Energy consumption of using HEPA-based portable air cleaner in residences: A monitoring study in Seattle, US

Abstract

Portable air cleaners (PACs), offering both auto and manual (adjustable) operation modes, are commonly used in residences. Compared with adjustable mode, auto mode’s advantage of reducing indoor PM_2.5 has been previously demonstrated. This study examines the energy consumption of such PACs in six residences recruited in Seattle, United States, and compares the power consumption between auto and adjustable modes. Each residence went through a one-week-long PAC filtration session under auto and adjustable modes, respectively. PAC power consumption, indoor PM_2.5, temperature, and relative humidity (RH) were measured at 10-second intervals in each residence. A linear mixed-effects regression (LMER) model was used to compare the PAC power consumption between the two modes after adjusting for indoor PM_2.5, temperature, and RH. Results show that the mean (standard deviation) PAC power consumption under adjustable and auto modes were 7.0 (3.5) and 6.8 (2.6) W, respectively. The average monthly energy consumption of continuous PAC operation was estimated to be ~5 kWh for both modes. Based on the LEMR model, PAC power consumption under auto mode was approximately 3% larger than that under adjustable mode, after adjusting for living-room PM_2.5, temperature, and RH levels. The implications for PAC operation mode selection in residential environments were discussed.

1. Introduction

In the United States (US), residential PM_2.5 is from a variety of sources, including outdoor infiltration and ventilation [1, 2], and indoor cooking fumes [3, 4], tobacco smoke [5], and other indoor activities [6]. Despite the relatively low outdoor PM_2.5 levels in the US, residential PM_2.5 levels can be up to 300 μg/m³ with indoor/outdoor ratios > 80 when cooking is present [4]. Considering that the US people spend nearly 70% of their time in a residence, the fraction of residential PM_2.5 exposure to total exposure is often dominant [7].

Common building ventilation and indoor PM_2.5 filtration strategies include natural ventilation with portable air cleaners (PACs) and mechanical ventilation with air filters [4, 8–15]. Building energy consumption of ventilation and filtration plays a critical role in global energy consumption and greenhouse gas emissions [16]. Comparing the indoor air quality and energy consumption of various ventilation and filtration strategies has attracted much attention [16–22]. Given that these ventilation strategies cannot fully replace each other, at least in the near term, it is of much value to optimize each ventilation mode and reduce associated energy consumption. Compared with mechanical ventilation with air filters, higher clean air delivery rates (CADRs) and lower indoor PM_2.5 levels can be achieved more easily under natural ventilation with PACs [17]. Also, it is quite flexible and convenient to use a PAC in homes. Although natural ventilation with PAC use is quite common in residences, few studies have examined the energy consumption of PAC operation in the real-world residential environment. Only two studies have involved measurement of PAC power consumption in Chinese residences and showed that the actual PAC power consumptions were much different from the manufacture-stated ones [18, 23]. Considering the substantial differences between Chinese and US residential environments, these two Chinese surveys do not work for US residences. Overall, there is minimal information on PAC energy consumption in real-world residential settings, especially in the US.

In addition to manual/adjustable operation mode, which refers to adjusting the fan speed manually, varieties of PACs offer an auto operation mode. Auto mode refers to a feature that automatically adjusts PAC fan speed according to the surrounding PM_2.5 levels. This feature is commonly used in real-world settings [23]. A recent study has validated the auto-mode feature and suggested that PACs operating in auto mode can reduce indoor PM_2.5 levels by ~20% compared to the adjustable operation mode with the presence of primary indoor sources [4]. Despite the potential of the auto-mode feature in removing residential PM_2.5, it remains unclear whether PACs operating in this mode consume more energy than the adjustable mode.

In this study, we analyzed the PAC power consumption data collected from a previous study that initially evaluated the impacts of using auto-mode PACs on residential PM_2.5 levels [4]. We aim to (1) examine the energy consumption of PACs operating in adjustable and auto modes in practical US residences, and (2) compare the PAC power consumption between adjustable and auto modes.

