SAREF-Compliant Knowledge Discovery for Semantic Energy and Grid Interoperability

—Modern cities are becoming "green and sustainable" once effective and optimized energy management of resources, gathering, and integration of data from smart energy and building are executed. The integration of heterogeneous technologies, and devices, require an interoperable solution to describe devices and data exchanged. SAREF is an ontology supported by the ETSI SmartM2M standard to achieve interoperability among IoT projects, architectures, etc. that can be extended to any IoT vertical domains such as smart buildings or energy. SAREF can be used to describe data sent through communication protocols or once that data must be processed (e.g., on the cloud, gateways, devices). ETSI does not provide tools yet that support the SAREF ontology to avoid engineers from developing from scratch. We design a SAREF-compliant sensor dictionary which also overcomes SAREF limitations. The sensor dictionary, applied to energy scenarios, is employed by a reasoner to infer meaningful knowledge from sensor data. Online demonstrators are available. The scenarios are also relevant for the Interconnect European-funded project that comprises 50 partners to design interoperable smart buildings and grids.


I. INTRODUCTION
Modern cities are becoming "green and sustainable" once effective and optimized energy management of resources, gathering, and integration of data from smart building, and energy are executed.The integration of heterogeneous technologies, and devices, require an interoperable solution to describe devices and data exchanged.
Several devices are designed to monitor energy consumption: the Linky meter is deployed in more than 30M French houses and building, and there are other solutions such as Hager, Schneider Electric, Legrand, Google PowerMeter, and Apple Reveals Smart-Home Energy Management.PowerMeter 1 was designed in 2009 to record the user's electricity usage in near real-time but abandoned two years later.Apple Reveals Smart-Home Energy Management Dashboard System2 in 2010.If the companies do not develop the full-stack compatible with 1) sensors, 2) the final application for consumers, 3) and other products; the designed product can fail to succeed.It illustrates the need for interoperability between devices, applications, and sensor data.In this paper, we focused on semantic interoperability on data to infer meaningful information.
SAREF is an ontology supported by the ETSI SmartM2M standard (M2M stands for Machine-to-Machine) to achieve interoperability among IoT projects, architectures, etc. that can be extended to any IoT vertical domains such as smart building, and energy [1].SAREF can be used to describe data sent through communication protocols or once that data must be processed (e.g., on the cloud, gateways, devices).ETSI does not provide tools yet that support the SAREF ontology to avoid engineers to develop from scratch.There is a need to design a SAREF-compliant sensor dictionary applied to energy scenarios.The dictionary is employed with a reasoner, to infer meaningful knowledge from sensor energy data.The scenarios, accessible via online demonstrators (see Table II), are also designed for the Interconnect European-funded project 3 .
We address the following research questions (RQ): • RQ1: What are the sensors relevant to the energy domains?Are there standardized sensor dictionaries for the energy domain?Are there ontology standards for the energy domain?• RQ2: What are the limitations of the existing ontologybased energy projects?Can we reuse the domain expertise designed in past projects?• RQ3: What are the rules and reasoning mechanisms to interpret energy sensor data to help developers design faster their IoT energy applications?
The main contributions of this paper are: • C1: Alignment of the sensor dictionary with the SAREF ontology and its extensions (SAREF4ENER, SAREF4BLDG) and the IEC 61360 -Common Data Dictionary standard 4 ; it addresses RQ1 in Section II.[2], and the AIOTI WG03 Standardisation 5 Semantic Interoperability Expert Group [3] [4] where the reasoner is taken as a baseline [5].SAREF designers are also involved within AIOTI WG03.
Structure of the paper: The sensor dictionary for energy is described in Section II.SAREF-compliant energy scenarios are introduced in Section III; the classification of the source of the knowledge to prove the veracity of our scenarios is included.The SAREF limitations are summarized in Section IV.The paper concludes and envisions future work in Section V.

