Adaptive and Predictive Control Method to Improve Energy Efficiency and Indoor Thermal Comfort for Intermittently Occupied Spaces

1. Introduction

2. Literature Review

3. Research Method

3.1. Building Concept

Research on the energy consumption intensity of buildings has been ongoing. Furthermore, data directly surveying the energy consumption of actual buildings and analyzing their characteristics based on a vast amount of statistical data is available. In particular, the U.S. Energy Information Administration’s Commercial Building Energy Consumption Survey (CBECS) report is arguably the most useful resource for this research. It categorizes buildings across the United States into 14 categories and examines their energy consumption by each category [31]. However, even with the CBECS report and other sources, there are certain buildings for which energy consumption is difficult to assess. Among these, research has been conducted on large-scaled buildings with a strong public character, such as airports, subway stations, and multi-purpose terminals with mixed uses [32~34].

This study aims to test a model to improve the indoor environment and energy consumption of spaces with unclear uses and low usability, which are commonly found in the aforementioned complex buildings. Therefore, the modeling of the architectural space follows the characteristics of typical small and medium-sized commercial spaces, but simulation tests will be conducted using a time schedule scenario with irregular occupancy times. Table 1. summarizes the characteristics of the space geometry, and Table 2. presents scenarios regarding space occupancy times.

Table 1.

Geometry information of a designed spaces

Table 2.

Scenario in use

3.2. Control Rule

To quantify the model’s energy consumption, the total energy use per unit time is calculated using a foundational thermodynamic balance equation. This equation accounts for the heat energy supplied by the HVAC system and the net heat transfer across the building envelopes [35,36]. By reformulating this process as the rate of change in the controlled indoor temperature over time, an essential differential equation is derived, which serves as the core formula for constructing the building’s thermal simulation module.

$\[ \frac{d{T}_{room }}{dt}=\frac{1}{m_{room }{C}_v}\mathrm{*}\left(\begin{array}{c}\left(\frac{T_{room}-T_{out }}{\frac{1}{h_{out }A}+\frac{D}{kA}+\frac{1}{h_{\in }A}}\right)\hfill \\ +\left({\dot{\mathrm{m}}}_{ht}{C}_p\left(T_{heater}-T_{room}\right)\right)\hfill \end{array}\right) \]$

(Eq. 1)

The PMV index, developed by Dr. P. O. Fanger and standardized by EN ISO 7730, serves as the primary metric for quantifying thermal comfort [35,36]. Its auxiliary indicator, the Predicted Percentage of Dissatisfied (PPD), is derived from the PMV equation, indicating the percentage of occupants expected to feel uncomfortable, ranging from 0% (near-satisfied) to 100% (fully dissatisfied). To calculate the PMV level within this simulation model, six core variables are input into a block module: dry bulb temperature (DBT), relative humidity (RH), indoor air speed (AS), mean radiant temperature (MRT), metabolic rate (MET), and clothing insulation (Clo) [37,38]. It is assumed that indoor the AS is 0.1m/s, the MRT is the same as the DBT with a 1 hour delay, the MET is a fixed value of 1.2 (light office work), and the Clo is 1.0 (ligh summer clothing).

$\[ PMV=3.155\left(0.303e^{-0.114M}+0.028\right)L \]$

(Eq. 2)

$\[ \begin{array}{c}L=q_{met, heat}-f_{cl}{h}_c\left(T_{cl}-T_a\right)\hfill \\ -f_{cl}{h}_r\left(T_{cl}-T_r\right)-156\left(W_{sk, req}-W_a\right)\hfill \\ -0.42\left(q_{met,heat}-18.43\right)\hfill \\ -0.00077M\left(93.2-T_a\right)\hfill \\ -2.78M\left(0.0365-W_a\right)\hfill \end{array} \]$

(Eq. 3)

where, M is the metabolic rate and L is the thermal load.

Then, the results of the three different control model are compared in terms of the level of energy use and the consistency of thermal comfort. In the case of energy use, the total sum of the amount of energy consumed for heating and cooling can be used because it is intuitive and uncontroversial. In the case of the PMV, since positive and negative values exist in the calculated result, it is difficult to evaluate the performance with a simple sums or average values, so the Root Mean Squared Error (RMSE) can be used as follows:

$\[ \mathrm{S}\mathrm{E}=\sqrt{\frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{P}\mathrm{r}edicted}_i-{Actual}_i\right)}^2}{\mathrm{N}}} \]$

(Eq. 4)

Through this method, the most comfortable state in the PMV value is ±0, which is used as the best fit model, and the distribution of the results of each model is compared as the indicator of maintaining consistency of thermal comfort.

