Presentation is loading. Please wait.

Presentation is loading. Please wait.

“Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model only Network 4 nodes MCU Power management in energy harvesting sensor networks,”

Similar presentations


Presentation on theme: "“Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model only Network 4 nodes MCU Power management in energy harvesting sensor networks,”"— Presentation transcript:

1 “Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model only Network 4 nodes MCU Power management in energy harvesting sensor networks,” ACM Trans. Embed. Comput. Syst., 2007 Model driven NetworkMCU “Design, modeling, and capacity planning for micro-solar power sensor networks.” IPSN’08 Model only Network 19 Tmotes MCU “Steady and Fair Rate Allocation for Rechargeable Sensors in Perpetual Sensor Networks.” SenSys’08 Model driven Network 150 Tmote MCU “SolarStore: Enhancing Data Reliability In Solar-powered Storage- centric Sensor Networks.“ MobiSys’09 Model driven NetworkEmbedded system ”Minimum Variance Energy Allocation for Solar-Powered System” DCOSS’09 Model driven Network 9 nodes Embedded system ”On the Limits of Effective Hybrid Micro-Energy Harvesting on Mobile CRFID Sensors.” MobiSys’10 Model only Individual system Embedded system “A Weather-Condition Prediction Algorithm for Solar-Powered Wireless Sensor Nodes” WICOM’10 Model only Individual system Embedded system Accurate Modeling and Prediction of Energy Availability in Energy Harvesting Real-Time Embedded Systems” Green Computing’10 Model only Individual system Embedded system Cloudy Computing: Leveraging Weather Forecast in Energy Harvesting Sensor System.” SECON10’ Model driven Network 5 TelosB MCU Networking Low-Power Energy Harvesting Devices: Measurements and Algorithms.” INFOCOM’11 Model driven Individual system/ Networks MCU

2 Networking Low-Power Energy Harvesting Devices: Measurements and Algorithm Gorlatova M., Wallwater A., Zussman G. INFOCOM 2011

3 Outline Introduction Model Measurement Energy profile Algorithm for predictable profile and stochastic profile Result

4 Introduction Achieve time-fair resource allocation since energy varies among different time Other research focus on fairness – Data generation rate (SenSys08) – among nodes (JSAC06) Proposed energy allocation algorithms across the different time slot to optimize – energy spending rate for single node – energy communication rate for a link Indoor irradiance measurements study.

5 Dimensions of algorithm design Environmental energy model Energy storage type Ratio of energy storage capacity to energy harvested Time granularity – nodes characterize received energy – Make decision from sec ~days Problem /Network size – energy harvesting affects nodes’ decisions – Link decisions, routing, rate adaptation

6 Environmental energy model Predictable profiles – Ideal – Accurate for the future Partially predictable Stochastic Model-free

7 Energy storage type Rechargeable battery – Ideal linear model – Changes of stored energy vs harvested power Capacitor – Non-linear model – Power harvested depend both on the energy provided and on the amount of energy stored

8 Contributions Indoor : partially-predictable energy model Time granularity : day, improving prediction Fair allocation of resources along the time – Energy spending rate for a node – Data rate for a link Predictable energy profiles – Lexicographic maximization – Utility maximization Stochastic model – Markov Decision Process

9 Model K slots time i = {0,1,…K-1} D=AηH Q(i)=q(D(i),B(i)) for capacitor

10 Objective Optimization energy spending rate s(i) for single node Optimization energy communication rate r u (i), r v (i) for a link Utility maximization frame work to find – spending rate s(i) – communication rate r u (i), r v (i) – α-fair function U(s(i)) = s(i) 1-α /(1-α), for α>0, α≠1 log(s(i)), for α=1 Consider predictable profile energy model

11

12 Dynamic programming-based algorithm h(i,B(i))=max[ U(s(i)) + h(i+1,min(B(i)+Q(i)-s(i),C))] Determine vector {s(0),…s(K-1)} maximizes h(0,B 0 ) Running time O(K[C/ △ ] 2 )

13 For linear storage q(D(i),B(i)) = D(i) Progressive Filling algorithm Running time O(K[K+Q T / △ ]),Q T = ΣQ(i)+(B 0 -B K ) If linear storage is large, s(i) = Q T /K

