智慧型節能:使用感測網路自動偵測異常空調 狀態之研究 Intelligent Sensing for Energy Saving : A Case Study on Detecting Abnormal Air-Conditioning States Using A Sensor Network Presenter : Min-Chia Chang Advisor : Prof. Jane Hsu Date : 2011-06 -30 Outline Introduction System Architecture Analysis Conclusion NTU CSIE iAgent Lab 2 Energy Saving Reason Policy NTU CSIE iAgent Lab 3 Power Consumption in a Building (source : Continental Automated Buildings Association, CABA) NTU CSIE iAgent Lab 4 Architecture of Central A/C System Chilled water host • Evaporator • Condenser Other devices • Pump • Cooling tower NTU CSIE iAgent Lab 5 Energy Conservation for Central A/C System Device setting • The setting of the chiller water [Zhao, Enertech Engineering Company] • Parameter optimization of the cooling tower [James and Frank 2010] Building automation system • Component • Energy saving controller • Infrared motion sensor (source : NTU 電機學系) NTU CSIE iAgent Lab 6 Power Consumption in NTU CSIE Total • 9,036.4 KWH/day ≒ 28,012 NTD/day ( January 2009 - April 2011 ) (source : NTU 校園數位電錶監視系統) Central A/C system ( July 2010 - May 2011 ) • 3,693.8 KHW/day • 40.88% of the total 7 Control of Central A/C System Central • Chilled water host • Off mode • On mode (All year on duty) Local • A/C controller • Off mode • Venting mode • Cooling mode NTU CSIE iAgent Lab 8 Abnormal A/C State in NTU CSIE Ideal A/C power consumption • Assumption : people number ∝ A/C power consumption KWH NTU CSIE iAgent Lab 9 Abnormal A/C State in NTU CSIE Real A/C power consumption • From electricity meter 20KWH A/C is turned off ? KWH NTU CSIE iAgent Lab 10 Abnormal A/C State in NTU CSIE Hot Cold NTU CSIE iAgent Lab 11 Outline Introduction System Architecture Analysis Conclusion NTU CSIE iAgent Lab 12 System Overview NTU CSIE iAgent Lab 13 Wireless Sensor Network NTU CSIE iAgent Lab 14 Sensors Platform : Taroko • Temperature and humidity sensor : SHT11 • Infrared motion sensor NTU CSIE iAgent Lab 15 Nodes in the Sensor Network Sender • (temperature, humidity, ID) • (preamble, motion value, ID) Receiver • Data saving : 1 minute Relay NTU CSIE iAgent Lab 16 Deploy Unit Room : divide into zones according to A/C controller Environmental data • temperature and humidity • vent • indoor • motion value vent indoor motion sensor NTU CSIE iAgent Lab 17 Deployment One server per floor (1F to 5F) Relays deploy around the corridors NTU CSIE iAgent Lab 18 Deployment Room • Class room : R104 • Computer class room : R204 • Professor room : R318 • Laboratory : R336 • Seminar room : R324, R439, R521 NTU CSIE iAgent Lab 19 A/C Mode Recognition NTU CSIE iAgent Lab 20 A/C Mode Mode • Off mode : blower= off , valve = off • Venting mode : blower = on , valve = off • Cooling mode : blower= on , valve = on A/C wind velocity A/C Off Venting Cooling mode power temperature setting NTU CSIE iAgent Lab indoor ≧ temperature < 21 A/C Mode Recognition GOAL : • Using machine learning to build the model for recognizing the A/C mode INPUT : • Feature vector OUTPUT : • A/C mode NTU CSIE iAgent Lab 22 Features Category Feature Dimension Type Temperature and