project - EIP-AGRI Operational Group
PREDIVÍ: big data-based wine harvest prediction model
PREDIVÍ: modelo de predicción de cosecha vitivinícola a través del big data
Objectives
The main objective of the project is to provide organisations and actors in the wine industry with decision-making support tools to obtain advance information on harvest predictions, particularly regarding:
1- Production volume per plot
2- Production quality parameters (grade, acidity, pH, etc.)
3- Classification of the plot qualitative potential
Objectives
el objetivo principal del proyecto es facilitar a las organizaciones y actores del sector vitivinícola herramientas de apoyo a la decisión para obtener anticipadamente información sobre predicciones de cosecha, particularmente en lo referente a:
1- Volumen de Producción por parcela.
2- Parámetros de calidad de la producción (grado, acidez, ph, etc.)
3- Clasificación del potencial cualitativo de las parcelas
Activities
The project has developed a model for predicting the quantity and quality of harvest with a margin of error similar to current methods, but which is much more efficient.
• If the historical data and those obtained automatically by the system (meteorological and satellite images) are used in conjunction with those from a single manual sampling (8 per field are possible in a conventional campaign) the margin of error is substantially reduced.
• The system engages in continuous learning, so the margin of error between the prediction and reality is expected to be increasing smaller in each campaign.
• The level of automation and the display systems for the results were considered optimal by the end users.
Activities
El proyecto ha desarrollado un modelo predictivo sobre la cantidad y calidad de cosecha con unmargen de error similar a los métodos actuales, pero más eficiente.
•Si a los datos históricos y los obtenidos por el sistema de forma automática se le añaden las correspondientes a un único muestreo manual (en una campaña convencional se pueden llegar a realizar 8 por parcela) el margen de error se reduce sustancialmente.
•El sistema está en aprendizaje continuo, por lo que se prevé que en cada campaña el margen de errorentre la predicción y la realidad sea menor.
•El nivel de automatización y los sistemas de visualización de los resultados se han consideradoóptimos por los usuarios finales.
Context
The project is based on the use of big data and machine learning technologies to develop prediction models that provide forecasts on quality parameters and harvest volumes in the wine industry. To that end, we used the information available to the companies and organisations participating in the project regarding production histories, plots, ripeness checks, sampling, etc. enabling the prediction to be made, together with the variables obtained in the campaign (meteorology, satellite images, etc.).
The model created greatly improved the efficiency (reduction of costs related to taking samples in the field) of the predictions of the two variables studied, with errors of around 10%.
The model has a powerful learning capability, which means the error will be reduced in future campaigns.
Project details
- Main funding source
- Rural development 2014-2020 for Operational Groups
- Rural Development Programme
- 2014ES06RDRP009 Spain - Rural Development Programme (Regional) - Cataluña
Location
- Main geographical location
- Barcelona
- Other geographical location
- Girona, Lleida
€ 191674.47
Total budget
Total contributions from EAFRD, national co-financing, additional national financing and other financing.
Resources
Contacts
Project coordinator
-
INNOVI Association of Innovative Companies
Project coordinator