2. Methods

2.1. Experimental design

The design of the present study has been described in detail elsewhere [4]. In brief, we recruited six residences (four apartments and two houses, labeled as R1–R6), based on a sample of personnel at the University of Washington in Seattle, Washington, US. From February to March 2019, each residence went through three one-week PAC filtration sessions (i.e., sham-mode, adjustable-mode, and auto-mode filtration session) with the order of the three filtration modes randomized. PACs (Air Purifier 2000i, Philips, Andover, US) containing a pre-filter, an active carbon filter, and a high-efficiency particulate air (HEPA) filter were utilized in the study. Based on the manufacture, the clean air delivery rate (CADR) of this PAC is 198 m³/h for dust, and the maximum power is 56 W [24]. The PAC power consumption data under sham-mode (i.e., no filters installed) filtration is not practical. Hence, we only compared the adjustable and auto modes in the present study. Specifically, adjustable-mode and auto-mode filtration refer to running PAC, with all the filters installed, in manual mode and auto mode, respectively. The manual mode includes five fan speed levels: sleep mode (minimum), levels 1–3, and turbo (maximum). We measured the PAC power consumption under each fan speed, ranging from ~4 to ~50 W (Appendix Table A1). Under adjustable-mode filtration, participants were allowed to adjust the PAC fan speed (five options) manually as long as the PAC was kept on. In contrast, they were not allowed to adjust the PAC under auto-mode filtration. All residences were naturally ventilated, with windows and doors kept closed during the experiment. The PAC was placed in the living room of each residence, at least 1 meter away from the walls.

2.2. Measurement

A high-accuracy power monitor (HOBO® Plug Load Logger Model UX120-018, Onset Computer Corp, Bourne, MA) was used to record each 10-second PAC power consumption. Real-time PM_2.5 monitors were used to measure 10-second resolved PM_2.5 mass concentration, relative humidity (RH), and temperature in the living room and kitchen of each residence. The PM_2.5 monitor, which utilizes a Plantower PMSA003 optical particle sensor, was calibrated in residence before the deployment (see more details elsewhere [4]). A time-activity diary in hour resolution was collected from each participant during the intervention, including cooking, cleaning, and candle burning.

2.3. Statistical analysis

The current analysis focuses on PAC power consumption as the PM_2.5 results were reported previously [4]. In addition to the operation mode, PAC power consumption was likely impacted by indoor PM_2.5 concentrations. Also, environmental temperature and RH could potentially impact the PAC electric motor windings’ resistance, and thus, the PAC power consumption. To compare the power consumption between adjustable and auto modes accurately, we used a statistical model adjusting for the PM_2.5, temperature, and RH levels in the living room. As indoor PM_2.5 levels can reflect the impacts of indoor activities (e.g., cooking), no indoor-activity variable was included in the model. Specifically, we used a linear mixed-effects regression (LMER) model as shown in Equation (1), which accounts for both fixed-effect factors (PAC operation mode, living-room PM_2.5 levels, temperature, and RH) and random-effect factor (residence). We used natural logarithm-transformed data in the models due to the right-skewed distribution of the measured PAC power consumptions. Also, we performed Spearman correlation tests between the independent variables. For all the analyses, the 10-second data were aggregated into hourly means.

$$\log \left(P_{it}\right)={\beta }_0+{\beta }_{1}FI{L}_{it}+{\beta }_{2}C{\left(in\right)}_{it}+{\beta }_{3}Temp{\left(in\right)}_{it}+{\beta }_{4}RH{\left(in\right)}_{it}+W_i+{\epsilon }_{it}$$ (1)

where β₀ – β₄ are the coefficients from the LMER model; log(P_it) is the natural log-transformed PAC power consumption in residence i at time t, W; FIL_it is PAC filtration mode – a binary variable (adjustable-mode filtration as the reference category, level = 0); C(in)_it is the living-room PM_2.5 concentration in residence i at time t, μg/m³; Temp(in)_it is the living-room temperature in residence i at time t, °C; RH(in)_it is the living-room relative humidity in residence i at time t, %; W_i is the random effect factor, which reflects a different baseline level of PAC power consumption for each residence; and ε_it is the residual.

Because the two operation modes resulted in different indoor PM_2.5 levels, the estimates of both β₁ and β₂ determined from the LMER model were used to predict the mean percent changes of PAC power consumption under auto mode relative to that under adjustable mode, as shown in Equation (2).

$$\Delta P\left(\%\right)=\left(e^{\left({\beta }_1+{\beta }_2\times \Delta C\left(in\right)\right)}-1\right)\times 100\%$$ (2)

where ΔC(in) is the mean absolute change of living-room PM_2.5 levels under auto mode in contrast to those under adjustable mode.

Six sensitivity analysis (SA) models were assessed via various combinations of independent variables: (1) a basic model including only the PAC filtration mode and residence-specific random effect factor; (2) SA model (1) plus the living-room PM_2.5 concentration; (3) SA model (1) plus the living-room temperature and RH; (4) SA model (2) plus residence floor size; (5) SA model (3) plus residence floor size; and (6) main model plus residence floor size. All statistical analyses were conducted using RStudio (Version 1.3.1093) with “nlme” and “stats” packages. P-value < 0.05 was taken as the indicator of statistical significance in this study.

