data treatment
Download
Skip this Video
Download Presentation
Data treatment

Loading in 2 Seconds...

play fullscreen
1 / 37

Data treatment - PowerPoint PPT Presentation


  • 78 Views
  • Uploaded on

Data treatment . Collect 257 days, (2010/12~2011/1, 2011/3/15~2011/9) Outage raw data 475 times Case I: timer setting => 253 times Case II: insufficient radiation => 73 times Case III: system/ human error => 149 times Determine whether interpolation or not Data treatment algorithm.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Data treatment' - cecil


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
data treatment
Data treatment
  • Collect 257 days, (2010/12~2011/1, 2011/3/15~2011/9)
  • Outage raw data 475 times
    • Case I: timer setting => 253 times
    • Case II: insufficient radiation => 73 times
    • Case III: system/ human error => 149 times
  • Determine whether interpolation or not
    • Data treatment algorithm
data treatment algorithm
Data treatment algorithm

Tstart: start time of outage, Tend: end time of outage, Tdur=Tstart - Tend

T={12:00, 13:00, 5:00, 7:00}, τ1= 10min, τ2=5min

Voff: the last voltage before outage Von: the first voltage after outage

Rd1<-- 1 day total radiation, Rd2 <-- 2 day total radiation

S(AB) <-- △VoltageAB/ △tAB, S(BC) <-- △VoltageBC/ △tBC

========================================

Case1

IfTstart = T{i} ± τ 1 ANDTdur < τ2 then interpolate

Included;

Case2

if {Voff <11.6 AND Von>12.0 AND date < 2010/3/14}

OR {Voff <11.2 AND Von>11.5 AND date > 2010/3/14}

if 4.53

else Included ;

Case 3

if |S(AB)-S(BC)|>0.015, then discard

elseifTstart ∩ {t| 7pm

else interpolate

Included;

outage

V

B

C

A

D

t

modeling
Modeling
  • B(t) = B(t-1) + λ(t) – μ(t) ,if λ(t) – μ(t) <24

Bmax, drop(t) = λ(t) – μ(t)-24 ≧ 0

  • B(t): energy in battery
  • λ(t): energy from radiation
  • μ(t): energy consumption
  • Bmax: energy in fully charged battery

B(t)

λ(t)

μ(t)

derive t t
Derive [ λ(t) – μ(t)]
  • Collect continued 2 days CASEII outage: 35
  • Lifetime vs Radiation w/o battery effect
  • Lifetime(hour) = 1.8516x(Radiation) + 0.7029
derive b max
Derive Bmax
  • Collect outage day due to insufficient radiation & work over last night : 18
  • Expected lifetime(hr) by radiation
  • B(t-1) = actual lifetime – {λ(t) – μ(t)} < 15.78
testing b max 15 78
Testing Bmax =15.78
  • Collect outage day after continuous days which past nights
  • 6 data sets (4/4, 4/17, 5/13, 5/24, 5/28, 9/21, )
slide9
Lifetime(hour) = 1.8516x(Radiation) + 0.7029
  • Sufficient for past one day : radiation > 12.58
  • Outage data 18 + past w/ insufficient radiation 31 = 49 day
derive b t
Derive B(t)
  • Collect CaseII outage and occur in night:
  • Record Lifetime vs Battery voltage(V)
  • Lifetime = 30.896V2 -727.51V + 4284
modified modeling
Modified Modeling
  • B(t) = B(t-1) + λ(t) – μ(t) ; if B(t-1) + λ(t) – μ(t)

Bmax, drop(t) = λ(t) – μ(t); if B(t-1) + λ(t) –

μ(t) ≧ Bmax

  • Bmax = 25hr
preliminary analysis
Preliminary analysis
  • Case I (253)=> interpolate 108 +81days
  • Case II (73)=> discard 18 days
    • Data ambiguous
    • High amount radiation but short life time
  • Case III (149)=> discard 8+9 days
    • Empty:2 +3 (12/29 30, 8/5 8 9 )
    • manual:3 (6/7, 7/16, 8/18)
    • unknown:1 (5/11)
    • no log:4 (12/10 14, 1/7, 3/14)
  • Keep 224 out of 257 days for modeling
data treatment algorithm1
Data treatment algorithm

