A bi-directional cross-channel RNN model for time-series forecasting of dairy production

Area of Science:

• Agricultural Science
• Data Science
• Machine Learning

Background:

• Precision livestock management requires accurate milk production forecasting.
• Analyzing historical cow data (health, milk quality, seasonality) is key.
• Multivariate time-series forecasting models struggle with complex dairy data dynamics.

Purpose of the Study:

• To develop an advanced recurrent neural network (RNN) for predicting dairy cattle milk production.
• To effectively capture temporal dependencies and interrelationships among dairy variables.
• To improve the accuracy and efficiency of milk income prediction.

Main Methods:

• Proposed a novel RNN architecture using bidirectional Gated Recurrent Units (GRUs).
• Applied GRUs bidirectionally along the channel dimension for efficient interaction modeling.
• Incorporated feed-forward layers to implicitly capture temporal dependencies.

Main Results:

• Achieved superior or competitive performance in predicting cumulative milk income.
• Demonstrated effectiveness across different lactation periods using various error metrics.
• Exhibited lower time complexity compared to Transformer and convolution-based models.

Conclusions:

• The proposed bidirectional GRU model offers an effective solution for dairy cattle milk production forecasting.
• This approach enhances precision livestock management through accurate and efficient data analysis.
• The model provides a valuable tool for optimizing dairy farm operations and profitability.