, it’s already rolling out—quietly, efficiently, one dispatch cycle at a time.
Jia Ying¹, Liu Hanli², Zhao Shuqi¹, Li Yang³, Han Pengfei⁴, Chen Biao²
¹School of Electrical and Information Engineering, Northeast Petroleum University, Daqing 163318, China
²Power No.1 Company, PetroChina Electric Energy Limited Company, Daqing 163453, China
³Purifier Branch, Daqing Oilfield Water Company, Daqing 163453, China
⁴Information and Communication Center, PetroChina Power Supply Company, Daqing 163453, China
Journal of Jilin University (Engineering and Technology Edition), 2023, Vol. 53, No. 4, pp. 709–716
DOI:10.13229/j.cnki.jdxbgxb20220417