Influence of microclimatic conditions on dairy production in an Automatic Milking System: Trends and Time-Series Mixer predictions

Heat stress, exacerbated by climate change, poses a significant challenge to the dairy industry. Advances in Artificial Intelligence provide tools to project future conditions based on historical data, offering crucial resources to tackle heat stress impacts. This study evaluates and forecasts the impact of microclimatic conditions on dairy cow milk production, using data collected from an Automatic Milking Systems (AMS). Over a 6-year and 7-month period, we analyzed the temporal relationships among milk protein, fat, production, temperature, and Temperature-Humidity Index (THI). Cross-correlation analysis revealed that as THI increases, there is a corresponding decrease in protein and fat content, with a 5-day time lag for protein. We introduced the innovative use of Time-Series Mixer (TSMixer), a cutting-edge deep learning algorithm, to forecast the long-term evolution of milk production and quality at high temporal resolution. TSMixer effectively predicted daily protein, fat, milk production and milk temperature over two-month temporal horizon with high accuracy (R2 values of 0.83, 0.81, and 0.80). It outperformed baseline models, capturing simple linear and periodic behaviors, demonstrating superior predictive abilities, especially for primiparous cows. TSMixer advances milk production forecasting by offering high-resolution daily predictions for both milk yield and quality, unlike typical deep learning models that focus on lower-resolution, long-term forecasts. Its use of lightweight Multi-Layer Perceptrons makes it suitable for real-time deployment in AMS pipelines, aiding farmers in forecasting individual cow performance.