2.1 Simulation of control algorithms

The first step in developing suitable control algorithms was to identify all the parameters that may influence the performance of an MM control algorithm. These included operative temperature, occupancy density and external conditions. For this research, change-over MM systems were used. It has also been shown that ceiling fans, which are routinely used in Indian residences, increase the operating range of NV (Babich et al. 2017a).

To illustrate the use of the proposed control algorithms and simulation approach, a validation study was conducted using data from an experimental chamber in India. Finally, a demonstration two-bedroom case study was used to analyse the performance of the proposed control algorithms using co-simulations.

2.1.1 Description of the two-bedroom demonstration case

The applicability of the newly developed control algorithms to real buildings was assessed to predict their energy saving potential. For this purpose, a typical two-bedroom apartment block was selected based on Rawal & Shukla’s (2014) identification of the most typical residential layout (Figure 2) built using reinforced concrete floor plates, with brick and concrete block masonry walls. Two different types of building envelopes were considered in this study: one denoted as business as usual (BAU) and the other from the Energy Conservation Building Code (ECBC), with envelope properties presented in Table 1, and air infiltration rate of 1 ach (air change per hour) (Rawal & Shukla 2014). Three different thermal zones were analysed in the present study (bedroom 1, bedroom 2 and the kitchen/living room), while other zones (bathrooms, balcony and hall) were considered to be isothermal relative to the adjacent zones. The three-dimensional (3D) model shown in Figure 2 gives the location of the windows/dampers for the two-bedroom apartment.

Figure 2 

Floor plan for the two-bedroom apartment case study.

Note: Dimensions = mm.

Table 1

Building envelope properties.

	BAU ENVELOPE PROPERTIES	ECBC ENVELOPE PROPERTIES
Wall	18 mm cement plaster (inner most) 230 mm uninsulated brick wall 18 mm cement plaster (outer most) U-value: 1.722 W/m2K	18 mm cement plaster (inner most) 230 mm uninsulated brick wall with XPS insulation 18 mm cement plaster (outer most) brick wall U-value: 0.44 W/m2K
Roof/floor	12 mm cement plaster (inner most) 150 mm concrete roof 12 mm tiles (outer most) U-value: 2.942 W/m2K	12 mm cement plaster (inner most) 150 mm concrete roof with XPS insulation 12 mm tiles (outer most) U-value: 0.409 W/m2K
Window	6 mm single-glazed anodised aluminium frame without thermal breaks U-value: 5.8 W/m2K	Double-glazed windows with thermal break U-value: 3.3 W/m2K

Note: BAU = business as usual; ECBC = Energy Conservation Building Code; XPS = extruded polystyrene.

Sources: Bureau of Energy Efficiency (2009), GBPN (2014), Rawal & Shukla (2014).

2.1.3 Mixed-mode control algorithms

Several control algorithms were developed for the different thermal zones depending on the occupancy levels and time of day. Presented in the form of flow charts (Figure 3), these prioritise the use of passive techniques whenever ventilation or cooling is needed during occupied hours. When the internal air temperature is between the heating and cooling setpoint, the control algorithms operate the dampers and the windows to maintain acceptable internal conditions. Dampers have the highest priority to operate over the windows, and only when the dampers are 100% open are the windows enabled. The modulation of the windows or dampers is a dynamic process where the control algorithms use a linear relationship to calculate the appropriate opening factor for the opening so as to provide the required airflow rate.

The windows were restricted to 20% of their maximum openable area during the night-time to use the well-known benefits of night-time ventilation whilst also maintaining security.

When outdoor and indoor conditions do not favour the use of NV, the control algorithm activates the mechanical system to meet the cooling requirements. When the internal air temperature is below the heating setpoint, and the external air temperature is below the internal air temperature, the windows and dampers are kept closed and the mechanical heating systems are activated. Whenever the control algorithms activate the dampers/windows, mechanical systems are switched off automatically. All the algorithms provide the option to the occupants to override the built-in algorithm either by selecting the manual override function or by adjusting the setpoints to control a specific component. In cases where the network or power communication is lost for a prolonged period (30 min), all the ventilation and cooling systems will move to a position defined by the safety protocol.