II. SAREF-COMPLIANT SEMANTIC SENSOR ENERGY DICTIONARY
We built a sensor dictionary for energy compliant with standards such as ETSI SAREF and IEC 61360 Common data Dictionary.
Sensor Dictionary for Energy: We designed a pattern to classify sensors for the energy domain: for each sensor, we provide the produced measurements and the associated unit; we also deal with synonyms.Furthermore, we referenced for each sensor the source of knowledge using it (e.g., past projects referenced within the ontology-based IoT project catalog (see Section III), and reasoning mechanisms to interpret energy sensor data (see the rule discovery project in Section III).
Sensor Dictionary Compliant with IEC 61360 -Common Data Dictionary: The sensor dictionary is extended to be compliant with the IEC 61360 -Common Data Dictionary standard 6 (see Table I, the last column).The sensor dictionary highlights the limitations of the IEC 61360 standard which references less than ten types of sensors.

SAREF-Compliant Semantic Annotation:
The sensor dictionary is compliant with the terms employed by SAREF when possible, as illustrated in Table I.The limitations of SAREF are summarized in Section IV.Sensor energy datasets (e.g., JSON, XML) follow the SenML format 7 to represent sensor measurement, its value, its unit, and its timestamp.Available demonstrators are providing code examples (see Section III).A simple rule repository for the semantic annotation is already available such as if "t" or "temp" or "temperature" and located in the room probably it is a room temperature and will be annotated following the dictionary mentioned above which is implemented as an ontology.

III. SAREF-COMPLIANT ENERGY SCENARIOS
We designed scenario demonstrators (collected in Table II) to answer the following competency questions by executing the reasoning and query engines: • Retrieving the sensors relevant for the energy domain.
The sequence diagram (illustrated in Figure 1) explains the interactions of the sensor dataset and its annotation, storage and execution of reasoning and query engines: 1) Semantically annotating the energy dataset to be compliant with ontologies, 2) Storing triples (semantic sensor data) to the triplestore, 3) Loading semantic datasets, 4) Loading ontologies, 5) Loading rules, 6) Executing the reasoning engine with the set of rules compliant with the ontologies which will update the triplestore, and 7) Loading and executing the SPARQL query to retrieve the semantic datasets and ontologies by selecting a subset of the triplestore content.III).Our survey is the result of a continuous enrichment of the LOV4IoT ontology catalog [6] since 2012 dealing with more and more expertise and synonyms (e.g., smart grid, renewable energy, power plant, micro-grid, etc.).We provide tools to support the reuse of the survey outcome (e.g., dump of ontology code, web services, and web-based ontology catalog) and release them for the AI4EU Knowledge Extraction for the Web of Things Challenge 11 .Meanwhile, we are aware of Systematic Literature Review (SLR) guidelines such as [7].The survey on energy ontologies is also reused within the reasoning discovery explained hereafter; both are frequently updated (see Table II for URLs).Manual extraction and semi-automatic analysis [8] [9] have been done to extract knowledge.

Knowledge Discovery and Reasoning for Energy (S-LOR Energy):
The sensor energy dictionary depicted in Table I described in Section II is employed within the energy rule discovery [34] (see Table II for URL).Each sensor for the energy domain can be automatically retrieved using SPARQL queries; the sensor dictionary is displayed in Figure 2, the content shown on the GUI is from Table I and Table III.
End-to-End Energy Scenarios: The knowledge acquired to infer meaningful information is implemented as rules.We simulated sensor data (e.g., power consumption data using the SenML format).We annotated the data to be compliant TABLE I: Subset of the SAREF-compliant sensor dictionary, applied to energy (similar sensor tables for building, weather, and air quality can be provided).We focused on sensors employed within the Interconnect project.with ontologies (the file can be access though the demos referenced in Table II).The semantic reasoner (e.g., Jena inference engine) is executed on the semantic sensor dataset and the set of rules all compliant with each other.For instance, after executing the reasoning engine, a rule has been executed that deduces that a mobile phone uses 5 Watt, as illustrated in Figure 3.We have more and more scenarios (Table II) including various sensors (introduced in Table I), as an example: • Energy Data: power, frequency, luminous efficiency, electric current, electrical potential, resistance, thermal con-Fig.2: Demonstrator: the sensor dictionary queried and used via web services and displayed on the GUI ductivity, etc. • Weather Data: solar radiation data with 5 rules for various thresholds (see Table IV or the S-LOR Energy online demo for a more exhaustive sensor dictionary (URL in Table II).• Home Data: thermometer, light sensor, occupancy detector, etc. are used in a cross-domain home scenario.• Air Quality Data: NO2, PM10, O3, PM2.5, SO2, CO.
• More and more scenarios are being added within the same TABLE III: Ontology-based energy and grid projects (total=22), more are referenced by the LOV4IoT-energy ontology catalog, including reasoning mechanisms and sensors employed.Some works are not referenced here due to space issues.Legend: Ontology Availability (OA), when the code is available, the ontologies are classified on the top.Then, the ontology-based projects are classified by year of publications.* means that the ontology has been sent to us but it is not publicly accessible online.Even if the ontology is not accessible, we are referencing the projects since reasoning mechanisms and sensors relevant to the energy domain are mentioned.
demos with different kind of sensor data (as referenced within Table I) and more and more rules to deduce high level information.
Technologies employed in the implementation: Our demonstrators (Table II) are implemented with the Jena Semantic Web framework that can deal with RDF, RDFS, OWL languages to implement the sensor dictionary.This dictionary is implemented within the M3 ontology which is refined for the energy domain; Java to develop REST web services with the JAX-RS library to hide the complexity of using semantic web technologies, and the Graphical User Interface with Ajax, JQuery, JavaScript, HTML, and CSS.IV: Subset of the SAREF-compliant sensor dictionary applied to the weather domain.

