windprediction/notes at master · npollard/windprediction · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
5/27/2014:
  - wind from 2012/2013 into test.m (as bus/gen/etc.)

  - train on alternate days
  - test vs remaining

  - search for wind dataset with specific locations
    - surfline
      - uses wunderground
    - power people
    - meteorologists

5/30/2014:
- start with prediction for a single station
  - two node BN: now/5 min from now
- WEKA
  - alternative to MATLAB
  - visualization
  - more (different) machine learning algorithms

6/4/2014:
- wunderground script
  - grabs from 12/31/2011 to 5/31/2014 for a hardcoded station
  - have KCASANTA132 (the wharf) and KCASANTE9 (somewhere in southern california)
- WEKA
  - installed, reading up on it
  - use CSV, but convert to ARFF
  - integrate with Java

6/11/2014:
- WEKA
  - walkthrough
  - according to doc, GUI works for basic usage but command line is recommended
- recommendation for brief overview of different algorithms? neural networks, Markov, Gaussian, etc.
- regression models
- Adaptive Gaussian Process for Short-Term Wind Speed Forecasting
  - NY area wind speed, Calamut, IL area wind speed
  - 10 minute interval
  - one year
  - their "adaptive gaussian process" outperformed Mycielski, ARMA, 2nd order Markov chain
- Data Mining II: Regression Analysis
  - WEKA
  - considered neural network, support vector machine, k-nearest neighbor
  - sort of just a walkthrough of the WEKA GUI
- A Decision Tree for Weather Prediction
  - WEKA
  - one year of pressure, humidity, temp, etc. in Hong Kong to predict temp
  - walks through how to nominalize date, temp values
  - used trees.SimpleCart algorithm
- Data mining and wind power prediction: A literature review
  - lists experiments done: what algorithm on what interval, etc.
  - very-short term: predict seconds to 30 min ahead
    - tested on mostly 5 or 10 minute interval datasets
    - artificial neural networks and adaptive neuro-fuzzy inference give best results out of those surveyed
    - however, problems listed with these studies
  - short term: predict 30 min to 6 hours ahead
- Short-Term Wind Power Forecasting Using Gaussian Processes
  - one day ahead prediction
  - Numerical Weather Prediction corrected by Gaussian Process
  - statistical models good for 1-4 hour ahead prediction (GP)
  - physical models good for up to dozens of hours (NWP)
  - use physical (NWP) as input for statistical (GP) -> day ahead predictions
- TODO:
  - start with simple regression, predicting 5 minutes out (f(x) = ax+b)
  - CART - decision tree into regression
    - if the wind is NW, then a1, b1

6/13/2014:
- BNT
  - Gaussian Bayesian Network
- CART

6/18/2014:
- BNT is for MATLAB
  - http://people.cs.ubc.ca/~murphyk/Software/bnsoft.html
- TODO:
  - look at GUI
    - SimpleCart
  - email data
  - multiple stations
  - wind direction
  - BNT
    - see if in WEKA
    - http://bnt.googlecode.com/svn/trunk/docs/usage.html#cg_model
  - use 5 minute datasets only
  - time of day

6/25/2014:
- wind direction 5 minutes from now

7/5/2014:
- KCADAVEN4 not nominalizing
  - "No upper limit has been specified for range"
- KCASANTA239 has numeric direction
- KCASANTA254 has issues
- runs out of heap space:
  - naive_2
  - naive_3
  - naive_4
  - naive_dir