Differences
This shows you the differences between two versions of the page.
| Next revision | Previous revision | ||
|
forecasting:meeting_minutes_october_25_2016 [2016/10/26 02:37] jobatake created |
forecasting:meeting_minutes_october_25_2016 [2021/09/19 21:59] (current) |
||
|---|---|---|---|
| Line 1: | Line 1: | ||
| - | ====== Forecasting Meeting Minutes for Week of October 18, 2016 ====== | + | ====== Forecasting Meeting Minutes for Week of October 25, 2016 ====== |
| - | ** Present Monday: Austin, Gordon, Brieanna, Jaimie ** | + | ** Present Tuesday: Austin, Gordon, Brieanna, Jaimie ** |
| - | ** Present Tuesday: Gordon, Brieanna, Jaimie ** | + | ** Present Wednesday: ** |
| ===== Updates ===== | ===== Updates ===== | ||
| - | * Ask Sharif to meet with us | + | * Sharif will meet with us on Wednesday |
| ===== Meeting with Dr. Kuh ===== | ===== Meeting with Dr. Kuh ===== | ||
| ===== Expectations ===== | ===== Expectations ===== | ||
| - | * 3D plot: 2011, 2012 normalized and unnormalized for comparison | + | * All of this is normalized (w/ Zenith Angle) |
| - | * LA more variation, but smoother | + | * X' = [X / 1 1 1 1 1 1] |
| - | * Moving Average and Decimation/Downsample | + | * X'.T = [X.T | 1] |
| - | * D and X from 2011: | + | * X'*X'.T = [__X*X.T__ | **X1**] |
| - | * X = 8:00-4:00 (input data) | + | * [1.T*X.T | m ] |
| - | * D = 9:00-5:00 (normalized and downsampled, column) | + | * (X'*X'.T)/m |
| - | * Y = Least Squares Equation = W.T(X) row = X.T(W) column | + | *** