Analysis of Office Rooms Energy Consumption Data in Respect to Meteorological and Direct Sun Exposure Conditions

Abstract

There are many papers concerning the consumption of energy in different buildings. Most describe residential buildings, with only a few about office- or public service buildings. Few articles showcase the use of energy consumption in specific rooms of a building, directed in different geographical directions. On the other hand, many publications present methods, such as machine learning or AI, for building energy management and prediction of its consumption. These methods have limitations and represent a certain level of uncertainty. In order to compare energy consumption of different rooms, the measurements of particular building-room parameters were collected and analyzed. The obtained results showcase the effect of room location, regarding geographical directions, for the consumption of energy for heating. For south-exposed rooms, due to sun radiation, it is possible to switch heating off completely, and even overheating of 3 °C above the 22 °C temperature set point occurs. The impact of the sun radiation for rooms with a window directed east or west reached about 1 °C and lasts for a few hours before noon for the east, and until late afternoon for the west.

1. Introduction

The aim of the paper is the development and verification of building energy consumption data collection and preliminary analysis methodology towards elaboration of algorithms for effective management of energy consumption. Electric use real measurements were utilized as the data source. The data was collected in 1-min intervals in selected office building rooms every day of 2020. The research subject was an office building named Gliwickie Ogrody Nauki of the KAMSOFT S.A. company located in Gliwice, Poland. The data regarding energy use of AC systems and heating for selected rooms located on different sides of the building in combination to meteorological data from the year period was collected. The research problem of the paper concerns evaluation of the building energy and meteorological data availability and suitability in respect of data mining-based algorithms development for building energy management. The further research problem components were whether it is possible to determine dependencies among the data by statistical analysis in selected time periods, as well as extracting building rooms energy consumption prediction rules based on the collected data. As it comes from the following state-of-the-art analysis and building energy consumption analysis, prediction and management are important scientific and economic issues.

Energetics is a basic element of national economics, and a warranty of strategic security of the country and the requirement for achieving balanced economic growth. Production and consumption of energy in Poland is ever growing; however, it requires constant improvements regarding effectiveness of energy use, the level of renewable energy use, energy saving, and CO₂ emission reductions [1]. To endure global climate change, emission reduction and energy saving is key. To reduce the influence of energy and the environment caused by buildings, in recent years many experts created studies on energy use and published their results. In many articles it has been shown that prognostics of energy use by buildings is an important step in solving various engineering problems [2,3]. It should be highlighted, however, that the issue of predicting energy use inside a building is dependent on many variables and its precise description is complicated. However, few works present measurement data created for real objects, especially office buildings. It is caused by the difficulty of conducting measurements over a long period of time and creating proper conditions for those measurements, for example disabling rooms from use. The behavior of users inside buildings is an important aspect of energy use [4]. In office buildings rooms are almost always fully occupied by users during the day, week, month, year. For example, Mahdavi et al. [5] compiled the results from 48 offices of three university buildings in Austria, consisting of 41 single person rooms and seven multi person rooms. The results showed that average use of those offices rarely exceeds 60% going by calculated 24-h long usage probability profiles. Another example is an office building located in Singapore, Peng et al. [6]. However, the use rate is not the only variable that influences electrical energy use in office buildings. Studies conducted by Masoso et al. [7] have confirmed that the energy used in free time is higher than energy used during work—56% and 44% respectively. It is caused primarily by keeping track of the movement of office equipment to the end of every day, no matter the usage rate. Nguyen et al. [8] also indicated that the behavior of users, who do not have energetic awareness, has caused an average of 30% higher energy use in buildings.

The pandemic situation has allowed us to conduct electricity use measurements inside an office building over the period of 1 year. This will allow us to not only optimize energy use inside the building, but also will allow us to propose energy saving solutions and to showcase strategies of controlled growth for cities [9]. Optimization and managing energy use require full understanding of building efficiency. It should foremost define the use of buildings and end users of various forms of power. Studies on end use of buildings show that they can be divided into six parts, including heating systems, ventilation, air conditioning (HVAC), lighting, and electrical outlets, and for special use such as in elevators, kitchens, and service rooms. [10] Based on multiple studies including those highlighted before, heating, ventilation, and air conditioning (HVAC) are responsible for most (40%) of the energy use inside a building [11].