The result from the thermal model and the process in input into the ANN learning model. In general, the structure of the ANN consists of fully connected multi-layer neural network as a multilayer perceptron. This type of network includes more than one layer of artificial neurons or nodes, which is used to solve more complex classification and regression model. The most typical structure is the three layers fully connected backpropagation model. The first layer consists of input neurons that send data on to the second layer as hidden layers. Then it, in turn, sends the output neurons as the third layer to get results from this neural networks. As described in Fig. 1., each neuron takes weighted inputs x₁, ... x_k multiplied by a weight term w_ai. Then, each result adds a bias term θ_b, then uses the result n_c as an input into an activation function g_d [39~41]. In the simulation to get a learning predictive model, the scaled conjugate gradient algorithm is adopted with a configuration of 1,000 iterations at 1 epoch each. In the results for controlling mass flow rate and temperature of supply air, the tested statistical significance of R² is 0.99175 and 0.98969, respectively.

Fig. 1.

Conceptual thermal model

The method of sending appropriate control signals into a system in certain situations has been studied very widely in the deterministic viewpoint. When the system operates over a certain period of time, the difference in two control signals before and after and the derivative of the difference have been preferred to determine the increase or decrease of the next control signal. And these approaches have been applied into in several algorithms to construct the initial deterministic structure.

For maintaining indoor thermal comfort, a deterministic process utilize the results of five different variables related to thermal comfort and energy use as input variables: difference of occupant, derivative of the difference of occupant, difference of PMV value, derivative of the difference of the PMV, derivative of the energy use. Then, by adopting the fuzzy algorithm, the output values of heating and cooling supply air are adjusted by summing the weights specified in each range according to the designed configuration like fuzzy membership functions. At this time, the total output value is combined by use of two elements: mass flow rate, temperature. Then, the adaptive control process optimizes the output values whether they are within or outside in designed ranges for energy use. For example, if the number of workers and visitors does not met a required number, and this increase or decrease rate do not exceed a certain designed range, the output is decreased by 20% to reduce energy use regardless of other conditions. However, in the next simulation phase, if the conditions of the occupant related and the comfort related values are increased while the rate of change in energy use is within a certain designed range, the output is decreased by 5%. In addition, No adjustment is made if all the conditions are within the designed ranges. This adaptive control process is conceptually described in Fig. 2. In order to integrate each of the above programmed thermal and statistical models into one model, Simulink in Matlab r2024b is used as illustrated in Fig. 3. The entire simulation block model consists of 6 different parts: Signal (adaptive), Control (thermostat and ANN), Supply (cooler and heater), Condition (indoor, outdoor, and people), Room, PMV Calculator.

Fig. 2.

Adaptive control process

Fig. 3.

Simulation block model

4. Result and Discussion

Table 3. shows the results of the thermal comfort consistency and the energy efficiency controlled by three different controls, and the performance of two models is improved as compared to the conventional thermostat model. As predicted in the indoor temperature control pattern, the numerical results confirm that the learning predictive model by the ANN algorithm is highly effective in maintaining the consistency of the PMV level. In the adaptive control model, through the adaptive control process of the thermal variables related to the PMV level, the high performance of its thermal comfort was achieved to be more than 50% as compared to the thermostat model, but the learning predictive model showed a very high performance exceeding the result. In particular, it shows a very high effectiveness of maintaining the consistency of thermal comfort in the intensive cooling period from July 25th to August 14th. In addition to this period, significant performance can be seen in the period from June 15th to June 24th and from September 5th to September 14th, which effectively suppresses parts of overshooting when the system turns on for cooling supply air.

Table 3.

