数据标准化法

Data standardization method

对数据进行标准化处理，包括命名统一、单位转换、数值范围调整等，以确保数据格式一致，使其适应模型训练的要求

Standardize data through processes such as uniform naming， unit conversion， and adjustment of numerical ranges to ensure consistent data formats， meeting the requirements for model training

列出数据格式标准，将所有数据字段的名称、数据单位和范围进行统一。但对于数据量较大、管理框架耦合度较高的农场，将历史数据标准化，会有较大的工作量

List data format standards by unifying the names， units， and ranges of all data fields. However， for large-scale farms with high coupling in the management framework， standardizing historical data may entail significant workload

特征选择法

Feature selection method

针对不同农场的数据特征进行筛选和选择，确定共同的特征，丢弃不一致或不相关的数据特征，建立更一致的数据集

Tailor the selection of data features based on the characteristics of different farms， identifying common attributes， discarding inconsistent or irrelevant data features， and establishing a more cohesive dataset

使用机器学习中的实体对齐模型对数据字段、数据单位进行匹配，根据匹配结果进行调整。但是训练一个准确度高的实体对齐的模型，也是不小的工作量

Utilize entity alignment models in machine learning to match data fields and units， and make adjustments based on the matching results. However， training a high-precision entity alignment model is also a non-trivial undertaking

数据映射法

Data mapping method

通过特征转换、映射或转换函数来调整数据，使其在数值范围或格式上更为一致，以帮助模型更好地理解和处理数据

Utilize transfer learning techniques to train historical agricultural models or relevant task models for new tasks， thereby reducing agricultural model training time

每个农场给出自己的数据字段、数据单位和数据范围，联邦学习模型训练的发起方根据这些自己模型训练所需数据，开发转换函数，完成字段格式、数据单位和数据范围的转换，并对缺失值作出插值

Each farm provides its own data fields， data units， and data ranges. The initiator of federated learning model training， based on the required data for their own model training， develops transformation functions to perform conversions of field formats， data units， and data ranges. Additionally， interpolation is applied to handle missing values

数据来源

Data source

可以从农业机构、学校、企业、各个农业平台（隐私保护后的模型梯度）和本地农场采集等获得数据

Collecting data from agricultural institutions， schools， businesses， various agricultural platforms （with privacy-protected model gradients） and local farms

可以从农业机构、学校、企业和本地农场采集。其他农业平台较为隐私保密的数据无法获取利用

Collecting data from agricultural institutions， schools， businesses and local farms. Data from other agricultural platforms， which are more privacy-sensitive， cannot be accessed or utilized

数据隐私和安全性

Data privacy and security

参与方可以在本地更新模型梯度，发送给中心训练服务器时会使用隐私加密算法对梯度进行加密，可以减少数据隐私泄漏的风险

Participants can locally update model gradients， and when sending them to the central training server， privacy encryption algorithms will be used to encrypt the gradients， reducing the risk of data privacy leakage

如果进行直接的数据共享，可能会存在着数据隐私泄漏的风险

If direct data sharing is carried out， there may be a risk of data privacy leakage

模型泛化能力

Model generalization capability

可以使用更多的数据进行模型训练，使得模型可以更好的适应不同的农业环境，提升模型性能

By utilizing a greater volume of data for model training， the model can better adapt to diverse agricultural environments， enhancing overall performance

数据量不够大，可能在某些场景下会出现过拟合，使得模型性能波动较大

If the dataset is not sufficiently large， there may be instances of overfitting in certain scenarios， leading to significant fluctuations in model performance

合规性和法规要求

Compliance and regulatory requirements

联邦学习训练模型时，参与方可以在本地训练模型，不需要传输原始数据。本地训练模型过程中，可以通过同态加密和差分隐私的算法防止训练数据泄漏。这样的方式，可以满足大部分数据安全管理的法律法规要求

In federated learning model training， participants can train the model locally without the need to transmit raw data. During local model training， algorithms such as homomorphic encryption and differential privacy can be employed to prevent the leakage of training data. This approach ensures compliance with most legal and regulatory requirements for data security management

在一些国家和地区，数据的共享受到严格的法律保护。传统智慧农业平台要想使用这批农业数据会非常的困难，甚至是无法使用

In some countries and regions， data sharing is subject to strict legal protection. Traditional smart agriculture platforms may find it challenging， or even impossible， to use this batch of agricultural data

智慧农业系统的 可持续性

Sustainability of smart agriculture systems

通过对数据的隐私保护，联邦学习可以打造更大的加密数据资源共享生态，为农业生产的可持续发展提供更好的支持

By ensuring privacy protection for data， federated learning can establish a larger ecosystem of encrypted data sharing， providing better support for the sustainable development of agricultural production

需要考虑数据中心的可持续性

Need to consider the sustainability of data center