Concluded Research Project

Artificial intelligence to accelerate the net-zero transition: sector coupling between electricity and transport

人工智慧加速淨零轉型：電力與運輸跨部門耦合

Duration
2022–2023
Funder
National Science and Technology Council (NSTC)
Grant no.
111-2222-E-002-019-

Uses artificial intelligence and interdisciplinary modeling to couple Taiwan’s electricity and transport sectors — long-term solar forecasting, solar-aware electric-bus charging strategies, and lithium-ion battery techno-economics — quantifying cross-sector decarbonization pathways toward net zero.

Key findings

  1. An LSTM model forecasts regional long-term solar capacity factor with R² = 0.796, outperforming XGBoost, MLP, and the published benchmark (0.724); Kriging interpolation of weather-station data materially improves accuracy.

    以長短期記憶（LSTM）模型預測各區域太陽能長期容量因子，R² 達 0.796，優於 XGBoost、MLP 及文獻基準（0.724）；克利金法空間內插顯著提升準確度。

  2. Shifting urban electric-bus fleets from overnight to daytime charging cuts operational emissions: at 40–80 GW national solar capacity, per-km emissions fall to 523–459 gCO₂e/km (a 14–24% reduction), saving roughly 220 tonnes of CO₂e per day.

    將市區電動公車車隊由夜間充電改為日間充電可降低營運排放：在全國太陽能裝置容量達 40–80 GW 時，每公里排放降至 523–459 gCO₂e/km（減少約 14–24%），每日約可減排 220 公噸 CO₂e。

  3. A two-stage learning-curve model projects EV battery packs reaching the US$80/kWh target around 2027 (LFP packs by 2023–24), with consistent price declines across cathode chemistries and an extension to stationary energy-storage systems.

    二階段學習曲線模型推估電動車電池組約於 2027 年達到每瓩時 80 美元目標（磷酸鐵鋰電池於 2023–24 年率先達標），各正極材料價格皆呈一致下降趨勢，並延伸至定置型儲能系統。

About this project

Reaching net zero requires the electricity and transport sectors — historically planned in isolation — to be designed together. This project applies artificial intelligence and interdisciplinary modeling to quantify the decarbonization potential of sector coupling in Taiwan, across three connected research thrusts.

1. Long-term solar forecasting. Land and climate constrain where solar can be built, so siting decisions hinge on knowing each region’s long-term generation potential. Combining ten meteorological inputs, Kriging spatial interpolation, and deep learning, the project forecasts regional solar capacity factor at weekly– yearly horizons. An LSTM model achieved an R² of 0.796, outperforming XGBoost, MLP, and the published benchmark — giving planners a tool to prioritize the most productive sites as Taiwan scales toward 40–80 GW of solar by 2050.

2. Solar-aware electric-bus charging. Using a bottom-up, hourly, regional grid emission model, the project shows that when an electric-bus fleet charges matters as much as that it electrifies. Moving urban fleets from overnight to daytime charging — to soak up midday solar — cuts per-kilometer operating emissions to 523–459 gCO₂e/km at 40–80 GW solar (a 14–24% reduction), saving on the order of 220 tonnes of CO₂e per day nationally.

3. Lithium-ion battery techno-economics. Battery cost is the key barrier to electrifying vehicles and storing renewables. A refined two-stage learning-curve model — separating raw-material, cell, and pack stages — projects EV battery packs reaching the US$80/kWh cost-parity target around 2027 (LFP packs by 2023–24), and extends to stationary energy-storage systems.

Together these results give policymakers a more forward-looking, quantitative basis for coupling clean-power growth with transport electrification. Work from this program also drew international attention: the principal investigator was invited onto the editorial board of the Nature-family journal Communications Earth & Environment, and the group’s low-carbon-mobility research was featured by BBC News and Finland’s Helsingin Sanomat.

邁向淨零，必須將過去各自規劃的電力與運輸部門整合設計。本計畫運用 人工智慧與跨領域建模，量化臺灣跨部門耦合的減碳潛力，涵蓋三個相互 連結的研究方向。

一、太陽能長期發電預測。 土地與氣候限制了太陽能的建置地點，因此選址必須 仰賴對各地區長期發電潛力的掌握。本計畫結合十項氣象因子、克利金法空間內插與 深度學習，預測各區域以週、月、年為尺度的太陽能容量因子。其中LSTM模型 的 R² 達 0.796，優於 XGBoost、MLP 及文獻基準，可協助決策者在臺灣邁向 2050 年 40–80 GW 太陽能目標的過程中，優先選擇發電效益最高的地點。

二、配合太陽能的電動公車充電策略。 透過自下而上、每小時、分區的電網排放 模型，本計畫顯示電動公車何時充電與是否電動化同等重要。將市區車隊由 夜間充電改為日間充電以吸收正午太陽能，可在 40–80 GW 太陽能下將每公里營運排放 降至 523–459 gCO₂e/km（減少 14–24%），全國每日約可減排 220 公噸 CO₂e。

三、鋰離子電池技術經濟分析。 電池成本是車輛電動化與再生能源儲存的關鍵 門檻。本計畫以精修的二階段學習曲線模型（區分原礦物、電芯與電池組階段）， 推估電動車電池組約於 2027 年達到 每瓩時 80 美元的成本平價目標（磷酸鐵鋰 電池於 2023–24 年率先達標），並延伸應用於定置型儲能系統。

這些成果為政策制定者提供更具前瞻性的量化依據，將綠電成長與運輸電動化加以耦合。 本計畫相關研究亦獲國際關注：計畫主持人受邀加入 Nature 旗下期刊 《Communications Earth & Environment》編輯委員會，研究團隊的低碳運具研究並 獲英國 BBC News 與芬蘭《Helsingin Sanomat》報導。

Related publications

Top