3. Results

3.1. Measured PAC power consumption

Fig. 1 shows the bin-specific and cumulative probability of hourly averaged PAC power consumption under each operation mode in the six residences. The hourly power consumptions were less than 11 W for 97% of the total running time under both operation modes, indicating that the PACs ran at level 2 (~10 W) and below most of the time in these residences. The maximums were 48 and 34 W for adjustable mode and auto mode, respectively, representing a mixture of level 3 (~20 W) and turbo (~50 W). The cumulative probability curves for the two modes overlap each other, mostly at lower fan speed levels (level 1 and below). In contrast, they have relatively more considerable differences at level 2 and above. In particular, the time proportion of PACs running at level 2 under adjustable mode is much larger than that under auto mode. It reveals that the PAC fan speed under auto mode increased rapidly when a significant indoor PM_2.5 source occurred.

Fig 1.

Bin-specific and cumulative probability of hourly averaged PAC power consumption under adjustable and auto modes in the six residences. Bars and colored lines in the plot represent bin-specific and cumulative probability, respectively.

Table 1 shows the hourly averaged PAC power consumption and living-room PM_2.5 levels under the two operation modes with the data for all residences pooled. The overall mean (standard deviation, SD) power consumption under adjustable mode was 7.0 (3.5) W, with a range of 2.1–47.6 W. In contrast, the mean power consumption under auto mode was comparable but slightly lower (3%). Based on the measured mean power consumptions, the estimated PAC energy consumptions per month (continuous running) under adjustable and auto modes were 5.0 and 4.9 kWh, respectively. On the other hand, the hourly averaged living-room PM_2.5 levels under the two modes were less than 10 μg/m³, with 18% lower under auto mode. Based on the time-activity diaries, 150 cooking events and 15 cleaning events were reported during the experiment. With cooking present, the mean PAC power consumptions increased by 24% and 29% compared with the overall averages under adjustable and auto modes, respectively. The more considerable increase under auto mode indicates a relatively better PAC running strategy in response to major indoor source occurrence. By comparison, no significant increases in PAC power consumptions were observed when indoor cleaning events happened, in compliance with the trend of PM_2.5 levels.

Table 1.

Summary of hourly average PAC power consumption and living-room PM_2.5 levels with different indoor activities.

Scenario	Mode	Power consumption (W)					Energy consumption per month (kWh)b	Living room PM2.5 (μg/m3)		N
		Min	Median (IQR)	Mean (SD)	Max	RD (%) a		Mean (SD)	RD (%) a
All data	ADJ	2.1	6.2 (1.0)	7.0 (3.5)	47.6	/	5.0	9.5 (11.5)	/	982
All data	AU	4.0	6.4 (0.6)	6.8 (2.6)	34.2	−3.1	4.9	7.8 (6.7)	−18.1	1007
Cooking	ADJ	4.0	6.3 (4.0)	8.7 (7.2)	47.5	/	/	17.9 (25.1)	/	81
Cooking	AU	4.7	6.8 (1.0)	8.8 (5.6)	33.8	0.4	/	11.7(15.1)	−34.8	69
Cleaning	ADJ	6.1	6.2 (0.4)	6.9 (1.6)	10.2	/	/	5.0 (2.3)	/	6
Cleaning	AU	4.1	6.3 (0.4)	6.5 (2.2)	11.7	−5.5	/	8.9 (4.0)	78.5	9

Definition of abbreviations: ADJ = adjustable-mode; AU = auto-mode; IQR = interquartile range; SD = standard deviation; RD = relative difference; N = number of measurement hours.

^aCompared with adjustable-mode filtration in the corresponding scenario.

^bEstimated based on mean power consumption: power consumption × 24 hours/day × 30 days/month.

Residence-specific results are presented in Appendix Table A2. The mean (SD) PAC power consumption under adjustable and auto modes ranged from 4.0 (0.2) to 10.2 (0.1) W and 6.2 (4.1) to 7.3 (2.1) W, respectively, among the six residences. The relatively larger variations under adjustable mode indicate the variations in PAC use behaviors among the residences. The estimated monthly PAC energy consumptions were ranging 3–6 kWh based on the measured mean power consumption. Additionally, the relative magnitudes of PAC power consumptions between the two modes varied among residences. On average, PACs running under auto mode consumed more power in Residences R1–R3 and less power in Residences R4–R6. In particular, the PAC power consumption and living-room PM_2.5 levels were both lower under auto mode in Residences R4 and R5.