Tstart: start time of outage, Tend: end time of outage, Tdur=Tstart - Tend

T={12:00, 13:00, 5:00, 7:00}, τ1= 10min, τ2=5min

Voff: the last voltage before outage Von: the first voltage after outage

Rd1<-- 1 day total radiation, Rd2 <-- 2 day total radiation

S(AB) <-- △VoltageAB/ △tAB, S(BC) <-- △VoltageBC/ △tBC

========================================

IfTstart = T{i} ± τ 1 ANDTdur < τ2 then interpolate

else if {Voff <11.6 AND Von>12.0 AND date < 2010/3/14} OR

{Voff <11.2 AND Von>11.5 AND date > 2010/3/14}

if 4.53

then discard

else if |S(AB)-S(BC)|>0.015, then discard

elseifTstart ∩ {t| 7pm

else interpolate

outage

V

B

C

A

D

t

iii outage
III outage
  • Occur 42 days in 165 days
    • Reboot: 6
    • Error: 3
    • Empty: 13
    • Manual: 3
    • Unknown: 4(5/7, 9, 10, 11)
    • No log: 12
slide21
8 days filtered in 42 days
    • Empty:2(12/29 30),
    • manual:1(6/7),
    • unknown:1(5/11),
    • no log:4(12/10 14, 1/7, 3/14)
    • 10,14,29,30 Dec., 7 Jan., 14 Mar., 11 May, 7 Jun.
slide22
開始

III類斷電

斷電前後電壓是否符合控制器設定

抓出每天斷電資料

No

斷電前後斜率差是否<0.015

No

Yes

斷電時間是否符合timer 設定

No

Yes

II類斷電

斷電時間是否跨越晚上(7:00pm~6:00am)

Yes

Yes

No

前一天日照累積是否>4.53

I類斷電

No

Yes

不能採用的資料discard

前二天日照累積是否<12.61

No

interpolation

interpolation

Yes

不能採用的資料discard

可以採用的資料

slide25
smooth最大斜率差0.015

=>只要斜率差>0.015,挑出來,不能用

related works
Related works
  • Design, modeling and capacity planning for micro-solar power sensor networks, IPSN08
  • Cloudy computing: leverage weather forecasts in energy harvesting sensor systems, secon10
  • Solarstore: enhancing data reliability in solar-powered storage-centric sensor network, mobisys09
  • Steady and Fair rate allocation for rechargeable sensors in perpetual sensor networks, Sensys08
  • Networking low-power energy harvesting devices: measurements and algorithm, infocom11
slide27
Mobisys09
    • 計算是否有多餘電量來讓系統備份data
    • If 目前電池電量>(平均消耗功率/平均充電功率)x電池容量, 就可以保持供電
    • Evaluation data loss (無備份,always備份,電池滿才備份)
slide28
Sensys 08
  • 用太陽能收到多少來決定data collection rate
  • 讓node都不會因為傳太快而沒電
  • Centralized/distributed algorithm
slide29
Infocom11
    • 室內光產生電源,實際測量8個環境
    • Model 分三類:可預測,半預測,隨意型
    • Ep(i) = a x E(i-1) +(1-a)Ep(i-1), 0
    • 但誤差還是很大
    • 所以在某些環境採用SECON10 天氣model,誤差比較小
ipsn08
IPSN08
  • Solar Model
  • Network & HW design
    • 19nodes for collect/receive 溫濕度
  • Sizing and selection
    • 實際去量測每個component profile
    • 比較用不同電池壽命多久
  • Evaluation for two test deployments
    • 市區佈建 VS 森林佈建
secon10
Secon10
  • 分析一整年的資料,搭配氣象局資料
  • 日照輻射(W/m2)與充電功率(W)=>線性關係
    • Solar power = 0.0444*radiation -2.65
  • 每日充電能量(W) vs時間(hr)=>2次方程式
    • Power = a*(Time+b)^2 + c
slide32
Solarstore: enhancing data reliability in solar-powered storage-centric sensor network, mobisys09
  • Networking low-power energy harvesting devices: measurements and algorithm, infocom11
outage iii algorithm
Outage III algorithm
  • if (abs(mAB-mBC)>Threshold) || (abs(mBC-mCD)>Threshold)

filter out

Threshold = 0.019

  • else if(13V<斷電前電壓) &&(duration > 19600 )

filter out

  • else if(12V<斷電前電壓<12.5V) &&(duration > 10800 )

filter out

  • else if (斷電前電壓<12V) &&(duration > 3600)

filterout

how to filter for iii
How to filter for III類
  • 一階微分用斜率來判斷
    • |mAB-mBC|< δ
    • |mBC-mCD|< δ
  • 二階微分
    • |F’’ABC- F’’BCD| < δ

D

C

B

A

C

B

D

A

ad