Window control was performed in accordance with the required airflow rate for delivering cooling in response to internal heat gains rather than in response to the internal/external temperature difference, which is the common approach for window operation in EnergyPlus (Angelopoulos et al. 2018). Modulation of the windows was implemented to match the calculated required effective area for the given conditions. To represent windows and dampers in the simulation model, the airflow network module in EnergyPlus (DOE 2018) was used to simulate the indoor–outdoor air exchange.

It was assumed that the ceiling fan could operate at three fan speeds, resulting in internal air velocities of 0.6, 0.9 and 1.2 m/s (Babich et al. 2017b). In the scenarios where ceiling fans were modelled, the setpoint temperature for NV was increased based on the air velocity. To represent the operation of ceiling fans in EnergyPlus, the setpoint temperature for cooling was increased according to the fan speed. ASHRAE Standard-55 (ASHRAE 2013) has incorporated in its documentation the advantages of the use of a ceiling fan by increasing the acceptable range for operative air temperature by 1.2, 1.8 and 2.2°C for internal air velocities of 0.6, 0.9 and 1.2 m/s, respectively. The energy consumption of the ceiling fan was assumed to be 50 W (Babich et al. 2017b).

2.1.4 Description of the simulation approach

Dynamic thermal modelling (DTM) tools are designed to handle annual energy performance simulation rather than to assess an advanced control strategy for passive or active systems. To improve the accuracy of DTM tools when assessing the controls of mechanical systems, co-simulations, coupling DTS with other programs, can be used (Nouidui & Wetter 2014; Angelopoulos et al. 2018; Borkowski et al. 2018). A standardised method to facilitate the exchange of information between the two simulation tools known as the Functional Mock-up Interface (FMI) (MODELISAR 2017) was used.

When performing a co-simulation, one simulation tool is considered the master simulator handling the communication time step and the exchange of variables, and the other tools are the slaves. The current study used the DTM EnergyPlus v8.6.0 and Dymola v2017 FD01 (Dymola 2018), which is the commercial simulation environment based on the Modelica language, along with the FMI standard v1.0. Dymola was used as the master simulation tool, and EnergyPlus was the slave. Dymola was used to handle the communication time step and variable exchange information (Bastian et al. 2011).

To access all the components needed for the control algorithms, the use of the Energy Management System (EMS) module in EnergyPlus was used (DOE 2017). In EnergyPlus, the building envelopes, mechanical systems, occupancy schedules, and internal heat gains were created and modelled whilst the new control algorithms were developed in Dymola, allowing for simulation flexibility as the same control algorithms could be used with different models without changing the control logic (Angelopoulos et al. 2018). Internal air temperature, internal heat gains, occupancy schedule and weather conditions were transferred from EnergyPlus to Dymola (Figure 4). Dymola then runs the control algorithms and sends information to EnergyPlus about the operation mode, position of the windows, ceiling fan speed, mechanical fan speeds, operation of the mechanical ventilation unit, operation of the dehumidification unit and setpoint temperatures. As mentioned above, the ceiling fan speed information was converted into an adjusted cooling setpoint temperature in EnergyPlus.

The simulation time step in EnergyPlus was equal to the sampling time in Dymola (600 s, or 10 min), allowing synchronisation of the two simulation tools. Information exchange between the two simulation platforms occurred automatically in each time step and throughout the whole simulation period (one year). The small time lag in practice caused by the time required for the control algorithm to activate the windows/dampers and the time required by the BMS to send them the signal is not present during the co-simulations, so the signal is assumed to be transferred instantaneously.

2.1.5 Simulated scenarios for the validation analysis

For the validation study, average hourly data were used to compare the simulated results with data measured in the experimental chamber (see Section 2.3). Steady-state data were used for the validation study to examine the validity of the proposed control algorithms and the co-simulation technique. By selecting hourly data, it was feasible to examine a variety of different outdoor and indoor conditions and windows/dampers configurations.

