Development of an Optimal Allocation System for Green Power Wheeling

Taiwan's export industries — ICT, semiconductors, metalworking — increasingly live or die on their access to certified green power. RE100 commitments from Apple, Google, and Microsoft, together with the EU's CBAM carbon border levy, have turned renewable procurement from a corporate-responsibility bonus into a hard market threshold. Yet Taiwan runs an isolated grid, and solar's intermittency produces a sharpening "duck curve": midday PV floods the system while the evening peak arrives after the sun is gone. As the market shifts from feed-in tariffs toward free trading and corporate power-purchase agreements, green power has become a scarce, tradable asset — and premium supply is largely locked up by large firms, leaving smaller users unable to buy, or unable to absorb their own surplus.

The renewable-energy retailer sits between generation and demand, and as wheeling (轉供) scales up, its allocation problem gets genuinely hard. Green-power wheeling is not bulk trading; it is a dynamic game among intermittent generation, fluctuating load, and Taipower's 15-minute settlement rule, played out across many generators and many consumers at once. Worse, most users hand over only monthly bills, so the retailer plans under real information asymmetry — over-commit and lose money, or stay conservative and miss the business. This project, commissioned by GREENET (天能綠電) and led by Prof. I-Yun Lisa Hsieh, treats wheeling as a multi-objective optimization problem — one that must satisfy regulatory rules, contractual promises, and profit at the same time — and builds a decision-support system around it.

The system is assembled from four modules. A demand-simulation module turns a user's sparse monthly bill (peak / half-peak / off-peak totals) into a high-fidelity 15-minute load curve, using load-shape logic for four archetypal users — a 24-hour factory, a weekday office, an all-week storefront, and an evening-and-weekend department store. A settlement engine encodes Taipower's official wheeling rules (Article 13 of the wheeling and direct-supply operating regulations) into verifiable logic: a first-pass 15-minute match taking the lesser of generation and demand, then a second-pass reallocation of surplus within the same price period, with standardized settlement reports. An optimization core then searches the multi-generator-to-multi-consumer matching space — tens of thousands of possible configurations — for the pairing that maximizes wheeling gross profit (or, switchably, total wheeled volume), under practical controls such as surplus-ratio thresholds and fixed-contract locks. Finally, a flexible-allocation module anticipates Taipower's sandbox program, letting the same engine switch between today's 15-minute matching and a parameterized, priority-weighted allocation for future policy scenarios.

Validated against Taipower's standard test cases, the system aims to replace slow, error-prone manual spreadsheeting with a repeatable, auditable, data-driven workflow: pre-contract risk assessment from simulated demand, compliant automated settlement, and profit-maximizing allocation that trims generation-side surplus and consumption-side shortfall alike. For the E3 Center, it carries its energy-systems and techno-economic modelling into the fast-moving business of green-power retail — and aims to leave behind a replicable technical benchmark for Taiwan's renewable-energy trading sector.