基于时序生长度单元分析的播种期优化模型
Scheduling Planting Time Through Developing an Optimization Model and Analysis of Time Series Growing Degree Units
摘要：
在缩短的育种周期内生产更高质量的作物可确保全球粮食供应和安全，但由于储存限制，这一改进加剧了种子行业在全年育种过程中的物流和生产力挑战。在2021年先正达作物分析挑战赛中，先正达提出了一个问题，即为2020年全年育种过程中的种植时间安排设计一个优化模型，以便每周有一致的收获量。他们发布了一个数据集，其中包含2569个种子种群及其种植窗口，收获所需的生长度单位，以及两个地点的收获量。为了应对这一挑战，我们开发了一个新的框架，包括天气时间序列模型和优化模型来安排种植时间。设计了一个深度递归神经网络来预测未来的天气，并在时间序列模型之上开发了一个高斯过程模型来模拟预测天气的不确定性。所提出的优化模型还将种子群体的种植时间安排在最少的周数上，并且具有更一致的周收获量。与原始种植时间相比，使用所提出的优化模型可以在站点0和站点1处将所需容量降低69%和高达51%。
英文摘要：
Producing higher-quality crops within shortened breeding cycles ensures global food availability and security, but this improvement intensifies logistical and productivity challenges for seed industries in the year-round breeding process due to the storage limitations. In the 2021 Syngenta crop challenge in analytics, Syngenta raised the problem to design an optimization model for the planting time scheduling in the 2020 year-round breeding process so that there is a consistent harvest quantity each week. They released a dataset that contained 2569 seed populations with their planting windows, required growing degree units for harvesting, and their harvest quantities at two sites. To address this challenge, we developed a new framework that consists of a weather time series model and an optimization model to schedule the planting time. A deep recurrent neural network was designed to predict the weather into the future, and a Gaussian process model on top of the time-series model was developed to model the uncertainty of forecasted weather. The proposed optimization models also scheduled the seed population's planting time at the fewest number of weeks with a more consistent weekly harvest quantity. Using the proposed optimization models can decrease the required capacity by 69% at site 0 and up to 51% at site 1 compared to the original planting time.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
