Introduction
Smart feeding systems are emerging in aquaculture. They are expected to bring improvement and optimization of feeding practices by digitizing real-time biological and environmental information, integrated with core biological knowledge, new technologies and machine learning to maximize profit while minimizing environmental footprint and maintaining fish welfare.
This digitization requires the automation of data collection, exchange, and decision support. In such a process, sensor data are collected, combined with other important data such as fish behaviour, feeding patterns, weather conditions, and finally processed by Artificial Intelligence (AI) algorithms to provide recommendations and management decisions, such as feeding management strategies and optimal harvest times.
The issue if that the offers are multiplying to provide vertically integrated solutions with limited capabilities for the end user to share data from one application to one another. This creates several drawbacks including the risk of vendor lock-in, the limited innovation capabilities or having a sub-optimal system not allowing data reuse.
This could be achieved by collecting and validating data from heterogeneous sources and organise it in an interoperable way so it can be used over several applications spanning over different domains. In this paper and following our previous work in [1], we will introduce our work toward the normalization of Aquaculture Data model for smart fish feeding support using standardised cross-domain specifications.
Data Platform
For managing data in a fish farming system, the proposed core data platform is based on FIWARE brings a curated framework of open-source software components which can be assembled and combined with other third-party platform components to build platforms easing the development of smart solutions and smart organizations.
In a nutshell, the NGSI-LD specification is based on data model and API. The NGSI-LD data model is an entity-based data model. An NGSI-LD Entity is an informational representation that is supposed to exist physically or conceptually. Relationships in NGSI-LD capture possible links between a subject which maybe an entity, a property, or another relationship on the one hand, and an object, which is an entity, on the other hand. NGSI-LD Property is a description instance, which associates a main characteristic to either an Entity, a Relationship, or another Property. Based on its cross-domain ontology, NGSI-LD covers several generic domains such as Mobility, Location, Temporal, System Structure and Behavioural Systems. The NGSI-LD API relies on this data model. It provides a set of operations on entities covering entity creation, update, deletion, retrieval, and subscription. The API also proposes operations that include temporal operations.
Normalising Aquaculture Data and NGSI-LD
A simplified overall schematic of Data management in a fish farming system is depicted in Figure 1. The main NGSI-LD entity in this fish farming system is the Fish Containment: Fish Cage, or Tank, are the equivalent concept for it. Data may be devised in different categories: Fish Behaviour, Feeding Processes, Water Quality and Weather. Weather Parameters (temperature, wind, cloud, waves height, etc.) will be added as a property to Weather Observed Entity. Water Parameters (pH, redox, dissolved Oxygen…) are added as properties to the Fish Containment entity.
We analysed a number of existing data models and ontologies including Agrovoc, WORMS, dbpedia; FOODI, Aquacloud, SmartdataModels to identify terms relevant for the aquaculture applications, list equivalence among the models and select a definition leading to the least unambiguous description of the term. This term is then mapped over the NGSI-LD interface so data exchanges can occur across the systems. The table below illustrates the discussion for one NGSI-LD entity and one NGSI-LD property. The complete property graph will be described in detail during the presentation.
Conclusion and Discussion
The NGSI-LD fish farming system is published in the smart data models repository. In this data model we have tried to normalize data from different sources and types in one common NGSI-LD model. We have also tried to link and map almost data model attributes to external knowledge and known thesaurus for interoperability issue.
Reference
[1] Abid, A., Dupont, C., Le Gall, F., Third, A., & Kane, F. (2019, June). Modelling Data for A Sustainable Aquaculture. In 2019 Global IoT Summit (GIoTS) (pp. 1-6). IEEE.