14 Measurement Long-term measurement of indoor irradiance Office buildings at Columbia Uni. since 2009/6 TAOS TSL230rd photometric sensors LabJack U3 DAQ

15 H d : mean of the daily irradiation σ : standard deviation r : bit rate, throughout a day when exposed H d – r = A(10cm 2 ) x η(1%) x H d /(3600x24)/(10e-9) – EnHANTs costs 1nJ/bit

16

17 How to predict H d ? Exponentially weighted moving-average (EWMA) – Error is relatively high For L-1, avg. prediction error > 0.4H d L-2, avg. prediction error > 0.5H d Outdoor, avg. prediction error = 0.3H d Weather forecast [secon10] may be improved – For L-1, correlation coefficient of H d and weather = 0.35 – For L-6, correlation coefficient =0.8

18 Work week pattern For L-2, student office, on shelf far from window H d = 1.63 on weekdays, Hd=0.37 on weekend 9.7hr/day lighting on weekday, <1hr on weekend Avg. error prediction error 0.5H d -> 0.26H d if separate weekdays and weekends. L-1 and L-5, correlation coefficient =0.58 L-1 and L-5 facing same direction, correlation coefficient =0.71

19 Short term energy profiles H T, T= 0.5 hr L-3, daylight- dependent variations L-2, either 0 or 45uW/cm 2 partially predictable energy model

20 Mobile measurements mobile device carried around indoor and outdoor locations Indoor : 70uW/cm 2 Outdoor : 32mW/cm 2 Poorly predictable Stochastic energy model

21 Link: optimizing Data rate

22 r u (i) = r v (i) = r(i)

23 Extension of TFR algorithm {r u (0),..r u (K-1)}, {r v (0,..r v (K-1)} maximize h(0,B 0u,B 0v ) Complexity : For linear storage, LPF algo.,

24 Decoupled Rate Control algorithm DRC algorithm Determine s u (i), s v (i) independently using PF algorithm r(i) = min(s u (i), s v (i))/(c tx + c rx )

25 Stochastic Energy model Energy harvested in a slot is and i.i.d (identical independent distribution)random variable D [d 1,..d M ] with probability [p 1,…p M ] Spending Policy Determination (SPD) problem – Given distribution D, determine s(i) Markov Decision Process(MDP)

26 Apply dynamic programming, from i=K-1 for each {i,B(i)} For each storage B(i), s(i) approached optimal Running time

27 Link Spending Policy Determination Problem(LSPD) Apply dynamic programming For each {i, B u (i), B v (i)} Maximization is over all {ru(i), rv(i)}, such that c tx r u (i) + c rx r v (i) = s u (i) ≤ B u (i) c tx r v (i) + c rx r u (i) =s v (i) ≤ B v (i) Complexity: O([C u / △ ] 2 [C v /] 2 M u M v K)

28 Numerical results Energy profile L-3 input s(i) are obtained by PF algorithm for linear storage(left) s(i) for TFU problem for nonlinear storage (right)

29 L-1, L-2 energy profile Optimal communication rate {ru(i), rv(i)}

30 Optimal energy spending policies (SPD) L-1 profile as random variable D Optimal s(i) Optimal communication rate r u (i), r v (i)

31 Conclusion First long-term indoor radiant energy measurements campaign that provides useful traces Developed algorithms for predictable environment that uniquely determine the spending policies for linear and non-linear energy storage models Developed algorithms for stochastic environments that can provide nodes with simple pre-computed decisions policies

