EV and AC Clusters Team Up to Solve Renewable Integration Puzzle
In a world racing toward carbon neutrality, the grid’s struggle to embrace renewable energy is fast becoming the invisible bottleneck behind the clean-energy revolution. Solar and wind—once heralded as near-magical solutions—are now exposing fundamental weaknesses in modern power systems. The sun doesn’t always shine, the wind doesn’t always blow, and when they do, sometimes they flood the grid at the worst possible moment. Enter two unlikely heroes: parked electric vehicles and humming air conditioners—not as energy consumers, but as agile, intelligent assets ready to balance the scales.
A groundbreaking study published in Power System Protection and Control reframes how we think about demand-side flexibility. The research, led by Hu Zhiyong, Guo Xueli, Wang Shuang from the Economic and Technological Research Institute of State Grid Nanyang Power Supply Company—alongside Xu Congming, Li Tingting, and Zhou Wei from Dalian University of Technology—demonstrates that large-scale EV and AC aggregations, when managed through a novel response willingness framework, can almost entirely eliminate curtailment of wind and solar power. And they do so not by brute-force control, but by respecting human behavior, uncertainty, and real-world variability.
This isn’t just another algorithmic exercise buried in academic notation. It’s a pragmatic blueprint for utilities grappling with daily overgeneration and reverse power flow—problems that have grown acute in regions like Henan, where distributed renewables now feed into the distribution grid in unpredictable surges. What makes this work stand apart is its psychological realism. Instead of assuming users will obey dispatch signals like obedient robots, the model quantifies how willing drivers and building occupants are to shift their behavior—based on battery levels, indoor temperatures, and, critically, the size of the incentive offered.
Imagine this: it’s a bright, breezy afternoon in central China. Rooftop solar panels and nearby wind farms are generating more electricity than local homes and factories can use. The substation buzzes—not with efficiency, but with stress. Without intervention, 5% of that clean energy must be curtailed, wasted, shut off at the source. It’s like turning off a fountain because the bucket is full—even though, just down the street, thousands of EVs sit parked with half-empty batteries and thousands of air conditioners cycle lazily in commercial buildings.
Traditional grid operators might respond by throttling generation or calling on expensive peaker plants later. But Hu and his team propose something radically different: ask the loads to dance.
Their method begins not with equations, but with empathy. A driver who just arrived home with 20% battery left is far more willing to plug in—even at moderate incentives—than one whose battery is already at 90%. Similarly, an office worker tolerates a one-degree indoor temperature shift during a heatwave far less readily than during mild spring weather. The researchers encode these human realities into a Takagi-Sugeno-Kang (TSK) fuzzy inference system, a well-established tool for modeling complex, imprecise relationships.
Rather than assigning fixed participation rates, the model treats response willingness as a dynamic, fuzzy variable—triangular in shape, with a most-likely value flanked by lower and upper bounds reflecting behavioral uncertainty. One EV owner might respond eagerly to a ¥0.20/kWh bonus; another, distracted or skeptical, may ignore the same signal. The TSK model absorbs this variance by blending multiple “if–then” rules: If SOC is low AND price is high, then willingness is high; If SOC is medium AND price is low, then willingness is moderate, and so on. Expert judgment calibrates the rule weights, bridging data and domain knowledge.
The brilliance lies in scaling this up. A single EV’s flexibility is negligible. But 5,000 EVs—modeled as an aggregated fleet—produce a dispatchable margin: a corridor of power that can be safely shifted upward or downward without violating battery constraints or user expectations. Same for 10,000 air conditioners. Crucially, the model links each cluster’s dispatchable margin directly to its real-time willingness level. When willingness is high, the corridor widens; when it’s low, the corridor narrows—keeping the system grounded in behavioral feasibility.
But willingness alone isn’t enough. To turn potential into action, the team layers on an incentive-based demand response strategy—a financial mechanism that rewards load adjustments in tiers. Think of it as a dynamic bonus system: the first megawatt of load reduction earns ¥0.05/kWh, the next tier ¥0.08, and so on, up to ¥0.50/kWh for deep, sustained shifts. This structure encourages participation without overpaying—crucial for economic viability.
The optimization objective is elegantly simple: minimize total operational cost, defined as
wind/solar curtailment cost + incentive payments + penalty for overloading tie-lines.
Curtailment cost isn’t abstract—it’s pegged to the real-world expense of firing up coal or gas units to compensate for lost renewables, plus environmental externalities. Tie-line penalties reflect physical grid limits: too much reverse flow, and transformers overheat.
What emerges is a self-correcting loop. When renewable overproduction looms, the optimizer computes how much EV charging and AC load can be safely increased (within their willingness-constrained margins), then sets tiered incentives high enough to attract that shift—but no higher. In deficit periods, it does the opposite: gently nudging loads down by offering modest payments for delay or reduction.
The simulation—set in a realistic distribution network with 5,000 EVs and 10,000 AC units—yields stunning results. Without any demand response, curtailment hits 4.77% of total renewable generation, peaking between 8 a.m. and 2 p.m., when solar output soars but industrial demand lags. With the proposed strategy in place? Zero curtailment. Every kilowatt-hour of wind and sun is absorbed—not by overbuilding storage or transmission, but by intelligently aligning flexible demand with supply.
Even more telling is the economic delta. Total system cost falls by over ¥16,000 compared to the baseline—primarily through reduced electricity purchases from the main grid. That may sound modest, but scale it to a city or province, and the savings climb into millions. And remember: this is achieved without forcing users into discomfort. EV drivers still leave home with full batteries; office occupants stay within ±1.5°C of their preferred temperature band—all while unknowingly performing grid-balancing heroics.
One of the study’s most insightful validations compares fixed-willingness assumptions against the dynamic TSK approach. When researchers assumed a flat 50% participation rate (conservative), total cost rose by ¥4,700. When they assumed 100% participation (optimistic), the model violated physical constraints: in several hours, the scheduled EV/AC loads fell below the minimum feasible power—meaning real-world execution would fail. In other words, overconfidence in user compliance doesn’t just inflate costs; it breaks the system.
This leads to another critical finding: uncertainty is manageable, but not ignorable. The team tested three levels of willingness uncertainty—none, mild (±5%), and high (±10%). As uncertainty grew, curtailment remained at zero, but external electricity purchases inched upward. Why? To hedge against the risk that some users won’t respond, the optimizer buys a little more backup power from the grid—like carrying an umbrella when the forecast says “40% chance of rain.” It’s a rational trade-off: slightly higher cost for guaranteed reliability.
Perhaps the most compelling insight involves synergy. The team ran four scenarios:
1) EVs + ACs both active
2) EVs only
3) ACs only
4) Neither
Results? EVs alone reduced curtailment cost by 99.97%—impressive, thanks to their high power and scheduling flexibility. ACs alone cut it by 57.6%. But together, they outperformed the sum of their parts. Why? Complementary profiles. EVs charge mostly at night and early morning, with sharp peaks; ACs run steadily during daylight hours, especially in summer. When renewables overproduce midday, ACs provide bulk absorption. When overproduction hits overnight (e.g., from wind), EVs swoop in. It’s a perfect temporal handoff—proof that diversity in flexibility is as vital as diversity in generation.
Zoom out, and this work speaks to a deeper transformation: the grid is no longer a one-way street. Power flows down, but intelligence and control must now flow up—from millions of endpoints, each with its own constraints and preferences. The old paradigm of top-down dispatch, where generators obey orders and loads passively consume, is fraying at the edges. In its place emerges a participatory grid—one that negotiates, incentivizes, and collaborates.
What Hu and colleagues have built isn’t a command system; it’s a marketplace for flexibility. EVs and ACs aren’t being coerced—they’re volunteering, in exchange for fair compensation. And because the model respects behavioral realism (via fuzzy willingness) and physical limits (via dispatchable margins), it avoids the classic pitfall of demand response: promised capacity that never materializes when called upon.
For utilities, the implications are profound. Distribution planners can now quantify how much EV and AC aggregation can contribute to renewable integration—hour by hour, season by season. Regulators can design incentive structures that align economic signals with grid needs. Automakers and HVAC manufacturers, meanwhile, gain a roadmap for embedding grid-friendly intelligence into future products: not just smart charging, but willingness-aware charging.
Already, pilots in Shanghai and Shenzhen are testing similar concepts—EV fleets responding to real-time price signals, commercial buildings pre-cooling before peak rates kick in. But most stop short of modeling uncertainty in human response. That’s where this study adds unique value. By treating willingness as fuzzy rather than binary, it injects humility into the optimization—acknowledging that humans are not devices, and flexibility has soft edges.
Yet challenges remain. The model assumes homogeneous fleets—same EV specs, same AC setpoints. Real-world heterogeneity (Tesla vs. BYD, inverter-driven vs. fixed-speed compressors) will require more granular clustering. Cybersecurity and data privacy loom large: to estimate real-time SOC or indoor temperature, systems need access to sensitive device data. And perhaps most critically, consumer trust must be earned. Incentive programs fail when users feel manipulated or shortchanged.
Future work, the authors suggest, could integrate peer-to-peer energy sharing—where EVs not only absorb surplus but feed it directly to neighbors—or couple willingness models with machine learning that adapts to individual behavior over time. One can imagine an app that learns your routine: if you usually leave at 8:30 a.m. with 80% charge, it won’t ask you to delay charging beyond 7:45 a.m., no matter how high the incentive.
But the core message is already clear: the flexibility we need isn’t locked in billion-dollar battery farms. It’s parked in garages and humming in ceilings—waiting, not for commands, but for the right reason to act.
As renewable penetration climbs toward 50%, 60%, even 80%, the grid’s greatest asset may not be its wires or transformers, but its users—empowered, compensated, and invited to be part of the solution.
That’s not just engineering. It’s energy democracy.
Hu Zhiyong, Guo Xueli, Wang Shuang, Xu Congming, Li Tingting, Zhou Wei
Economic and Technological Research Institute of State Grid Nanyang Power Supply Company, Nanyang 473000, China; School of Electrical Engineering, Dalian University of Technology, Dalian 116024, China
Power System Protection and Control, Vol. 51, No. 15, 1 August 2023
DOI: 10.19783/j.cnki.pspc.221699