Most existing building automation systems are operated with rule-based settings. These systems are wasting a lot of energy because the systems can not properly cope with changes in indoor/outdoor environments. In this paper, we propose hybrid inference for inferring indoor environments in the building using real-time stream data coming from BAS. Hybrid inference consists of Runtime Stream Processing and Semantic Lift Processing. Runtime Stream Processing deduces occupancy and thermal comfort using machine learning technology with historical data. Semantic Lift Processing uses the semantic inference to extract new knowledge based on inferred results from Runtime Stream Processing. On the basis of stored semantic-based data in the ontology, Semantic Lift Processing derives energy waste space based on occupancy and thermal comfort by using semantic technology.
Korea is also under pressure to achieve its 37% reduction target of greenhouse
gas(GHG) emissions (comparison on BAU) by 2030, due to Paris Climate
Agreement. In Korea, GHG emission accounted for industry (50.1%), buildings (25.2%),
transportation (17.6%) and others (7.1%)[
that change settings in a time-dependent manner. The latest buildings also use this method in same way. There are two major problems in BAS. The rst problem is that precise control is impossible due to the lack of micro monitoring for individual space. The second problem is that it is di cult to create and apply a control algorithm for individual space.
In this paper, we analyze energy waste space based on real-time data from Researcher Hotel of Alto University in Espoo, Finland. The Researcher Hotel is a dormitory-type building for the researchers with total oor area of 8,206 m2 and each oor area of 2,450 m2. The building automation system is composed of the Fidelix BAS system in Finland and can collect data in real time using the SOAP protocol.
Classification
Scale
Usage Total ground area
Finish material
Area Summary
First Floor Second Floor Third Floor Fourth Floor Fifth Floor
Researcher Hotel
Details 5 floors 2,450㎡ 2,450㎡ 2,450㎡ 2,450㎡ 2,450㎡ 8,206㎡ Concrete
Total 4,199 data points which consist of HVAC, indoor/outdoor temperature, illuminance, and other environment properties are connected to our inference system. In this paper, we determine the internal thermal comfort based on each apartment's temperature and humidity. Also, we detect the occupancy status which is highly related to the CO2 concentration level. The extracted semantic based information is updated, integrated with the building meta information and the real-time sensing information that is constructed by the ontology. Then we generate the SPARQL query for this ontology of the wasted space inside the building, to visualize on a web service. By providing such user interface to the building manager, the manager can easily identify the status of the building and learn the elements necessary for management at a glance which can lead to energy savings.
The Internet of Thing (IoT) produces the data from the sensors by connecting to the Internet. Although, the BAS, is still under controversy whether it is IoT technology or not. If the BAS transmits its data via Internet, it could be an excellent example of IoT.
A variety of building energy saving technologies have been studied since the Paris
Climate Agreement. In most cases of building energy savings, research topic is
how to use energy more e ciently than before. We claim that the most important
problem in energy saving is to determine whether the space is occupied or not. A.
Caucheteux[
Another research on building energy is about state inference technology
through ontology connection of existing BAS data. Hendro Wicaksono[
Some of other IoT-based ontology studies[
The BAS system at the Researcher Hotel provides SOAP-based interface, so that data can be retrieved easily from outside through SOAP request. The Linked Open Data Adapter in Fig. 2 transmits these SOAP requests periodically and collects real-time data at 10 seconds intervals through the response. Collected data is stored through two interfaces: OpenTSDB3for storing historical data and Apache Fuseki's4 triple store for the semantic-based data repository. OpenTSDB is used to analyze runtime streaming of history-based time series data, and triple store stores the data needed for Semantic Lifting. Fuseki is used as an ontology
repository since it is not suitable for storing history data. Therefore our system only updates the latest value of BAS data on the ontology.
The IoT(Internet of Things) adapter shown in Fig. 2 could be connected to various legacy systems and IoT devices. It can translate the sensing data to RDF schema using the semantic annotation.
The system proposed in this paper uses hybrid inference of Semantic Lift Processing and Runtime Stream Processing as shown in Fig. 3. Runtime Stream Processing is composed of two algorithms: one is to determine occupancy state according to historical data stored in OpenTSDB, and the other is to calculate internal thermal comfort by using real-time temperature and humidity data. The semantic-based information generated by Runtime Stream Processing and the latest data of BAS are stored in the triple store and will be provided to the building manager. Stored semantic-based information is used to perform inference of energy waste space according to Description Logic in Semantic Lift Processing. This will make the building manager can identify which space wastes the energy by using the ontology.
