SBIR Phase I: High Fidelity Climate Simulation Powered by Generative Adversarial Networks

这个小企业创新研究（SBIR）第一阶段项目的更广泛/商业影响是创建一个广泛的（1000个结果），超局部（小于3公里）气候模拟档案，电网规划者和能源行业投资者可以使用它来更好地了解电网可靠性和可再生能源资产可行性的前瞻性风险。这些模拟数据将在德州电子可靠性委员会（ERCOT）电网内的所有地点进行预先计算，使规划者和投资者能够快速模拟不同可再生能源容量路径和不同电气化趋势的概率影响。最终，这些数据将支持更可靠的电网和更快的能源转型，因为决策者将能够访问包含极端事件、自然变化和气候变化的未来天气数据的单一来源。这个小企业创新研究（SBIR）第一阶段项目建议创建一个气候模拟引擎，为许多地点和许多天气变量生成每小时的综合当地天气模式（所有这些都需要对能源资源进行建模，如公用事业需求、风力发电和太阳能发电）。该项目将不依赖基于物理的全球气候模型，因为这些模型的计算强度很大，而且需要模拟当地而不是区域或全球天气。相反，该项目将研究统计模拟与人工智能（AI）的创新结合，利用各自的优势来弥补对方的弱点。例如，统计仿真模型是精确的，但不能扩展，而人工智能仿真模型几乎可以无限制地扩展，但不精确。该项目研究将研究一种新方法，通过已知的统计数据对人工智能模式生成施加精度，从而产生大规模的高保真气候模型。该项目的预期技术成果是创建一个模拟引擎，可以模拟德克萨斯州每小时超局部天气的1000个结果，其精度与纯统计模型基准相似，同时保持较低的云计算资源成本。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase I project is the creation of a broad (1,000 outcome), hyperlocal (less than 3 km) climate simulation archive that can be used by power grid planners and energy industry investors to better understand forward-looking risks to grid reliability and renewable energy asset viability. This simulation data will be pre-computed for all locations within the Electronic Reliability Council of Texas (ERCOT) power grid, enabling planners and investors to quickly model the probabilistic impact of different renewable energy capacity pathways and different electrification trends. Ultimately, this data will support a more reliable grid and faster energy transition because decision-makers will have access to a single source of future weather data that incorporates extreme events, natural variability, and climate change. This Small Business Innovation Research (SBIR) Phase I project proposes the creation of a climate simulation engine that generates synthetic hourly local weather patterns for many locations and many weather variables (all that are needed to model energy resources such as utility demand, wind generation, and solar generation). The project will not rely on physics-based global climate models due to the computational intensity of those models and the need to model local rather than regional or global weather. Instead, this project will research an innovative combination of statistical simulation with artificial intelligence (AI), leveraging the strengths of each to compensate for the weaknesses of the other. For example, statistical simulation models are precise but do not scale, while AI simulation models can scale almost without limit but are not precise. The project research will investigate a new method to impose precision (via known statistics) on AI pattern generation, yielding a high-fidelity climate model at scale. The expected technical result of the project is the creation of a simulation engine that can simulate 1,000 outcomes of hyperlocal hourly weather over the state of Texas--with accuracy similar to a pure-statistics model benchmark while keeping the cost of cloud computing resources low.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.