Comparison of the thermal comfort by RMSE

Table 4. displays the result of the cooling and heating load controlled by tree different controls. Although the short periods of early summer and early autumn are included on the designed period, it can be considerable for the heating by the configuration of heating set-point temperature. Therefore, this result needs be considered when early morning or late night working hours are included to increase its operational economy. As predicted by the RMSE results, the control characteristics from August 5th to Au-gust 24th are clearly different from other period. The energy use of the learning predictive model, which significantly reduced the overshooting and the fluctuation of the controlled pattern, is confirmed to be higher than that of other periods. In particular, from August 15th to September 14th, it is confirmed that high increases from 15% to 25% for both cooling and heating. It is confirmed that this pattern is effectively maintained from August 25th to September 4th. If this is inferred by the comparison of the graphs, the controlled pattern of the thermostat model clearly has a regular up and down based on the designed set-point temperature, whereas the learning predictive model works continuously supplying heating and cooling energy derived from adjusted values above the set-point temperature (22℃ for heating and 26℃ for cooling). It is predicted that the indoor temperature was controlled a bit higher than the heating set-point temperature and slightly lower than the cooling set-point temperature, according to the adaptive control process of analyzing the number of users and the changes in the PMV values. However, from August 5th to August 14th, it is confirmed that a significant amount of cooling energy use is reduced. This implies a fact that the adaptive control process optimizes the supply of cooling energy responding to the decrease of the number of users in extreme hot temperature in early August. Therefore, accordingly, it can be inferred that the energy efficiency has increased to a significant level in the period when the importance to maintain the thermal comfort is decreased due to changes in designed thermal situations. Overall, the energy use intensity by the adaptive control model slightly increases by about 0.91% as compared to the thermostat model. It can be a reasonable result of the mathematically deterministic model which increases the cooling energy in summer and heating energy in winter in order to maintain the thermal comfort level above the designed levels. On the other hand, it can be inferred that the ANN learning algorithm which made it possible to predict the next situations makes the learning predictive model reduces about 7.35% of energy use.

Table 4.

Comparison of the energy use by thermal loads

Fig. 4. and Fig. 5. display the result of the cooling and heating load controlled by the thermostat model. Depending on the set-point temperature and the dead-band range, highly regular energy supply patterns are confirmed. Fig. 6. and Fig. 7. show the improved performance of the model reflecting the adaptive control process. As predicted, in the thermostat model, the repetitive on/off controls occur in the maximum range of 50MJ for cooling and 100MJ for heating, but the adaptive control model controls to a maximum of 25MJ for cooling and a maximum of 23MJ for heating. In particular, it can be seen that the overshooting that occurs at the beginning of operation is also effectively suppressed to a maximum of 48MJ for cooling and a maximum of 95MJ for heating. Energy use increases in some periods and decreases in others, resulting in a slightly increased total sum of energy use. However, as seen in Fig. 8. and Fig. 9., the learning predictive model controls to a maximum of 16MJ for cooling and to a maximum of 12MJ for heating. In addition, it can be seen that it effectively suppresses the overshooting at the beginning of operation to a maximum of 25MJ for cooling and a maximum of 95MJ for heating. It is inferred that energy use could be effectively suppressed, unlike the thermostat model, by reducing maximum values of the control range and the overshooting in spite of the consistent and continuous control. In the viewpoint of improving the space types sustainability, this result can have an opportunity without consuming any additional resources.

Fig. 4.

Cooling load by the baseline (thermostat) control

Fig. 5.

Heating load by the baseline (thermostat) control

Fig. 6.

Cooling load by the adaptive control model

Fig. 7.

Heating load by the adaptive control model

Fig. 8.

Cooling load by the learning predictive control model

Fig. 9.

Heating load by the learning predictive control model

However, even though the model with increased statistical validation by learning the adaptive control process and improved control results was used, there are still some undefined overshoots and some periods in which the patterns of increase or decrease in energy use are not identified. This should be tested with more effective regression models using more weather conditions and space models reflecting various physical conditions. In addition, it is necessary to improve the statistical validation of input data for the ANN learning process and to increase the precision of deterministic algorithms by diversifying membership functions and variables.

5. Conclusion

This study investigated the performance of a proposed control model designed for an intermittently occupied space to enhance occupant thermal comfort without compromising operational energy efficiency. The methodology involved combining a conventional thermostat model with an adaptive control process that refined output signals by analyzing real-time data on occupant size and thermal comfort levels. The resulting control data were then fed into a learning algorithm to develop a predictive model.

The study concluded by comparing the performance of three different control models based on the consistency of maintaining thermal comfort and the reduction of energy use. The learning predictive model demonstrated significant improvements: it effectively maintained thermal comfort consistency by approximately 84.8% and achieved an energy use reduction of about 7.4%. These results confirm the model’s advantage: it allows for significant energy savings while simultaneously increasing the usability of building types where operational economy is important.

However, the analysis of performance improvement highlighted several areas for future research: the need to incorporate more detailed thermal dynamics, increase statistical validity through the use of more variables and extensive datasets, and analyze unforeseen practical issues by applying the model to actual, functioning buildings. The anticipated rise in the spaces necessitates improved operational methods focused on human-centric issues. The adaptive control process proves valuable in increasing the effectiveness of managing the deterministic trade-off between thermal comfort and energy use. The learning predictive model offers a distinct advantage in maintaining thermal comfort consistency while effectively reducing energy consumption.