The adapter layer in Fig. 3 shows the architecture of hybrid inference. In
this architecture, IoT adaptor translates the data to semantic-based
information using the domain ontology and semantic annotation. Time series data and
semantic-based information will be published to the Pub/Sub Message Broker
for delivering to the OpenTSDB, triple store, and the upper application.
Runtime Stream Processing extracts semantic-based information from historical
time series data by using machine learning. In this paper, there are two ways of
analysis processes: one is occupancy detection based on CO2, and the other is
that analyzes thermal comfort based on temperature and humidity. Occupancy
is monitored by CO2 data transmittied from Researcher Hotel. We refer to the
Tachikawa's equation[
C = C0 + (CS
C0)e VQ t + (1 e VQ t) M
Q (1)
In the Seydel's Equation which is Equ. 1 above, the amount of M pollution emission can be calculated as the concentration of CO2 emitted by a person, which is changed to k(amount of CO2 emission per person) and n(number of people), and the ventilation e ciency coe cient is applied to the ventilation amount of Q to determine the number of occupants. By solving the equation(1)'s n (number of people) in terms of other variables, we get Tachikawa's equation(2) shown below.
n =
aQ k(1 e aVQ t)
C
C0 (CS
C0)e aVQ t (2)
In Equ. 2, the variables consist of C(Current Pollution Level), C0(Lowest Pollution Level), CS (Normal Pollution Level), Q(Ventilation Amount), V (Space Size), k(CO2 emission amount per person), (Ventilation E ciency). Based on history based CO2, the lowest, average, and current pollution levels of past CO2 concentrations are selected to determine the number of occupants on Tachikawa's equation basis. Q, V , k, and are xed values belongs to the space, thus they are set to constant values based on building information. Based on the above equation, we compute the coe cients of each room and calculate the number of occupants with CO2 data based on the coe cients.
Since CO2 concentration baseline values are di erent in night, daytime, weekday, weekend, and semester breaks, we need to take di erent criteria on past historical data. Then, we can get a precise answer by implementing the algorithm. The number of occupants in each room is calculated as follows when the algorithm is executed. If the occupancy rate is 0.75 or more, it is judged that occupancy of the individual space is occupied or it is judged that it is vacant. The reason is that because the number of occupants increases as the concentration of CO2 increases, the standard of around 0.75 can be judged more quickly. 4.2
Thermal comfort algorithm The spatial thermal comfort analysis algorithm calculates temperature and humidity from BAS data. The features used in this algorithm are PMV(Predicted Mean Vote) and PPD (Predicted Percentage Dissatis ed) from the ISO 7730 standard5. The PMV index is represented in real number from -3 to +3, with 0 being the most pleasant, negative being cold, and positive being positive. The PPD is a dissatis ed index indicating how many percentage people are currently dissatis ed with the PMV thermal comfort index.
PMV is the value that is calculated by various variables such as amount of activity, external work, insulation value of clothes, heat transfer coe cient and so on. This value predicts the average temperature sensed by people who are in the same exposed environment. We calculate the PMV by using real-time transferred temperature value from BAS. Other various variables are used as
input by de ning constant values for each season. PMV is the thermal comfort that a person can feel, PPD predicts how many people are dissatis ed. By using these two indicators, it is possible to estimate how much the individual space consumes the energy of heating/cooling properly compared to the season and outdoor temperature, and the energy consumption of the individual space can be e ciently controlled. 5
In Semantic Lift Processing, an instance of a BAS data point of the Researcher Hotel is constructed by the ontology. The building semantic model proposed in this paper extends the SAREF6 (the Smart Appliance REFerence ontology) to build the ICBMS 7 ontology. In addition to SAREF, OM8 (Ontology of Units of Measure) and Time Ontology9 are used for the unit and time of data measured by the sensor. However, ppm units of CO2 concentration are not de ned in OM, so we de ne it through ICBMS ontology. And sensor types not provided by SAREF are also extended and de ned.
and Information Sharing among ICBMS (IoT, Cloud, Big-Data, Mobile, Security) Platform". That's why we use "ICBMS" as the name of the ontology.
VacantState or OccupiedState. hasThermalComfortState represents the thermal comfort index of seven stages as the PMV.