Humidity TI , HI , TV , HV , TO , HO 6 Float ΔTI,V , ΔHI,V ΔTI,O , ΔHI,O ΔTO,V , ΔHO,V 6 Float Host 3 {0,1} Leaving Temperature 1 Float Rotation Speed of Pump 1 Float Building 2 {0,1} Floor 5 {0,1} Room Type 5 {0,1} Area 1 Float Delta Parameters (Central A/C ) Spatial NTU CSIE iAgent Lab 23 Annotation Period Place Annotation November 2010 R104, R204, R318, R324, R336, R439, R521 Method1 December 2010 January 2011 R204, R324, R336 Method2 February 2010 March 2011 R104, R204 Method2 Method 1 • Control on purpose Method 2 • Record by camera • Temporal feature NTU CSIE iAgent Lab 24 Dataset Place Total Data Label = off Label = venting Label = cooling Missing Data 204_1 17,345 6,371 6,817 4,157 4,686 27.0% 204_2 17,106 6,569 8,168 2,369 4,603 26.9% 204_3 16,694 6,357 5,381 4,956 4,576 27.4% 204_4 16,415 8,592 5,014 2,809 4,632 28.2% 204_5 17,487 6,569 0 10,918 6,054 34.6% 204_6 15,616 6,794 5,843 2,979 9,158 58.7% 336_2 20,889 6,439 10,100 4,350 5,931 28.4% NTU CSIE iAgent Lab 25 Experiment Setting Each zone builds a model 1. 3-fold cross validation 2. The weather pattern in testing data doesn’t exist in training data • Does not collect all the weather patterns NTU CSIE iAgent Lab 26 Steps of the Experiment 2 Cluster • Algorithm : k-means (k=4) • Feature: outdoor temperature, outdoor humidity Leave-one-out Cross Validation outdoor humidity 336_2 outdoor temperature NTU CSIE iAgent Lab 27 Preprocessing Missing data treatment • Encoding • Recognize the data is missing or not • Linear interpolation • The change of temperature or humidity is linear Normalization • Min-max normalization : [0,1] • It prevents features with large scale biasing the result NTU CSIE iAgent Lab 28 Experiment Result Result • The model achieves high accuracy • The model can recognize the data with the weather pattern not included in training data Zone Experiment 1 Experiment 2 • 204_5 has the highest accuracy 204_1 98.5% 87.0% 204_2 89.1% 85.0% 204_3 99.8% 98.4% 204_4 97.9% 90.2% 204_5 99.9% 99.0% 204_6 93.9% 86.3% 336_2 93.2% 92.1% NTU CSIE iAgent Lab 29 Thermal Comfort Calculation NTU CSIE iAgent Lab 30 Thermal Comfort Calculation GOAL : • Find the thermal comfort range to determine the indoor temperature being too cold or too hot INPUT : • Questionnaire OUTPUT : • Thermal comfort range NTU CSIE iAgent Lab 31 PMV Predicted Mean Vote model [Fanger 1970] • Calculated analytically by 6 factors : [-3, +3] • Metabolic rate • Clothing insulation • Air temperature • Radiant temperature (Outdoor temperature) • Relative humidity • Air velocity NTU CSIE iAgent Lab 32 Thermal Sensation Scale Thermal sensation scale [ASHRAE Standard 55] • Adaptive method to get PMV • Thermal sensation vote (TSV) • Constraints • Metabolic rate : 1.0Met - 2.0Met • Clothing insulation : ≦ 1.5 Clo • Comfortable or not • -1, 0, +1 : yes Scale Thermal sensation +3 Hot +2 Warm +1 Slightly warm 0 Neutral -1 Slightly cool -2 Cool -3 Cold • -2, -3, +2, +3 : no NTU CSIE iAgent Lab 33 Thermal Comfort - Linear Regression Field survey • Collect thermal sensation vote • Outdoor temperature has the highest relevance 