寻求农业作物产量和价格时间序列预测建议
英文摘要
A data professional at a major berry company is seeking community advice on machine learning-based time series forecasting for agricultural applications. Their role requires forecasting weekly crop harvest volumes and future commodity pricing using USDA and industry datasets. They have experimented with SARIMA, XGBoost, and Holt-Winters models and are now looking for production-grade libraries, suitable model architectures, and feature engineering strategies that incorporate weather, seasonality, acreage, and imports. The data is weekly and exhibits strong seasonality, with weather and supply shocks as key drivers. The post solicits recommendations on frameworks, papers, and practical lessons for forecasting in agriculture.
中文摘要
一位大型浆果公司的数据专业人员正在寻求农业领域机器学习时间序列预测方面的社区建议。其职责包括使用美国农业部和行业数据预测每周作物收获量和未来商品价格。他已尝试了SARIMA、XGBoost和Holt-Winters模型，现寻求可用于生产的代码库、合适的模型架构以及融合天气、季节性、种植面积和进口等因素的特征工程策略。数据为周频且季节性显著，天气和供应冲击是关键驱动因素。该帖文希望获得农业预测方面的框架、论文和实战经验推荐。
关键要点
The individual works at a major berry company and is responsible for forecasting weekly crop harvest volumes and future prices.
该人士在一家大型浆果公司工作，负责预测每周作物收获量和未来价格。
They have experimented with SARIMA, XGBoost, and Holt-Winters models using USDA and industry data.
已使用美国农业部和行业数据尝试了SARIMA、XGBoost和Holt-Winters模型。
Seeking recommendations on production-grade libraries/frameworks, model architectures for agriculture, and feature engineering incorporating weather, seasonality, acreage, and imports.
寻求关于生产级库/框架、适用于农业的模型架构以及融合天气、季节性、种植面积和进口的特征工程建议。
Data is weekly, highly seasonal, with weather and supply conditions as major drivers.
数据为周频、高度季节性，天气和供应条件是主要驱动因素。