A follow-up study will be initiated to reinforce these strengths and address the identified limitations, specifically including a regression model analysis of various control patterns and energy supply characteristics. If this human-centered control method is successfully integrated with broader research in urban planning, material science, and recycling, the methodological spectrum for improving the overall sustainability of various spaces is expected to expand considerably.

References

• G. Huang et al., State of the art review on the HVAC occupant-centric control in different commercial buildings, Journal of Building Engineering, 96, 2024, 110445. [https://doi.org/10.1016/j.jobe.2024.110445]
• J. Ahn, D. Chung, S. Cho, Performance analysis of space heating smart control models for energy and control effectiveness in five different climate zones, Building and Environment, 115, 2017, pp.316-331. [https://doi.org/10.1016/j.buildenv.2017.01.028]
• M. Abuhussain et al., Adaptive HVAC system based on fuzzy controller approach, Applied Science, 13, 2023, 11354. [https://doi.org/10.3390/app132011354]
• J. Ahn, S. Cho, Dead-band vs. machine-learning control systems: Analysis of control benefits and energy efficiency, Journal of Building Engineering, 12, 2017, pp.17-25. [https://doi.org/10.1016/j.jobe.2017.04.014]
• J. Ahn, S. Cho, D. Chung, Analysis of energy and control efficiencies of fuzzy logic and artificial neural network technologies in the heating energy supply system responding to the changes of user demands, Applied Energy, 190, 2017, pp.222-231. [https://doi.org/10.1016/j.apenergy.2016.12.155]
• E. Barbadilla-Martun et al., Analysing the indoor temperature ranges for an efficient control of HVAC systems, Energy and Buildings, 301, 2023, 113681. [https://doi.org/10.1016/j.enbuild.2023.113681]
• X. Liu et al., Enhancing HVAC energy management through multi-zone occupant-centric approach: A multi-agent deep reinforcement learning solution, Energy and Buildings, 303, 2024, 113770. [https://doi.org/10.1016/j.enbuild.2023.113770]
• C. Thakre, R. Vijay, Impact of low-frequency noise on a residential zone due to commercial HVAC systems, Journal of Building Engineering, 91, 2024, 109616. [https://doi.org/10.1016/j.jobe.2024.109616]
• B. Grillone et al., A data-driven methodology for enhanced measurement and verification of energy efficiency savings in commercial buildings, Applied Energy, 301, 2021, 117502. [https://doi.org/10.1016/j.apenergy.2021.117502]
• G. Barone et al., A new thermal comfort model based on physiological parameters for the smart design and control of energy-efficient HVAC systems, Renewable and Sustainable Energy Reviews, 173, 2023, 113015. [https://doi.org/10.1016/j.rser.2022.113015]
• Z. Xu et al., Physics guided wavelet convolutional neural network for wind-induced vibration modeling with application to structural dynamic reliability analysis, Engineering Structures, 297, 2023, 117027. [https://doi.org/10.1016/j.engstruct.2023.117027]
• S. Zhou et al., A comprehensive review of the applications of machine learning for HVAC, Decarbon, 2, 2023, 1000023. [https://doi.org/10.1016/j.decarb.2023.100023]
• F. Ciampi et al., Energy consumption prediction of industrial HVAC systems using Bayesian Networks. Energy and Buildings, 309, 2024, 114039. [https://doi.org/10.1016/j.enbuild.2024.114039]
• S. Yoon, J. Ahn. Comparative analysis of energy use and human comfort by an intelligent control model at the change of season, Energies, 13, 2020, 6023. [https://doi.org/10.3390/en13226023]
• J. Ahn, S. Cho, Development of an intelligent building controller to mitigate indoor thermal dissatisfaction and peak energy demands in a district heating system, Building and Environment, 124, 2017, pp.57-68. [https://doi.org/10.1016/j.buildenv.2017.07.040]
• C. Turhan et al., An integrated decision-making framework for mitigating the impact of urban heat islands on energy consumption and thermal comfort of residential buildings, Sustainability, 15, 2023, 9674. [https://doi.org/10.3390/su15129674]
• A. Agouzoul et al., Advancing sustainable building practices: Intelligent methods for enhancing heating and cooling energy efficiency, Sustainability, 16, 2024, 2879. [https://doi.org/10.3390/su16072879]
