Adaptive training load optimization for track and field athletes: A reinforcement learning approach

Area of Science:

• Sport Science
• Artificial Intelligence
• Machine Learning

Background:

• Optimizing athlete training involves balancing performance enhancement with injury prevention.
• Current methods often struggle to dynamically adapt to individual athlete needs and recovery states.

Purpose of the Study:

• To develop and validate an AI-driven framework for offline optimization of athlete training loads.
• To create a system that dynamically prescribes training adjustments (intensity, volume, recovery) based on real-time physiological and performance data.

Main Methods:

• Utilized a Deep Q-Network (DQN) architecture with a data-driven digital twin for training simulation.
• Incorporated physiological data (HRV, sleep quality) and performance metrics (ACWR, weekly performance) from 25 athletes.
• Employed a feedforward neural network for Q-value estimation and optimized adaptive policies with a dual reward function.

Main Results:

• The AI model successfully reduced error rates and loss functions, converging towards zero.
• Demonstrated effective management of training load and control of injury risk.
• Dynamically adapted training prescriptions to maintain optimal athlete performance and health.

Conclusions:

• The proposed AI framework offers a robust solution for personalized athlete training optimization.
• It effectively balances short-term performance goals with long-term athlete well-being and injury prevention.
• This data-driven approach overcomes simulation gaps and ethical concerns of real-time experimentation.