3.2. Model estimates

Appendix Fig. A1 shows the Spearman correlations among the independent variables in the main LMER models. Except for the moderate correlation between living-room temperature and RH (r = −0.5), the correlations were generally negligible (|r| ≤ 0.1). Because the current study focuses on the effects of operation modes on PAC power consumption while other factors remained constant, the main results are not affected by the moderate correlation between living-room temperature and RH variables. Table 2 summarizes the main LMER model results. All the regression coefficients were significant (p-value = 0.000), with positive values for the intercept (β₀), auto-mode PAC use (β₁), living-room PM_2.5 levels (β₂), and RH (β₄), and negative values for living-room temperature (β₃). The coefficient of determination (R²) for the main model was 0.57, indicating that the model can explain ~60% of data variances. Based on the model results, Table 2 also shows the predicted percent change in PAC power consumption associated with the change in those independent variables and the predictor (β₁+β₂×ΔC(in)). In particular, compared with the adjustable-mode operation, auto-mode operation increased PAC power consumption by 3.2% [95% confidence interval (CI): 1.3 – 5.2%], after adjusting for living-room PM_2.5, temperature, and RH levels.

Table 2.

Summary of the results for the linear mixed-effects regression (LMER) models.

Factor	Coefficient	Unit	Estimate	SE	p-value	R2	Percent change in power consumption (%)a
Intercept	β0	W	1.847	0.099	0.000	0.568	/
FIL	β1	/	0.036	0.010	0.000	0.568	3.7 (1.7 – 5.7)
C(in)	β2	W/(μg/m3)	0.003	0.001	0.000	0.568	0.3 (0.2 – 0.4)
Temp(in)	β3	W/(°C)	−0.019	0.002	0.000	0.568	−1.9 (−2.4 – −1.4]
RH(in)	β4	W/(%)	0.019	0.001	0.000	0.568	1.9 (1.7 – 2.2)
Predictor	β1+ β2×ΔC(in)	/	0.032	0.010b	/	/	3.2 (1.3 – 5.2)

Definition of abbreviations: FIL = filtration mode; C(in) = living-room PM_2.5 concentration; Temp(in) = living-room temperature; RH(in) = living-room relative humidity; SE = standard error.

^aMean with a 95% confidence interval of percent changes in PAC power consumption associated with a 1-unit increase in each variable.

^bCalculated based on error propagation functions; ΔC(in) equals −1.7 μg/m³ based on Table 1.

As shown in Appendix Table A3, all the sensitivity analysis models result in < 3.2% differences in PAC power consumption between adjustable and auto modes. The current main model, which results in a considerable R² (0.57) and all significant regression coefficients (p-value = 0.000), was reported as the main results.

4. Discussion

Few studies have examined the energy consumption of PAC operation in the real-world residential environment. To our knowledge, the current study is the first to measure the PAC energy consumption in US residences, and the first to compare the PAC power consumption between auto-mode and adjustable-mode operations after adjusting for indoor PM_2.5, temperature, and RH levels. Results show that the measured mean (SD) PAC power consumption under adjustable and auto modes were 7.0 (3.5) and 6.8 (2.6) W, respectively, for the recruited residences. The average monthly energy consumption of continuous PAC operation was estimated to be ~5 kWh for both modes in these residences. Based on the LEMR models, PAC power consumption differences between adjustable and auto modes were relatively small (< 4%) after adjusting for confounding factors.

Our previous analysis shows that, with the presence of primary indoor sources, auto-mode PAC filtration significantly reduced the indoor PM_2.5 levels by ~30% and 20% compared with sham-mode and adjustable-mode filtration [4]. With the present study results, auto mode seems to be a better operation choice in US residences than the adjustable or manual mode. An overall 20% decrease of indoor PM_2.5 level can be gained by switching adjustable-mode operation to auto-mode operation, accompanied by a 3% increase in operating power consumption. Considering the minor increase of energy consumption, approximately 0.2 kWh per month (~0.02 US Dollar in Seattle), the benefits of running the PACs in auto mode are apparent. Despite a limited sample size, the current evidence favors the operation of PACs in auto mode for the non-expert public.