• A. Skala et al., Implementation of building a thermal model to improve energy efficiency of the central heating system-A case study, Energies, 16, 2023, 6830. [https://doi.org/10.3390/en16196830]
• J. Ahn, S. Cho, Anti-logic or common sense that can hinder machine’s energy performance: Energy and comfort control models based on artificial intelligence responding to abnormal indoor environments, Applied Energy, 204, 2017, pp.117-130. [https://doi.org/10.1016/j.apenergy.2017.06.079]
• X. Liu, Z. Gou, Occupant-centric HVAC and window control: A reinforcement learning model for enhancing indoor thermal comfort and energy efficiency, Building and Environment, 250, 2024, 111197. [https://doi.org/10.1016/j.buildenv.2024.111197]
• J. Moon, J. Ahn, Improving sustainability of ever-changing building spaces affected by users’ fickle taste: A focus on human comfort and energy use, Energy and Buildings, 208, 2020, 109662. [https://doi.org/10.1016/j.enbuild.2019.109662]
• A. Youssef et al., Development and comparison of adaptive data-driven models for thermal comfort assessment and control, Total Environment Research Themes, 8, 2023, 100083. [https://doi.org/10.1016/j.totert.2023.100083]
• A. Laouadi, A new general formulation for the pmv thermal comfort index, Buildings, 12, 2022, 1572. [https://doi.org/10.3390/buildings12101572]
• USEIA, Commercial buildings energy consumption survey final results, http://www.eia.gov/consumption/commercial/
• B. Barbosa et al., Energy efficiency and thermal comfort analysis in a higher education building in Brazil, Sustainability, 16, 2024, 462. [https://doi.org/10.3390/su16010462]
• A. Baba et al., Indoor environmental quality for comfort learning environments: Case study of Palestinian school buildings, Buildings, 14, 2024, 1296. [https://doi.org/10.3390/buildings14051296]
• M. Rodiriguez et al., Thermal comfort in educational buildings: The classroom-comfort-data method applied to schools in Bogota, Colombia, Building and Environment, 194, 2021, 107682. [https://doi.org/10.1016/j.buildenv.2021.107682]
• K. Strelets et al., Analysis of microclimatic comfort conditions in university classrooms, Sustainability, 16, 2024, 3388. [https://doi.org/10.3390/su16083388]
• D. Custodio et al., Thermal comfort in university classrooms in humid subtropical climate: Field study during all seasons, Building and Environment, 258, 2024, 111644. [https://doi.org/10.1016/j.buildenv.2024.111644]
• G. Settimo et al., CO₂ levels in classrooms: What actions to take to improve the quality of environments and spaces, Sustainability, 16, 2024, 8619. [https://doi.org/10.3390/su16198619]
• J. Ahn, Development of energy performance metrics for airport terminal buildings using multivariate regression modeling, Raleigh, NC: North Carolina State University, 2016.
• J. Ahn, S. Cho, D. Chung, Development of a statistical analysis model to benchmark the energy use intensity of subway stations, Applied Energy, 179, 2016, pp.488-496. [https://doi.org/10.1016/j.apenergy.2016.06.065]
• J. Ahn, A benchmark methodology to assess the energy performance of train station complexes, KIEAE Journal, 19, 2019, pp.5-12. [https://doi.org/10.12813/kieae.2019.19.6.005]
• S. Kim, H. Shin, J. Ahn, Energy performance analysis of airport terminal buildings by use of architectural, operational information and benchmark metrics, Journal of Air Transport Management, 83, 2020, 101762. [https://doi.org/10.1016/j.jairtraman.2020.101762]
• The Engineering Toolbox, Building elements - heat loss vs. thermal resistivity, https://www.engineeringtoolbox.com/heat-loss-buildings-d_194.html, , 2003.
• F. Incropera et al., Fundamentals of heat and mass transfer, New York, NY: Wiley, 2018.
• The Engineering Toolbox, Predicted Mean Vote Index, https://www.engineeringtoolbox.com/predicted-mean-vote-index-PMV-d_1631.html, , 2010.
• National Institute of Building Science, Space types, Whole Building Design Guide, https://www.wbdg.org/space-types, , 2018.
• K. Ahamed, S. Akthar, A study on neural network architectures, Computer Engineering and Intelligent Systems, 7, 2016, pp.2222-1719.
• R. Vamsi, What’s a neural network?, https://medium.com
• D. Palladio et al., Artificial neural network for the thermal comfort index prediction: Development of a new simplified algorithm, Energies, 13, 2020, 4500. [https://doi.org/10.3390/en13174500]