We use the internal structural information of Researcher Hotel to create the instance of the whole building, oor, room, and sensor unit. Then, we build the state information of each room into the ontology. This will make such structure that enables SPARQL queries can query information such as occupancy state of the room, the room that has good thermal comfort, and etc.
The system creates an instance using the received data from the BAS based
on the ICBMS ontology. The semantic-based information of the BAS data is
converted by a semi-automatic format similar to the internal label mapping
tool[
Fig. 8 shows a IoT Adaptor for collecting BAS data generated by buildings. The legacy adapter provides connectivity through the protocol interface (SOAP) used by BAS. The semantic annotation converts data as RDF schema according to BAS naming rules, and stores them in triple store. However, storing historybased sensing data in the triple store causes performance degradation. Therefore, only the latest data value is updated in the triple store, and the historical data is stored in the Time Series Database which is called OpenTSDB.
By using the constructed ontology, we can infer the space in which energy is currently wasted inside of the building. We infer the energy use state of space inside the building using Description Logic. Equ. 3 is the Description Logic for inference; it infers a place where has good thermal comfort even if a person is not in the place and a place where has excessive thermal comfort when a person is in the place. Our software does not use semantic reasoning such as SWRL(Semantic Web Rule Language)10. Our inference software loads the semantic information into the memory and deduces the relationship in Equ. 3.
u(8hasThermalComfortState.NeutralState ) or
(3)
List 1.1 shows the SPARQL statement for querying the data stored in the ontology. This query can be used to verify the sensing information and the inference result of each room.
PREFIX r d f s : <h t t p : / /www. w3 . o r g / 2 0 0 0 / 0 1 / r d f schema#> PREFIX om : <h t t p : / /www. wurvoc . o r g / v o c a b u l a r i e s /om 1.8/> PREFIX r d f : <h t t p : / /www. w3 . o r g /1999/02/22 r d f s y n t a x ns#> PREFIX xsd : <h t t p : / /www. w3 . o r g /2 0 0 1/XMLSchema#> PREFIX r d f s : <h t t p : / /www. w3 . o r g / 2 0 0 0 / 0 1 / r d f schema#> PREFIX t i m e : <h t t p : / /www. w3 . o r g /2 0 0 6/ t i m e#> PREFIX icbms : <h t t p : / / k e t i 1 . e n e r g y i o t l a b . com/ icbms#> 10 https://www.w3.org/Submission/SWRL/
PREFIX s a r e f : <http : / / o n t o l o g y . tno . n l / s a r e f#> PREFIX i c : <http : / / imi . i p a . go . jp / ns / c o r e /210#> ?room icbms : hasNumberOfPeople ?nop . ?room icbms : hasObjectId ? o i d . ?room icbms : hasPMV ?pmv . ?room icbms : hasPPD ?ppd . ?room icbms : hasOccupancyState ? o c c s t a t e .
Listing 1.1. SPARQL query
Fig. 9 shows the results of the query in List 1.1. In the application of intelligent building, this query is used to get the required data for visualization to show the internal state of building.
Based on the hybrid inference system described above, we develop a service that could be monitored by the building manager. On the screen, the administrator can grasp each internal state that inferred from data in the building. As shown in Fig. 10, based on the CO2, temperature, humidity, the air quality of the room, the occupancy status, and the thermal comfort index information of the each room are implemented through the ICMBS ontology. The service is con gured to extract and monitor a speci c state of the space using the SPARQL query. In this sevice, various kinds of sensor data are visualized so that the building manager can monitor the entire state at a glance.
The user can select the desired energy data characteristics(Occupancy, CO2, Temperature, Humidity, PMV, PPD, Energy Loss) and shows by colors
Show energy data of selected room (CO2(ppm), Temperature(℃), Humidity(%), number of people, PMV, PPD(%))
In this paper, we have studied intelligent building which prevents energy waste by hybrid inference based on machine learning and the semantic ontology. However, it does not directly control HVAC and heating/cooling facilities inside the building. In the future, we plan to expand the study to verify the e ect of intelligent inference to control the inside of the building by using IoT technology. In addition, we plan to study intelligent building optimized for energy saving by utilizing neural network and reinforcement learning based on historical data.
Outdoor temperature and Season A button that allows the user to select only the layer the user wants Room number and status
This work was supported by Institute for Information and communications Technology Promotion(IITP) grant funded by the Korea government(MSIT) (No.2015-0-00274,Development of Smart Mediator for Mashup Service and Information Sharing among ICBMS Platform).