2. Data Analysis Methodology

2.1. Case Building Description

The office building Gliwickie Ogrody Nauki has been built for the terrain department of the company KAMSOFT S.A. localized in Gliwice (GPS 50.28047058207462, 18.685625245420002), view of the building shown on Figure 1.

The building was designed and finished according to the ordinance of the infrastructure minister from 12 April 2002 regarding technical requirements which buildings need to comply with, and its positioning based on art. Seven resolution 2.1 images from 7 July 1994—building law [12] describe the classifications for fire safety and human safety.

It building is two-floor building with a basement of a total height of 12 m and total surface area of approx. 9500 m². Main building functions are:

• Office space for 328 office workers (41 rooms).
• Training and lecture hall for 250 users.
• Server room.
• Technical rooms necessary for building function.
• Employee parking lot and for other users with 166 spaces.

External dimensions of the main buildings is shown in

Table 1. External dimensions of the main buildings.

External Dimensions of the Building	External Dimensions of the Roof	High of the Building with Overhang	Maximum High with Lights
58.32 m × 58.32 m	61.20 m × 61.20 m	9.00 m	10.70 m

.

The building was equipped with the following electrical and low power installations:

• SS6 structural network, category 6.
• KD control access.
• SAP fire signaling.
• CCTV television monitoring.
• ODM smoke removal.
• IEL electrical.
• KNX intelligent building automation.
• SUGG constant fire extinguishing system using gas.
• DCR data center processing center with power of 500 kW.

The building had proecological features such as good heat insulation (EPS 70-040 14 to 16 cm thick) encompassing all external walls, sound absorbing insulation, and is also equipped with a heat capture systems inside the ventilation systems, installed in required areas of the building. All office space was lighted equally with the use of windows, each with dimensions of 150 × 225 cm, with a glass surface of around 2.70 m². Windows were made of aluminum profiles filled with flat compound glass, clear with a U value of 1.1 W/m²K.

The heating installation was designed using water pump heating in a two-pipe trio system with working parameters of 70/50 °C. Floor heating was used inside the office spaces (five circuits with heating parameters of t_z/t_p = 50/40 °C). Total heating requirements of the building equal Q = 410 kW. For floor heating with electrothermal splitters were used (connected with KNX automation) and flow meters. Regulation of heating medium temperature was kept through a mixing system controlled by a heating node.

Office rooms were equipped with fan-coil which were used for air cooling during summer. Fan-coils were used to keep temperature parameters at 23 °C. Cassette fan-coils with two-pipe systems were used. The devices were steered through KNX automation, each fan-coil was equipped with three-way valves mixed with servos regulating the valves and cut off valves on input and output. Power during summer was supplied by ice water with temperatures 8/14 °C. The cooling source for the fan-coil systems was a cooling aggregate with a cooling power of Q = 172 kW, situated in an underground facility outside the building.

Table 2. Most important building parameter.

Parameter	Symbol	Value	Unit
Heated area	Af	8380.6	m2
Heated cubature	Ve	29,965.4	m3
Shape factor	Af/Ve	0.303	m−1
Thermal capacity	Cm	1,996,718	kJ/K
Ventilation heat transfer coefficient	Hve	1511.84	W/K
Annual unit energy demand for heating and ventilation	QH/Af	39.4	MJ/m2

shows the most important building parameter.

2.2. Localization of Rooms and Localization of the Weather Station

Eight office rooms were selected for the research, differing only in their location, as shown in Figure 2.

Inside

Table 3. Parameters of selected rooms.

Room Number	Area, m2	Volume, m3	Direction	Number of People
104	42.06	126.8	S	0
123	42.06	126.8	E	0
124	42.06	126.8	E	0
125	42.06	126.8	E	0
138	42.06	126.8	W	0
139	42.06	126.8	W	0
140	42.06	126.8	W	0
220	42.06	126.8	N	0

the parameters of all analyzed rooms and their position regarding to geographical directions.

2.3. Building Energy Consumption Forecasting

Building energy utilization was influenced by the building envelope, heating, ventilation, and air conditioning (HVAC), occupants’ behavior and lighting. The building energy forecasting can be sorted into six categories: heating energy, cooling energy, heating and cooling energy, whole-building energy, and others [13]. In this research, energy consumption is studied from the electrical demand approach.

Table 4. Parameters of selected rooms.

Room Number	ΦT, W	ΦV,min, W	ΦV,inf, W	Φ, W	ΦHLmax, W
104	1077	2385	341	3462	4154
123	1212	2415	345	4154	4352
124	1213	2415	345	3627	4354
125	1213	2415	345	4352	4354
138	1217	2415	345	3628	4358
139	1213	2415	345	4354	4354
140	1213	2415	345	3628	4354
220	1303	2385	341	4354	4427

where: Φ_T—heat loss through transfer, Φ_V,min—heat loss through minimal ventilation, Φ_V,inf—heat loss through infiltration, Φ—total projected heat loss, Φ_HLmax—maximum heating strength at 100% valve opening is 20% higher than the projected heat use.

summarizes the heat demand and losses for the analyzed office spaces.

2.4. Meteorological Parameters

External environment parameters were monitored using an outdoor weather station Meteodata 140 S KNX (Figure 3). The ambient temperature of the device operation ranged from 20 °C to 55 °C, although its sensor could register temperatures in wider range from −30 °C to 60 °C. The device could also measure the wind speed ranging 2–30 m/s, as well as register the brightness from 1 to 100,000 lx. It was also equipped with a heated rain sensor, which could detect the precipitation. The above-mentioned weather station typically consumed 0.7 W, but maximal energy consumption could achieve 5.5 W. The protection rating of this device was IP 44.

This data was used for automatic control of window vent flaps placed on the roof. In the case of main hall heating, to limit AC costs during the night when the temperature was lower outside than inside. The rain sensor was integrated with the ventilation system of the main hall and closed the ventilation flaps in case of rainfall. The station was equipped with four threshold canals and connected with external KNX sensors and six logic channels.

2.5. Data Description

Data from the KNX automation installation was collected through a KNX ISE gate connected to the buildings main line and written down in daily files on an SD storage card in the form of YYYY_MM_DD_TP1.xml files. A schematic of this is shown on Figure 4.

The gate was connected to the local building structural network. Access to the data collected in the SD memory was conducted through the internal company network, and through an external VPN connection via a public network. All connections were encrypted. As a result, access to this data was conducted from different areas of the building working in a single inside network from any place in the world.

An additional function of the gate was remote service, diagnostics, and KNX system configuration installed inside the building through ETS programming. Through the visualization server gate, it was possible to configure work status of the building and to edit settings of work-spaces, and to visualize those changes.

Security connection to the gate was created with the use of internal networks or external VPN connections. This data was saved on a computer with the following specifications: Intel i5-760CK CPU 3.9 GHz 16 GB RAM using the Windows Server 2019 Standard equipped with SSD NVMe 250 GB, 250 GB drives SSD 120 GB, 250 GB and a 1TB HDD. To import the data KNX Data extractor, a C# program was prepared. Through this, importing, decoding, and data was saved to an MS SQL server.

The database was located on a NAS QNAP TS-451+ server equipped with a quad core 64 bit intel processor with a QTS 4.2 system, two WD Red 3.5″ 4TB SATA 600 MB/s 64 Cache drives and with an extension frame QNAP TR-004 were equipped with two sets of WD Red 3.5″ 4TB SATA 600 MB/s 64 Cache drives). All drives worked in RAID 1. Volumes inside the drives were encrypted and additionally saved with snapshots. Figure 5 shows a KNX Data Extractor block diagram.

KNX data processing required reading telegrams which are the core of data transmission in the system. Gateway collected information and stored it in hexadecimal format in XML files, which had to be decoded for further analysis. For this purpose, a program was developed that allows the user to read these files and provide data integrity as they grow.

The application read the files by mapping their structure using XML tags: in order to speed up this process it used multithreading. Then from the selected tags the values of fields Date and Time (Timestamp), Service (Service), Telegram Frame Type (FrameFormat), and Data (RawData) were read.

Based on the length of the hexadecimal data, a decoding method was chosen—in the above solution, only telegrams with the right number of fields were taken into account: 24, 22 and 20 i.e., 48, 44, and. 40 bytes, respectively. Other telegrams that were analyzed did not have data that were important for the purpose of the study—for example, communication between couplers. The following properties were retrieved from the fields: telegram type (TelegramType), message code (MessageCode), EIB control field (EIBControlField), source address (SourceAddress), group address (GroupAddress), data length (DataLength), TPCI/APCI, data (Data), and checksum.

The individual fields had a variable number of bits in their properties, e.g., a group address with a length of 2 bytes was divided into 5 bits for the first part of the address, 3 bits for the next part, and the remaining 8 bits for the least significant part. The data field also had a variable length, which was indicated in the last bits of the DataLength field and ranges from 1 bit to 2 bytes. The read addresses were used to recognize the device type in order to correctly read the data values, assign the unit, and determine the device type/name. The data was decoded with respect to the data characteristics (DataPoints) into binary (Boolean), integer or floating point (Float) values. The final step was to save the record in the database and mark the file as loaded and decoded.

The whole building data was collected from 15 August 2020 to 14 August 2021, thus the data collection database contained above 72,000,000 total encoded telegrams (lines). For daily analysis of rooms, approximately 19,000 lines were saved. In studies over the period of 30 days (from 8 February to 10 March 2021) the analyzed number of records was approx. 600,000. On

Table 5. Sample data structure.

TID	Expr1	Group Address	Service	Frame Format	Data	Device Name	Source Address	Length	Day	FileName	Processed	RawData	Raw Data Length	Expr2	TimestampS
28075563	28075563	12/5/16	L_Busmon.ind	CommonEmi	21.9	124 office room temperature reading	1.4.44	2 bytes	18.02.2021	2021_02_18_TP1.xml	1	2B090301070604B1DEFC21BC142C6510D300800C4716	44	12/5/16	2021-02-18 T00:09:09.108Z
28075055	28075055	12/5/23			22.6	140 office room temperature reading	1.5.44	2 bytes	16.02.2021	2021_02_16_TP1.xml	1	2B0903010306044C91AEF3BC152C6517D300800C6A3D	44	12/5/23	2021-02-16 T00:07:04.372Z
28074624	28074624	4/3/12			1.568627	125 office space heating	1.4.51	1 byte	18.02.2021	2021_02_18_TP1.xml	1	2B090301030604AC69A620BC1433230CD20080041D	42	4/3/12	2021-02-18 T00:08:08.059Z
28074601	28074601	12/5/9			23	104 office room temperature reading	1.3.25	2 bytes	10.02.2021	2021_02_10_TP1.xml	1	2B090301030604471D5503BC13196509D300800C7E04	44	12/5/9	2021-02-10 T00:08:44.954Z
28074448	28074448	4/3/10			6.27451	123 office space heating	1.4.37	1 byte	19.02.2021	2021_02_19_TP1.xml	1	2B090301050604DC242897BC1425230AD200801019	42	4/3/10	2021-02-19 T00:08:37.740Z
28074130	28074130	12/5/9			23.2	104 office room temperature reading	1.3.25	2 bytes	18.02.2021	2021_02_18_TP1.xml	1	2B090301050604A8FA4B03BC13196509D300800C88F2	44	12/5/9	2021-02-18 T00:07:29.642Z
28073724	28073724	4/3/11			3.921569	124 office space heating	1.4.44	1 byte	20.02.2021	2021_02_20_TP1.xml	1	2B09030104060404C68801BC142C230BD200800A0B	42	4/3/11	2021-02-20 T00:07:48.079Z
28073472	28073472	12/5/9			22.5	104 office room temperature reading	1.3.25	2 bytes	12.02.2021	2021_02_12_TP1.xml	1	2B090301060604973570E6BC13196509D300800C651F	44	12/5/9	2021-02-12 T00:06:52.498Z
28073440	28073440	12/5/22			22.3	139 office room temperature reading	1.5.37	2 bytes	16.02.2021	2021_02_16_TP1.xml	1	2B0903010506044EEE1F88BC15256516D300800C5B04	44	12/5/22	2021-02-16 T00:07:30.780Z

where: TID—identifier of the decoded telegram; Expr1—verification of the TID number and its correctness; GroupAddress—group address; Service—name of the service sending the data; FrameFormat—telegram gateway format; Date—temperature, on/off, percentages etc.; DeviceName—device (source) name; SourceAddress—source address, Length—type of the data field of the telegram frame; Day—date of creating the telegram; FileName—daily xml data file; Processed—fag which means the processing of a telegram, RawData—telegram saved in hexadecimal form, RawDataLength—telegram length, Expr2—duplicate group address used to recognize the type of data sent by the sensor, TimestampS—time in the format Year-Month-Day T Hour: Minute: Second.Thous SecondsZ.

an exemplary data structure system is shown.

Grouping of such a large amount of data requires appropriate filtering. The basic filtering parameter, a side room number were physical addresses of devices (Table 5 Column 8—SourceAdress), and group addresses of measured parameters (Table 5 Column 3—GroupAdress). Inside

Table 6. Physical addresses.

Room Number	Sensor Taster 2fach + Schneider Electric Industries SAS MTN612-04xx	Sensor Weather Stations Theben AG Meteodata 140	Actuator Heating Fan Coil Theben AG FCA1	Actuator Theben AG RMG8S
104	1.3.25	1.2.2	1.3.22	1.3.21
123	1.4.37	1.2.2	1.4.34	1.4.33
124	1.4.44	1.2.2	1.4.41	1.4.40
125	1.4.51	1.2.2	1.4.48	1.4.47
138	1.5.30	1.2.2	1.5.27	1.5.26
139	1.5.37	1.2.2	1.5.34	1.5.33
140	1.5.44	1.2.2	1.5.41	1.5.40
220	1.10.14	1.2.2	1.10.11	1.10.10

physical addresses of all used devices are shown, and in

Table 7. Group addresses.

Room Number	Temperature Set Point 23 °C	Heating Valve Position	Temperature Reading	Insolation
104	12/5/61	3/3/4	12/5/9	-
123	12/5/62	4/3/10	12/5/15	-
124	12/5/63	4/3/11	12/5/16	-
125	12/5/64	4/3/12	12/5/17	-
138	12/5/65	5/3/8	12/5/21	-
139	12/5/66	5/3/9	12/5/22	-
140	12/5/67	5/3/10	12/5/23	-
220	12/5/68	10/3/4	12/5/48	-
Weather station	-	-	12/7/2	12/7/1

group addresses.

3. Results and Discussions

The aim of the chapter is to evaluate the collected data in respect to possible estimation and prediction of whole building energy consumption as well as its distribution for particular rooms. It is assumed that there are two types of energy source for most rooms. The main energy source was heating controlled by the valve, whereas the other could be a window exposed to solar radiation. The point is to determine if there is a relation between room internal temperature and insolation of the building. The data for that quantity has been collected in a single place and there is no information concerning this parameter in respect to a particular cardinal direction.

There were four rooms selected each representing a single geographical direction. For eastern and western location, the sun exposure time was similar for both the selected rooms.

The analysis of room energy parameters data was performed in two stages. First, the data concerning the whole measurement period covering about a month was elaborated. For each day of the analyzed period, an average of four parameters was calculated. External temperature and insolation were the parameters common for all rooms, whereas internal temperature and heating valve percentage were individual for each of the rooms. The above-mentioned results are depicted in Figure 6, Figure 7, Figure 8 and Figure 9, respectively.

The second stage of the performed analysis concerned comparison of correlation between heating of the room and changes of internal temperature during night hours as well as during 24 h including night and day. The detailed results are discussed in further paragraph of the paper.

In Figure 9 there is no heating valve position waveform shown. This is due to switching off the heating in this room for the measurement period.

Basing on the data presented in Figure 6, Figure 7, Figure 8 and Figure 9 the statistical parameters of heating were calculated and compiled in

Table 8. Heating statistic parameters collected basing on daily sum of samples of valve open percentage.

Parameter	Heat. Sum	Heat. Avg.	Heat. Std. Dev.	In. Temp. Avg.	In. Temp. Std. Dev.
Room 220 (N)	89.12	3.07	0.70	22.09	0.05
Room 124 (E)	98.22	3.39	1.26	21.82	0.22
Room 140 (W)	81.43	2.81	0.77	22.23	0.08
Room 104 (S)	0	0	0	23.74	1.38

.

According to the calculated sum of daily average heating valve position samples, it could be noted that room 140, located on the west, showed the lowest demand for energy, whilst the average temperature of this room was slightly higher than the room 220 which was located in the northern part of the building. Simultaneously, the standard deviations of heating valve position and internal temperature of the described rooms were very close; however, in a room with a window facing the west, the average temperature was minimally higher.

Therefore, in room a with a window exposed to sun radiation, even only for a few afternoon hours, it was possible to achieve lower energy demand during the sunny days, and the direct sun radiation was a reason for the increase in the average room temperature during the observed month.

This is due to the fact that the eastern wall of the building was not shielded from the wind in any way, which may have resulted in faster cooling of this side of the building despite identical insulation on all external walls. These conditions were reflected in the higher consumption of heating energy for this room and in the greater fluctuations in the room internal temperature, shown as the standard deviation.

The statistics presented in Table 7 for room 104 prove that a sufficient amount of thermal energy coming from the sun exposed window, providing an appropriate room internal temperature or even exceeding it. In both parameters, average room temperature and standard deviation of this quantity could be noticed. Among all analyzed rooms located on different sides of the building, the above-mentioned parameters values were highest for south side of the building.

As has been mentioned, the second stage of the result analysis concerned the assessment of influence of sun radiation on room internal temperature for different cardinal directions. For the purpose of the analysis, a 24-h insolation measurement period was selected. Similarly to the previous stage, the hourly averages of particular parameters were calculated. The results of room and environmental thermal parameters for 24-h periods are presented in Figure 10, Figure 11, Figure 12 and Figure 13, respectively.

Analyzing data for the representative rooms shows a relationship between insolation and an increase in temperature in a relatively simple way. This is clear for room 104, which had a window facing south, and the heating switched off, as shown in Figure 13. For other rooms, the influence of both energy factors could be observed. It should be noted that control of heating also resulted from changes in the outside temperature. It is therefore more difficult to estimate the contribution of individual factors to the effect on the room temperature changes.

The method of assessing the impact of a given energy source on the room temperature was the determination of the correlation coefficient between the variables describing these quantities, available in the collected data. An important issue was the selection of an appropriate period of data analysis leading to the minimization of the share of the second factor of energy generation for a given room. Each room was characterized by a certain thermal inertia. Moreover, for each room, the heat losses resulting from the building insulation depended, in direct proportion, on the outside temperature. Therefore, it is possible to determine the sole effect of heating on the increase in room temperature during the night hours; however, it should be taken into account that the energy demand was higher due to the lower outside temperature compared to the daylight hours. Thus, during the exposure of a room to sunlight, the room temperature was influenced by both factors, while for rooms in which windows were directly exposed to sunlight, the impact of this type of energy source should be more noticeable. Due to the inertia of the heating system, in these rooms the temperature exceeded the set value.

Considering the above-mentioned conditions, the following principles of data analysis were proposed:

The twenty-four-hour period was divided into two parts. For the first 12 h, heating was the only energy source for a particular room. Thus, correlation coefficient between the hourly average heating valve position and the room’s internal temperature was calculated. That calculations were made for all of three rooms, excluding room 104, where heating was switched off. Then, analogous calculations were made for the entire 24-h period. The achieved correlation coefficients are presented in

Table 9. Correlation coefficients between valve position and current value of room internal temperature collected during night as well as night-and-day period.

Parameter	Correlation between Heat. and In. and Temp (Night)	Correlation between Heat. and In. Temp (Night and Day)
Room 220 (N)	−0.96	−0.95
Room 124 (E)	−0.92	−0.87
Room 140 (W)	−0.94	−0.86
Room 104 (S)	-	-

.

High absolute values of the correlation coefficients indicate the existence of a relationship between the analyzed quantities, while a negative value of each of these coefficients indicates that the relationship is inversely proportional. This is consistent with a typical room temperature control algorithm. When the room temperature reaches the set point, the heating is turned off, and if the temperature drops below the minimum value, the heating is turned on.

On the basis of the obtained values of correlation coefficients, it can be noticed that for the room 220, located in the northern part of the building, impact of heating on room temperature remained at the same level at night as well as during the whole 24-h period. In the case of other rooms (124 and 140), there was a lower correlation between the degree of opening of the heating valve and the temperature in the room during the whole 24-h period, compared to the night hours only. The windows of these rooms face east and west, respectively. It can therefore be concluded that the temperature in the room over the entire 24-h period did not depend solely on the degree of opening of the heating valve in the room, as it did at night when sunlight had no effect on the temperature.

Due to the inability of referring to the position of the heating valve in the room located on the southern side, for room 104 and other rooms exposed to solar radiation (124 and 140), an additional analysis was carried out. The results of it are presented in Figure 14.

Figure 14 shows the waveforms of the internal temperature of particular rooms, which were directly exposed to solar radiation, referring to the hourly recorded average value of insolation for all cardinal directions. The figure shows that the temperature in the room with the window facing south was higher than in the others, despite the heating being turned off. In addition, an increase in temperature could be noticed during the hours of the highest insolation, despite the presence of cloudiness. Cloudiness was indicated by the lower value of insolation at noon than 2 h earlier. A similar situation could be observed in the case of rooms exposed on the east and on the west; however, the insolation impact on the room’s internal temperature was significantly lower than in the previously described room, while the maxima of internal temperatures were shifted earlier and later, respectively, depending on the sun’s position in the sky.

It is known that energy use inside a building is dependent on many variables such as weather, heat insulation of the building outside, user behavior, and geographical directional placement. Moreover, electricity used by buildings is characterized by obvious seasonal patterns and uncertainties [14].

At the same time, many prognostic methods for electricity use have been proposed, such as Wei et al. [10], Li et al. [15], and others [16,17,18,19,20,21,22,23,24,25,26,27].

General methods of prediction of energy use in buildings can be divided into three categories:

• White box methods, based on physical parameters of the building and using simulation systems such a DOE-2, EnergyPlus, and DeST [28]. The advantage of this method is the fact that it does not require historical data regarding energy consumption. However, the disadvantage of the white box method is the fact that it cannot be calibrated using real historical data [15].
• Grey box methods which related to a statistical prediction method which combines historical data with physical information about the building [29].
• Black box methods, also based on data. This method is largely based on large amounts of historical data regarding buildings. It uses precise arguments of mathematical input for data prediction: When historical data is sufficient and precise this method is preferable due to its accuracy.

In recent years, many scientists have used black box methods, based on their own data, for predicting power use inside buildings. Data-driven methods skip the process of physics modelling [30]. There exist three main methods that use models based on data to create prognosis of time periods, which include statistical analysis, machine learning, and deep learning. The most common method of statistical analysis is ARIMA. Because the ARIMA method largely depends on historical data, if the data is highly variable, they are not the right choice for predicting long term time periods [31]. Among methods of machine learning in predicting time periods the most common are k-near neighbors (KNN), artificial neural networks (ANN), and supported vector machines (SVM) [32,33,34,35,36,37,38,39,40,41,42]. All proposed models use as entry data historical measurements of buildings use. Because of this, in all the solutions, precision is largely dependent on the reliability and accuracy of data. In our opinion, precise prognosis of energy use inside an office building requires additionally considering weather and positioning of the rooms within the building related to geographical direction.

Basing on the collected data, it can be concluded that a window exposed to solar radiation may be an additional source of thermal energy for selected rooms. Currently, this factor is not taken into account in the building heating system, and its participation is coincidental. In addition, it should be noted that the collected data also allow for the development of accurate thermal models for particular rooms. These models could also take into account the influence of the main thermal source as an impact of additional energy coming from an external window exposed to the sun radiation. Predictive tools such as ARIMA models and artificial neural networks can be used to model and analyze the impact of the described energy sources on room temperature changes. Applying accurate thermal models of rooms, it is possible to avoid exceeding the set temperature in the room, and by basing the forecast on the outside temperature and insolation, an attempt could be made to lower the energy cost by reducing the heat inflow from the heating system during solar exposition of the room window.

The key element which influences the energy consumption of the building is the level of the valve opening of the heating system. One of the important factors which determines the valve level is the insolation. Therefore, it is related to the geographical location of the building, or, more precisely, to the location of the windows of the building which expose the room for the solar radiation. In order to determine how much it influences energy consumption, we decided to estimate key factors influencing the position of the valve. In order to estimate the key factors, we performed an experiment in which we built a prediction model aimed at predicting the level of valve opening based on parameters such as outdoor temperature, indoor temperature, the hour of the day, and insolation of identical rooms with identical settings but located on different sides of the building: north, west, east, and south.

The experiments were conducted using RapidMiner software and consisted of training random forest regression [43] algorithm to estimate the level of throttle opening. In the experiments, we performed a 10-fold cross-validation test [44] to assess the performance of the prediction model and optimize the size of the forest. The use of random forest meant we could extract feature importance indicators as factors, determining which of the input parameters had the highest impact on the valve opening. The experiments were performed independently for three rooms, namely 124, 140, and 220, located on the eastern, western, and northern side of the building, respectively. We were unable to perform the experiment for room 104 because the valve opening was always set to 0. The input data consisted of all four parameters described above collected for one month. The only difference between these rooms was the geographic location (data was recorded during the COVID-19 pandemic when there was no one in the office). Therefore, it can be assumed that all the differences between the feature importance factors were related to the room location.

The obtained results of the described experiment, based on the preliminary prediction model, confirmed previously conducted statistical analysis of the data for particular rooms. However, for more precise results of prediction, a more sophisticated model is required. Thus, development of such a model is a proposed subject of future research.

4. Conclusions

From the analyzed data, we concluded that there is no direct or indirect influence of the insolation on internal temperature for office rooms located on the northern side of the model building, whereas the impact of insolation for south-located rooms of the building was significant. Based on the recorded temperature of the room 104, the window of which was exposed on the south, the average temperature of the room was more than 1.5 °C higher than in rooms located in northern, eastern, and western parts of the building, even though the heating in the room was switched off during the whole measurement period. The area, volume, and a window size for all of the analyzed rooms were the same. Simultaneously, the dispersion of internal temperature was from 6- up to 27-times higher than in other rooms. As there was no heating in room 104, the temperature set point could not be considered, but in comparison to other rooms’ temperature set points at 22 °C, up to 3 °C overheating occurred because of sun radiation. In case of east- and west-directed rooms it was noticed that the period of the insolation influence is shorter than in south directed room. Moreover, the increase of internal temperature caused by sun radiation was up to 1 °C, beginning at about 10 a.m. with a peak about 1 p.m. for the eastern room, whereas the temperature in the western room begin to rise at 3 p.m. achieving a peak at about 5 p.m.

The novelty of the described research results from taking into consideration only external conditions, such as outside temperature and insolation of the building, with a complete lack of impact of persons working in rooms and moving between them. Such an approach was possible due to pandemic conditions, causing almost the whole personnel of the company to work remotely. Thus, the collected data and extracted rules could be a reference model for future research performed in typical building operation conditions. Although there are many of the papers concerning similar issues, there is a lack of publications referring to the abovementioned situation. Moreover, what makes this research exceptional to previous publications is the location of the building with respect to the moderate climate zone of middle eastern Europe and to the detached character of the office building. It is also worth mentioning that the building was relatively low, with quite a large area of the roof covered with photovoltaic panels.

This performed research could be helpful in the development of building energy consumption data collection and analysis scenarios, and for applications of data mining and machine learning methods in research of medium- and long-term energy use prognosis.

The conclusions drawn could facilitate future microscale changes of energy use for a particular building through appropriate retrofitting and the inclusion of renewable energy technologies. It also paves an avenue to explore potential in macro-scale energy-reduction with consideration of customer demands. All these will be useful to establish a better long-term strategy for urban sustainability.