To quantify the discrepancies in the results, two statistical indices were used to provide evidence of the validation method:

• Mean bias error (MBE), which captures the mean difference between the measured and the simulated data, and is a good indicator of the overall bias in the model (Coakley et al. 2014). A limitation of this method is that the positive and negative errors cancel each other out when summed, which might lead to positive bias compensating for negative bias:
  
  (1)

  $\text{MBE} (\%)= \frac{{{\displaystyle \sum }}_{i=1}^{N_P}\left(m_i-s_i\right)}{{{\displaystyle \sum }}_{i=1}^{N_P}\left(m_i\right)}$

  M1 \documentclass[10pt]{article} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \usepackage{pmc} \usepackage[Euler]{upgreek} \pagestyle{empty} \oddsidemargin -1.0in \begin{document} \[ {\rm{MBE}}\left(\% \right) = \frac{{\mathop \sum \nolimits_{i = 1}^{{N_P}} \left({{m_i} - {s_i}} \right)}}{{\mathop \sum \nolimits_{i = 1}^{{N_P}} \left({{m_i}} \right)}} \] \end{document}
  
  where mi is the measured data; si is the simulated data; and NP is the number of data points.
• Coefficient of variation of root mean square error (CVRMSE), which captures the error between the measured and simulated data without the cancellation effect described for MBE (Granderson & Price 2014):
  
  (2)

  $\text{CVRMSE} \left(\text{\%}\right)= \frac{\sqrt{\frac{{{\displaystyle \sum }}_{i=1}^{N_P}{\left(m_i-s_i\right)}^2}{N_P}}}{\overline{m}}$

  M2 \documentclass[10pt]{article} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \usepackage{pmc} \usepackage[Euler]{upgreek} \pagestyle{empty} \oddsidemargin -1.0in \begin{document} \[ {\rm{CVRMSE}}\left({\rm{\% }} \right) = \frac{{\sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^{{N_P}} {{\left({{m_i} - {s_i}} \right)}^2}}}{{{N_P}}}} }}{{\bar m}} \] \end{document}
  
  where mi, si and NP are as defined in equation (1); and m is the average of the measured data. CVRMSE was recommended by ASHRAE Guideline 14 (ASHRAE 2014) and previous studies (Coakley et al. 2014; Granderson & Price 2014). The simulation results were evaluated against the measured data from the experimental chamber.

To determine suitable conditions for the validation experiments, co-simulations were performed using the geometry of the experimental chamber and the weather file of Ahmedabad, and the control algorithms presented in Figure 3. Findings were analysed and clustered into groups based on outdoor air temperature. Outdoor air temperatures were grouped into 1°C intervals for each group; the number of hours for which the windows and/or dampers operated was counted. This allowed the calculation of the frequency that each scenario occurred. The cases with the highest occurrences were selected to be simulated in the experimental chamber (Table 3).

2.3 Experimental configuration and methods

Experimental measurements were conducted in a full-scale low energy characterisation chamber located at CEPT University, Ahmedabad. The facility consists of an outer chamber (10 × 8 × 8 m) and an inner chamber (5.05 × 3.95 × 3.00 m), which represent the external and internal environments, respectively. The inner chamber represents a typical Indian bedroom. Two air-handling units are installed, one per chamber. The outer chamber can maintain a wide range of conditions (10–45 ± 0.3°C air temperature, 15–95% ± 3% relative humidity—RH). The inner chamber is equipped with a motorised window (1.2 × 1.2 m) and four motorised vents (0.4 × 0.4 m). Building materials and properties are presented in Table 4.

A total of 13 flexible air temperature sensors were used, arranged using a 3 × 3 × 3 m grid (Figure 5). Four air temperature and RH sensors and six CO₂ sensors were installed and calibrated in an externally accredited laboratory. A CO₂ source used as a tracer gas was placed at the centre of the inner chamber to represent the internally generated CO₂. CO₂ sensors were placed in the centre of the low- and high-level vents and in the outer chamber.

Before each individual experiment was performed, the process comprised a ‘warming up and stabilisation’ period to reach a steady state (phase A), followed by the release of the CO₂ (phase B), the experimentation period (phase C) and finally the period required to cool the chamber in preparation for the next experiment (phase D). In phase B, the fan, window and AC were set up as per the configurations shown in Table 5, and all internal loads were maintained as per Table 3. The experiment used tracer gas-decay measurement techniques to estimate air exchange rates of the chamber under different scenarios (ASTM International 2011). The tracer gas-decay technique injects known concentrations of tracer gas into a space and solves the mass balance equations on the injection rates and measured concentrations. These techniques and their uncertainties are well documented in the literature, with the detailed methodology and governing equations provided for each technique (McWilliams 2003; Persily 1988, 2016). The CO₂ tracer flow was maintained to achieve 2000 ppm (parts per million) inside the indoor chamber. In phase C, the indoor conditions (temperature, RH and CO₂ level) were monitored for more than four hours (Figure 6). The experiment collected data from all the sensors are listed in Table 5.

3.1 Validation of the dynamic thermal model and control algorithm

Validation of the co-simulation method (Figure 7) shows a high level of agreement. For most scenarios, air changes per hour (ach) were predicted to be higher for the co-simulations relative to the measured values, although within the error margins. This was due to wind being taken into consideration in the simulations, whereas only buoyancy-driven cases were evaluated in the chamber, resulting in the prediction of lower ventilation rates. The wind was not removed from the weather files because the intention was to validate the co-simulation approach and then use the same model and inputs to predict the energy savings.

The highest impact of the wind occurred in scenarios 3 and 6 where the highest wind speeds were observed (5.9 and 7.2 m/s, respectively).

Simulations overpredicted the air change rate by approximately 5–11% compared with measurements, due to being dynamic and steady state, respectively. Parameters such as external air temperature and heat gains, etc. varied constantly over time, but for the determination of the results from the experiment, hourly averages were used. This further explains the small discrepancies in the results.

The analysis showed that the MBE and CVRMSE were 7.25% and 20.64%, respectively. Based on ASHRAE Guideline 14 (ASHRAE 2014), the acceptable limits for the hourly calibration of building energy simulation models were 10% and 30% for MBE and CVRMSE, respectively. Hence, both indices are within the proposed limits, providing evidence that the co-simulations provided acceptable results.

3.2 Predicted thermal comfort conditions

The use of more night-time NV has benefits because it can take advantage of the cooler outside air to cool the thermal mass of the building and possibly to mitigate the risk of uncomfortable internal conditions due to high internal air temperature during the next day. However, the use of night-time ventilation is not always favourable, mainly for safety and noise reasons. When a 20% night-time restriction was imposed for security reasons, mechanical cooling was used more during the day due to the systems’ inability to use the diurnal temperature difference. In cases without any restrictions, the algorithms calculated the opening area to be approximately 50–60% of their maximum allowable opening area during the night-time. This resulted in lower internal air temperatures, which favoured the use of NV during the daytime, as preconditioning had occurred. Similar window-operation patterns were predicted for all four climates.

For all four climatic conditions, the use of ceiling fans significantly increased the operation of the windows due to the higher setpoint temperatures for NV (see Section 2.3), leading to predictions of larger openable areas for windows/dampers compared with cases without ceiling fans. In Bangalore, with the most moderate climate compared with the other cities, the co-simulations predicted a more frequent operation of the windows/dampers throughout the year. On the other hand, in Chennai, with high outdoor air temperatures and RH, the co-simulations predicted that the windows/dampers operated less frequently, approximately 5–35%, compared with the other cities.

The hourly average zone internal air temperature was compared with the hourly comfort temperatures proposed by the IMAC model for 90% acceptability. Table 6 summarises the hours of the year that NV was able to provide ventilation/cooling and maintain the internal conditions within the comfort limits due to the controls. NV hours are the hours of the year that the co-simulations predicted exclusively NV/cooling mode. Comfort hours are the total hours that the internal operative temperature was within the comfort limits, provided by the IMAC thermal comfort model, during the NV operational mode.

The use of ceiling fans can result in 8–25% more use of NV compared with VCS 2 for all the cities studied, which is supported by the findings of Omrani et al. (2021). The higher percentages of thermal comfort were predicted in the city of Bangalore due to the moderate climatic conditions.

The most significant improvement from case VCS 1 of the predicted hours of NV was observed for the city of Chennai, with a warm and humid climate. The introduction of the dampers and ceiling fan, VCS 3, increased the predicted hours of NV by 1329 h (74%) compared with VCS 1, which aligns well with that predicted by others (e.g. Li & Chen 2021). Chennai has a high mean yearly outdoor air temperature, hence the use of only windows was unable to maintain a thermally comfortable internal environment. Dampers (VCS 2) and ceiling fans (VCS 3) used NV and increased the predicted hours for thermal comfort by 18%.

The use of stand-alone dehumidifier units leads to decreased NV due to the thermodynamic process of a dehumidifier. The humid air of the room enters the dehumidifier, where it is cooled to its dew point and releases its moisture. The dry air is then heated, due to latent heating, which is a natural process that occurs in the condenser. Also, the compressor releases heat, which can result in higher air temperature of the exhaust air compared with the inlet temperature. This results in the higher internal air temperature and reduced predicted hours of NV for VCS 4 compared with VCS 3.

Regardless of the climatic conditions or cooling system, the analysis showed that for approximately 60–75% of the hours when NV was used, the internal operative temperature was within the comfortable limits. In Bangalore, with moderate climatic conditions, the co-simulations predicted the highest percentage of comfort hours for both cooling systems, although these were outside the desirable limits. The use of windows/dampers alone could not provide the adequate provision of outside air. In addition, considering the strict comfort limits imposed by the IMAC model, the analysis showed that it is not feasible to rely only on the windows/dampers to achieve comfort conditions. The introduction of ceiling fans in the bedrooms and living rooms resulted in higher setpoint temperatures, suggesting that NV was selected for a wider range of external temperatures. This justifies the higher percentages of thermal comfort conditions predicted for VCS 3 over VCS 1 and VCS 2. The application of ceiling fans in the control algorithms increased the number of comfort hours for all cities and cooling systems. The most significant improvement was observed for both the moderate climate of Bangalore and the hot and dry climate of Ahmedabad. For no window night-time restriction, 90% acceptability was met for Bangalore and Ahmedabad. For Chennai and Delhi, up to 85% comfort hours were predicted, which is very close to the acceptability limit.

3.3 Energy savings for the ventilation and cooling strategies

The co-simulations were performed for the whole year, and the predictions of the total energy demand were compared against the base case, which was the fully mechanical mode case with no window operation. To calculate energy savings, all systems (mechanical heating/cooling, ceiling fans, humidifiers) were included.

The total energy for the base case (fully AC mode) scenario is shown in Figure 8 by the red bar, while the percentages show the energy savings for each scenario. For all VCSs, the difference between having the night-time window restriction and not is approximately 5–10%. As expected, when there is no restriction at the openings, the total energy demand is lower as the building thermal mass cools during night-time, relying less on mechanical cooling during the following day. However, the difference in the energy savings between these two cases is not significant, mainly because of the small effective window area (assumed to be 50% free area of the geometric area of the window). As expected, the co-simulations predicted that the highest energy savings can be achieved in Bangalore (Figure 8b), since the moderate climatic conditions favour the use of NV for a longer period compared with the rest of the cities.

The use of ceiling fans (VCS 3 and VCS 4) resulted in the highest percentages of energy savings for all cities: approximately 40–55% compared with the base Case (fully AC mode) scenario. This is comparative with the work of Omrani et al. (2021). The ventilation and cooling setpoints due to the air movement proved to be the most efficient measure to reduce the use of mechanical cooling systems. For instance, by including the ceiling fan as part of the control strategies, the co-simulations predicted an additional 20–25% energy savings compared with the cases without the ceiling fan (VCS 2). The positive effects of the use of the ceiling fans are more significant in warm and humid climatic conditions, such as Chennai, since the air movement increases the convective and evaporative heat exchange between the occupant and the environment. The convective heat exchange is driven by the temperature difference, whilst the evaporative heat exchange by the water vapour pressure difference. For the same temperature difference, the air motion has a greater effect on increasing thermal comfort in the humid climates compared with the drier climates. The occupants in warm and humid climates are more likely to use higher fan speeds to maintain the same levels of thermal comfort for the same temperature compared with the occupants in drier climates. This can also be reflected in the predictions of energy savings, since the co-simulations calculated that the inclusion of the ceiling fans can result in up to 25% additional energy savings for Chennai and Delhi, whilst for Bangalore or Ahmedabad the additional savings were up to 20%. Overall, the energy saving potential by using the ceiling fan is higher than the additional energy consumption by operating the ceiling fan. The ceiling fans consumed a fraction compared with overall energy consumption by the rest of the mechanical systems. Additionally, the study showed that the additional heat gains by the ceiling fans had a minimum impact on the overall indoor conditions.

Additionally, the use of stand-alone mechanical ventilation and dehumidifier units slightly decreased the energy saving potential for VCS 4 compared with VCS 3 (Figure 8), due to the thermodynamic process of a dehumidifier.