IV. KEY CONTRIBUTIONS AND LESSONS LEARNT
We designed a sensor dictionary and reasoner compliant with those standards: SAREF (SAREF-Core, SAREF4ENER, and SAREF4BLDG), IEC 61360 Common Data Dictionary, and W3C SOSA/SSN, employed within our semantic datasets, SPARQL queries, and rules.We have analyzed the following limitations of SAREF specifications which highlights the need of our unified dictionary explained in Section II. to SAREF, is to link and unify existing ontologies to achieve semantic interoperability.We added links such as rdfs:subclassOf, owl:equivalentClass, and rdfs:seeAlso.SAREF or W3C SOSA do not consider the IEC 61360 -Common Data Dictionary standard 13 that we address in Table I and Table IV.• IoT Alliance: Our work is taken as a baseline in AIOTI 14(Alliance for the Internet of Things Innovation) WG03 Standardization Semantic Interoperability white papers [3] [5].

V. CONCLUSION AND FUTURE WORK
Designing energy efficiency applications requires crossdomain knowledge acquired from heterogeneous communities (e.g., home/building, weather, air quality).Integrated machine interpretable knowledge implemented within ontologies help when domain experts are not available.Our experience and expertise are shared within standards (e.g., editors of the ISO/IEC 21823-3 IoT semantic interoperability).Future work will consider "How to do the real actuation?"(switch on/off devices) once the reasoner is providing the suggestion.
Interconnect Use Cases: The SAREF-compliant sensor dictionary and rule-based reasoning is a first step towards the development of more sophisticated applications, left for future work such as: 1) reducing electricity bill, 2) energy consumption diagnosis, 3) voltage control of residential building, 4) enabling smart parking services, and 5) need of minimum X kW immediately for emergency, Y kW until 6 pm, Z kW surplus.
As a long-term solution, we will infer meaningful value from IoT data generated by devices using more sophisticated Artificial Intelligence technologies.Ideally, we also plan to contribute and be fully compliant with more standards such as W3C SSN/SOSA, W3C Web of Things, and iot.schema.org.

TABLE II :
Available demonstrators: ontology catalog, rule discovery, and full scenarios for energy, building, and weather

TABLE
Section III and TableIII)).Similarly, the S-LOR project (see Section III) suggests reasoning mechanisms (e.g., rules) for specific sensor type.It is not provided by SAREF.•Interlinking ontologies: our proposed solution, compared • Unifying sensor metadata: We structure sensor data in TableIand Table IV: 1) sensor name, 2) the produced measurement, 3) the associated unit (e.g., Watt to be more explicit than saref-core:PowerUnit).There is a need of domain experts to verify synonyms (e.g., solar radiation UV, Solar Panel, Photovoltaic).Each row of the Table (sensor, measurement, unit),