Autonomous Digital Twin of Plants for Precision Farming

Overview

The Autonomous Digital Twin of Plants (ADTP) is an intelligent knowledge-based cyber-physical system designed to simulate and forecast crop growth and yield. By combining ontology-driven modeling, multi-agent systems, and machine learning, the platform captures the complex adaptive behavior of plants across multiple growth stages and parts. It enables farmers, agronomists, and agricultural technology providers to make data-driven decisions on planting schedules, resource allocation, and yield optimization. Initial prototypes have been validated against wheat and broccoli cultivation datasets from Russia and Taiwan, demonstrating strong adaptability to environmental events and progressively improving forecast accuracy.

Technical specifications

Core capabilities:

• Ontology-driven knowledge base that formalizes expert agronomic knowledge about plant growth stages, parts, and interdependencies
• Multi-agent model representing plant stages and parts as autonomous agents with collective decision-making and negotiation protocols for resolving resource-demand conflicts
• Non-equilibrium thermodynamics modeling of resource utilization, reflecting the dynamic balance between plant demands and environmental supply
• Machine learning integration to discover hidden interdependencies between input and output parameters and expand data collection fields
• Adaptive self-synchronization enabling real-time adjustment of plant behavior in response to unforeseen events such as heavy rains or cold weather
• Yield forecasting and schedule simulation with reported accuracy improving from 15-20% early-stage variance to 5-10% as the simulation progresses

Validation status:

• Tested with wheat and broccoli datasets in Russia and Taiwan
• Simulations compared against real historical crop cultivation data and expert agronomist models
• Experiments conducted with disruptive weather events at various growth stages

Technology readiness level

The technology is currently at a prototype stage (TRL 4-5). A first-generation Smart Digital Twin of Plants has been built and experimentally validated with real crop datasets, confirming feasibility and adaptability. The research team is advancing toward an Autonomous Digital Twin with real-time self-synchronization, expanded ontologies for plant parts, enhanced multi-agent negotiation protocols, and machine learning-driven parameter discovery. Planned validation for 2024-2025 includes expanded testing on wheat and additional crop types, with the goal of broadening the knowledge base and improving simulation accuracy for diverse agricultural applications.