Besides the operation modes, the PAC power consumption was impacted by indoor PM_2.5, temperature, and RH levels to varying degrees. The LMER models in this study did not include variables of indoor events, such as cooking, because the indoor PM_2.5 levels reflected the presence/absence of indoor events. In this study, indoor PM_2.5 levels were positively associated with PAC power consumption under both operation modes, which does not necessarily mean an increase in indoor PM_2.5 concentration always led to increased PAC power consumption. The environmental temperature was negatively associated with PAC power consumption, reflecting that the resistance of the PAC electric motor windings (mainly made of copper) increases with temperature [25]. In contrast, an increase in environmental RH, or moisture, likely decreases the overall insulation system resistance and increases PAC power consumption [26]. Without adjusting the indoor PM_2.5, temperature, and RH levels, it would be challenging to compare the PAC power consumption under different operation modes. The LEMR models enabled us to make the comparison more accurately after controlling for major confounding factors. With the confounding factors adjusted, the PAC power consumption differences between adjustable and auto modes reflect the contrasts between the “smart device” (i.e., auto-mode PAC) and human operating behaviors. Although the participants generally tended to increase the PAC fan speed and sustain it for a certain amount of time when major indoor sources (e.g., cooking fumes) occurred, there could be a mismatch between indoor PM_2.5 levels and PAC power output. For instance, users might forget to turn down the PAC fan speed levels when indoor PM_2.5 decayed to the background level, leading to larger power consumption than the “smart device” operating mode, which adjusted its fan speed continuously, from turbo to level 1, based on the surrounding PM_2.5 concentrations.

As mentioned earlier, there exists a growing interest in comparing the benefits and costs of various ventilation and air filtration strategies, including natural ventilation, natural ventilation + PACs, and mechanical ventilation + air filters. The current study does not aim to answer such questions. However, given their unique merits, these strategies will probably co-exist in the real world for a long time. In the scenarios where natural ventilation with PAC use applies, the present study shows that continuous running of PACs under auto mode in the recruited US residences consumed ~5 kWh per month, which is approximately 0.6% of average monthly electricity consumption for a US home [27]. Therefore, it seems cost-effective to run a HEPA-based PAC under auto mode in US residences. Nevertheless, this cost should be compared with those from other studies cautiously because it only refers to PAC operation cost rather than heating, ventilation, and air conditioning (HVAC) cost, which can be much higher.

The present study was conducted during non-wildfire seasons when outdoor PM_2.5 levels in Seattle are generally lower than 20 μg/m³ [28, 29]. In contrast, the outdoor PM_2.5 concentrations during wildfire seasons can be 10 times higher. As a result, indoor PM_2.5 levels would be significantly higher than those reported in this study; therefore, the PAC energy consumption would be more extensive. Similarly, the results herein do not directly apply to regions with relatively much higher outdoor PM_2.5 levels. However, the outdoor-infiltrated indoor PM_2.5 concentrations are generally higher when outdoor levels are higher [30, 31]. Based on our previous study, auto-mode PAC filtration significantly reduced the indoor PM_2.5 levels by ~ 20% compared with adjustable-mode filtration [4]. Therefore, it can be inferred that the absolute difference between auto mode and adjustable mode should be larger when outdoor PM_2.5 levels are higher. Based on the prediction function in this study (Equation (2)), the PAC power consumption under auto mode might be smaller than that under adjustable mode. In both scenarios, the results do not change the conclusion that the current evidence favors the operation of PACs in auto mode for the non-expert public. Nevertheless, future analyses of examining auto-mode PAC power consumption for different scenarios/regions and larger sample sizes, using a method parallel to that applied in the current study, are warranted. On the other hand, a potential difference between lower and higher outdoor PM_2.5 scenarios is the noise during sleep periods. In the present study, the auto-mode PACs generally ran at level 2 and below at nighttime, and the participants did not report a noise issue. It remains unclear whether the noise would be an issue under auto-mode PAC filtration in higher outdoor PM_2.5 scenarios, and thus, requires further study.

5. Conclusions

This study shows that the average monthly energy consumption of continuous PAC operation was ~5 kWh for the recruited US residences. It suggests that running such PACs is not a massive burden to total energy consumption in the US. With indoor PM_2.5, temperature, and RH fully adjusted, PAC power consumption under auto mode was approximately 3% larger than that under adjustable mode. Given the substantially lower indoor PM_2.5 levels under auto-mode operation than adjustable-mode operation [4], the findings favor taking auto mode as the primary PAC operation mode in residential environments. Such analysis provides practical guidance to the PAC use in the US.

Highlights.

• Portable air cleaner (PAC) power consumption was monitored in 6 US homes for weeks.

• PAC power consumptions between auto and manual operating modes were compared.

• Multivariate mixed-effects models were used to adjust for confounding factors.

• Continuous running of PACs in the recruited residences consumed ~5 kWh per month.

• PAC power consumption under auto mode was approximately 3% larger than the other.