1. TC = 17.8 + 0.31TO (Worldwide) [deDear and Brager 1998] 2. TC = 18.3 + 0.158TO (Hong Kong) [Mui and Chan 2003] 3. TC = 15.5 + 0.29TO (Taiwan) [Lin et al. 2008] NTU CSIE iAgent Lab 34 Questionnaire Thermal sensation scale : {-3, -2, -1, 0 ,+1, +2, +3} Direct question : {comfortable, not comfortable} Metabolic rate : {after sport, static activity} VALID ! Insulation : {sleeveless, shirt-sleeve, long-sleeve, thick coat} NTU CSIE iAgent Lab 35 Data Collection R204 (computer class room) R336 (laboratory) Period March 2010 - July 2010 December 2010 - February 2011 Number 1,745 1,033 -3 -2 -1 0 +1 +2 +3 Comfortable 55 (46%) 10 (24%) 283 (86%) 1604 (98%) 308 (70%) 39 (43%) 16 (14%) Not Comfortable 65 (54%) 32 (76%) 48 (14%) 34 (2%) 131 (30%) 52 (57%) 101 (86%) NTU CSIE iAgent Lab 36 Result Linear regression equation • TC = 20.6+ 0.107TO 1. TC = 17.8 + 0.31TO (Worldwide) 2. TC = 18.3 + 0.158TO (Hong Kong) 3. TC = 15.5 + 0.29TO (Taiwan) NTU CSIE iAgent Lab 37 PMV - PPD Predicted of Percentage Dissatisfied model [Olesen and Bragen 2004] • Typical standard : 80% acceptability, (PMV, PPD)= (±0.85, 20) • Higher standard : 90% acceptability, (PMV, PPD)= (±0.50, 10) NTU CSIE iAgent Lab 38 Thermal Comfort Range Regression • Indoor temperature • Mean thermal sensation vote (PMV) during each ℃ 2.67 NTU CSIE iAgent Lab 39 Thermal Comfort Range 2.67 NTU CSIE iAgent Lab 40 A/C State Evaluation NTU CSIE iAgent Lab 41 A/C State Evaluation GOAL : • Classify the room’s A/C state to normal or abnormal INPUT : • Each zone • Occupancy state • A/C mode • Indoor temperature • Thermal comfort range OUTPUT : • A/C state NTU CSIE iAgent Lab 42 A/C State people in the room Y A/C = cooling mode N A/C = turned on N Y indoor temperature ? comfort range lower within higher abnormal normal abnormal normal Y abnormal NTU CSIE iAgent Lab N normal 43 Outline Introduction System Architecture Analysis Conclusion NTU CSIE iAgent Lab 44 Analysis of Abnormal A/C States Abnormal A/C States Detecting System normal/ abnormal history data analysis NTU CSIE iAgent Lab useful information User 45 Target Room Room • Class room : R104 • Computer class room : R204 • Professor room : R318 • Laboratory : R336 • Seminar room : R324, R439, R521 NTU CSIE iAgent Lab 46 Valid Data From January 2011 to May 2011 Place January February March April May R104 36,525 82% 33,045 82% 36,958 83% 34,076 79% R204 39,135 88% 15,401 38% 28,123 63% 31,167 72% 13,958 31% R318 33,444 75% 32,053 79% 35,742 80% 31,978 74% 34,806 78% R324 30.993 69% 26,722 66% 29,890 67% 24,978 58% 29,212 65% R336 35,277 79% 28,872 72% 43,088 97% 39,920 92% 40,604 91% R439 39,681 89% 24,284 60% 35,212 79% 30,658 71% 34,158 77% R521 386,59 87% 34,927 87% 38,171 86% 26,992 62% 18,657 42% NTU CSIE iAgent Lab 6,072 14% 47 Professor Room - R318 State (April 2011) 0 : no people but A/C is turn on(abnormal) 1 : too cold (abnormal) Number Percentage Color 2,201 6.9% Yellow 242 0.8% Blue 1 0% Red 29,534 92.4% Green 2 : too hot (abnormal) 3 : others (normal) weekday distribution during a week NTU CSIE iAgent Lab weekend 48 Class Room – R104 State (April 2011) Number Percentage Color 628 1.8% Yellow 3,062 9.0% Blue 1 0% Red 30,386 89.2% Green 0 : no people but A/C is turn on(abnormal) 1 : too cold (abnormal) 2 : too hot (abnormal) 3 : others (normal) weekday distribution during a week NTU CSIE iAgent Lab weekend 49 Computer Class Room – R204 State (April 2011) 0 : no people but A/C is turn on(abnormal) 1 : too cold (abnormal) Number Percentage Color 5,582 17.9% Yellow 18,080 58.0% Blue 14 0% Red 7,491 24.0% Green 2 : too hot (abnormal) 3 : others (normal) weekend weekday NTU CSIE iAgent Lab 50 Class Room – R336 State (April 2011) 0 : no people but A/C is turn on(abnormal) Number Percentage Color 15,559 39.0% Yellow 0 0.0% Blue 5,360 13.4% Red 19,001 47.6% Green 1 : too cold (abnormal) 2 : too hot (abnormal) 3 : others (normal) weekend weekday NTU CSIE iAgent Lab 51 Seminar Room State (April 2011) R324 R439 R521 0 : no people but A/C is turn on(abnormal) 3.6% 8.6% 5.9% 1 : too cold (abnormal) 5.5% 2.9% 4.4% 2 : too hot (abnormal) 0.0% 0.0% 0.0% 90.9% 88.4% 89.6% 3 : others (normal) R324 R439 NTU CSIE iAgent Lab R521 52 Result Professor room • Cooling mode is too cold for the professor, so State 0 happens Class room • Administrator decreases a lot of abnormal states Computer class room • Students does not change the A/C mode even if State 1 happens Laboratory • State 0 happens often in midnight Seminar room • State 0 and State 1 takes about 10% NTU CSIE iAgent Lab 53 R204 and R336 R204 • State 1 takes up a big percentage in every month R336 • State 0 takes up a big percentage in every month • When the weather became warmer, state 2 would happen more frequently NTU CSIE iAgent Lab 54 Outline Introduction System Architecture Analysis Conclusion NTU CSIE iAgent Lab 55 Conclusion and Contribution Data collection : more than five months and continuously A/C mode recognition model : accuracy is higher than 85% Thermal comfort range : 19.32℃ and 24.67℃ Abnormal A/C states • Professor room : 7.7% • Class room : 10.8% • Computer class room : 75.9% • Laboratory : 52.4% • Seminar room : 10.3% NTU CSIE iAgent Lab 56 Future Work Improve the quality of the wireless sensor network Use persuasive technology to provide the results for users Recognize the activity level of NTU CSIE in each time interval NTU CSIE iAgent Lab 57 Thank You Q&A NTU CSIE iAgent Lab 58 2010/10/14 NTU CSIE iAgent Lab 59 NTU CSIE iAgent Lab 60 時間 入 01:00 X 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 出 X 7 3 1 1 3 10 59 90 86 77 155 173 144 101 33 67 175 85 24 24 16 4 5 6 1 2 16 24 50 84 190 88 69 133 41 204 167 71 40 88 累積人數(從06:00開始) 19 10 9 5 0 2 4 47 113 149 142 107 192 267 235 227 90 98 112 96 32 NTU CSIE iAgent Lab 61 Abnormal A/C State in NTU CSIE Ideal power consumption KWH NTU CSIE iAgent Lab 62 Abnormal A/C State in NTU CSIE Real power consumption A/C is turned off ? KWH NTU CSIE iAgent Lab 63 Deployment One server per floor (1F to 5F) Relays deploy around the corridors NTU CSIE iAgent Lab 64 標記 地點 時間 時間長度(分鐘) R104 11/14 00:00 – 11/15 17:27 2487 R113 11/12 14:17 – 11/15 17:21 4504 R204 11/13 11:43 – 11/15 10:35 2812 R318 11/13 11:34 – 11/15 17:22 3228 R324 11/13 12:33 – 11/15 10:34 2761 R336 11/12 08:47 – 11/15 11:45 4498 R439 11/13 11:30 – 11/15 10:33 2823 R513 11/12 14:27 – 11/15 14:23 4316 R521 11/14 14:55 – 11/15 17:30 1595 2010/12/09 NTU CSIE iAgent Lab 65 2010/10/14 NTU CSIE iAgent Lab 66 context data index dimensions Value 室內溫/濕度 (raw) 1-2 2 Float 出風口溫/濕度 (raw) 3-4 2 Float 室外溫/濕度 5-6 2 Float 冰水主機 7-9 3 {0,1} 出水溫度 10 1 Integer 泵浦轉速 11 1 Float 舊館/新館 12-13 2 {0,1} 樓層 14-18 5 {0,1} 房間類型 19-24 6 {0,1} 區域編號 25-30 6 {0,1} 建積 31 1 float 星期幾 32-38 7 {0,1} 周間/周末 39-40 2 {0,1} 學期中/寒暑假 41-42 2 {0,1} 小時 43-66 24 {0,1} 室內溫/濕度 (Interpolation) 67-68 2 Float 出風口溫/濕度 (Interpolation) 69-70 2 Float 室內溫/濕度 (Encode) 71-72 2 Float 出風口溫/濕度 (Encode) 73-74 2 Float 差距(室內, 出風口) 75-76 2 Float 差距(室內, 室外) 77-78 2 Float 差距(出風口, 室外) 79-80 2 NTU CSIE iAgent Lab Float 67 Dataset D={(xn,yn)}, where n=1 to N • each minute of labeled period (original : intersection of vent and indoor) • labeled by camera (original : controlled on purpose by duck) • size = 77,439 2010/12/01 NTU CSIE iAgent Lab 68 perhaps bringing up Structural Risk Minimization versus traditional Empirical Risk Minimization as it relates to the avoidance of local minima and overfitting NTU CSIE iAgent Lab 69 Zone(Experiment 1) SVM(linear) SVM(RBF) Additive Logistic Regression 204_1 97.6% 98.5% 98.4% 204_2 89.0% 89.1% 96.2% 204_3 99.8% 99.8% 99.8% 204_4 94.3% 97.9% 98.0% 204_5 99.9% 99.9% 99.9% 204_6 93.7% 93.9% 94.2% 336_2 93.2% 93.2% 93.4% NTU CSIE iAgent Lab 70 Cross-Validation : 3-Fold T(V) TH(V) TH(V), TH(I) TH(V), TH(I), TH(V), TH(I), TH(O), TH(O) AChost , ACdegree , ACspeed 204_1 0.68 0.80 0.84 0.96 0.98 204_2 0.70 0.85 0.86 0.97 0.99 204_3 0.68 0.82 0.83 0.97 0.97 204_4 0.71 0.84 0.88 0.97 0.98 204_5 0.87 0.87 0.88 0.97 0.99 204_6 0.60 0.63 0.81 0.96 0.98 336_2 0.87 0.87 0.87 0.96 0.98 Avg. 0.77 0.81 0.85 0.97 0.98 NTU CSIE iAgent Lab 71 NTU CSIE iAgent Lab 72 PPD=100-95*e-0.03353*PMV^4-0.2179*PMV^2 NTU CSIE iAgent Lab 73 Thermal Comfort Range NTU CSIE iAgent Lab 74 A/C State Abnormal • No people in the room but there exists at least one zone’s AC not closed • People in the room and there exists at least one zone where the AC is cooling mode and cooling below lower bound of the comfort range • People in the room and there exists at least one zone where the AC is cooling mode but warmer above upper bound of the comfort range Normal • Other states NTU CSIE iAgent Lab 75 R324 Event R104_T 0:不正常 (無人,空調開啟) 3.6% (898) 1:不正常(有人,空調開啟且過冷) 5.5% (1381) 2:不正常(有人,空調開啟且過熱) 0% (0) 3:正常(其他使用情形) 90.9% (22699) weekday weekend distribution during a week NTU CSIE iAgent Lab 76 Seminar Room – R439 (April 2011) Percentage State 2,650 (8.6%) 0 : no people but AC not closed (abnormal) 897 (2.9%) 1 : too cold (abnormal) 0 (0.0%) 2 : too hot (abnormal) 27,111 (88.4%) 3 : others (normal) distribution during a week weekday NTU CSIE iAgent Lab weekend 77 R521 Event R104_T 0:不正常 (無人,空調開啟) 5.9% (1594) 1:不正常(有人,空調開啟且過冷) 4.4% (1196) 2:不正常(有人,空調開啟且過熱) 0.0% (4) 3:正常(其他使用情形) 89.6% (24198) weekday weekend distribution during a week NTU CSIE iAgent Lab 78