32 “Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model only Network 4 nodes MCU Power management in energy harvesting sensor networks,” ACM Trans. Embed. Comput. Syst., 2007 Model driven NetworkMCU “Design, modeling, and capacity planning for micro-solar power sensor networks.” IPSN’08 Model only Network 19 Tmotes MCU “Steady and Fair Rate Allocation for Rechargeable Sensors in Perpetual Sensor Networks.” SenSys’08 Model driven Network 150 Tmote MCU “SolarStore: Enhancing Data Reliability In Solar-powered Storage- centric Sensor Networks.“ MobiSys’09 Model driven NetworkEmbedded system ”Minimum Variance Energy Allocation for Solar-Powered System” DCOSS’09 Model driven Network 9 nodes Embedded system ”On the Limits of Effective Hybrid Micro-Energy Harvesting on Mobile CRFID Sensors.” MobiSys’10 Model only Individual system Embedded system “A Weather-Condition Prediction Algorithm for Solar-Powered Wireless Sensor Nodes” WICOM’10 Model only Individual system Embedded system Accurate Modeling and Prediction of Energy Availability in Energy Harvesting Real-Time Embedded Systems” Green Computing’10 Model only Individual system Embedded system Cloudy Computing: Leveraging Weather Forecast in Energy Harvesting Sensor System.” SECON10’ Model driven Network 5 TelosB MCU Networking Low-Power Energy Harvesting Devices: Measurements and Algorithms.” INFOCOM’11 Model driven Individual system/ Networks MCU

33

34

35 Model only/model driven objectiveNetwork/ individual system MCU / embedded P. Corke, et al. “Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model onlyDeployed 2 years, estimation component efficiency Network 4 sensor nodes MCU J. Taneja, et al. “Design, modeling, and capacity planning for micro- solar power sensor networks.” IPSN’08 Model onlyBattery comparison and experience of deployment Network 19 Tmotes MCU K.Fan, et al., “Steady and Fair Rate Allocation for Rechargeable Sensors in Perpetual Sensor Networks.” SenSys’08 Model drivenDeployed outside for 2 month maximum data collection rate Network 150 Tmote MCU

36 Model only/model driven objectiveNetwork/ individual system MCU / embedded Y. Yang,et al. “SolarStore: Enhancing Data Reliability In Solar-powered Storage-centric Sensor Networks.“ MobiSys’09 Model drivenData replication, outdoor/ indoor testbed NetworkEmbedded system D.Noh, et al, ”Minimum Variance Energy Allocation for Solar- Powered System” DCOSS’09 Model drivenEnergy allocation for time slots(1hr) Network 9 nodes Embedded system C.Moser, et al., “Power Management in Energy Harvesting Embedded Systems with Discrete Service Levels” ISLPED’09 Model drivenOptimization for scheduling by dynamic programmin g Individual system Embedded system Jeremy G., et al. ”On the Limits of Effective Hybrid Micro-Energy Harvesting on Mobile CRFID Sensors.” MobisSys’10 Model onlyPower consumption benchmark Individual system Embedded system

37 Model only/model driven objectiveNetwork/ individual system MCU / embedded Zhaoting J., et al. “A Weather- Condition Prediction Algorithm for Solar-Powered Wireless Sensor Nodes” WICOM’10 Model onlyEWMA vs Weather Conditioned EWMA Individual system simulation Embedded system Jun. L., et al.”Accurate Modeling and Prediction of Energy Availability in Energy Harvesting Real-Time Embedded Systems” Green Computing’10 Model onlyPrediction of harvested energy Individual system simulation Embedded system N.Sharma, et al. ”Cloudy Computing: Leveraging Weather Forecast in Energy Harvesting Sensor System.” SECON’10 Model drivenForecast weather energy model, lexicographic ally fair rate Network 5 TelosB MCU M.Gorlatove, et al.”Networking Low-Power Energy Harvesting Devices: Measurements and Algorithms.” INFOCOM’11 Model drivenDerive algorithm for energy spend rate/ link rate Individual system/ Network simulation MCU

38 Alpha ->0, globally optimize Alpha ->1, proportional fairness ->2, harmonic mean fairness -> infinite, generalized max min allocation

39 Introduction Perpetual energy harvesting wireless device – Solar, piezoelectric and thermal energy harvesting – Ultra-low-power wireless communication – Rechargeable sensor networks Focus on devices that harvest environmental light energy Energy-harvesting-aware algorithm and system – Lack of data and analysis of energy availability for indoor/outdoor environment – 16 months indoor light energy measurement light measurement and resource allocation


Download ppt "“Long-duration, Solar-powered Wireless Sensor Networks.” EmNets’07 Model only Network 4 nodes MCU Power management in energy harvesting sensor networks,”"

Similar presentations


Ads by Google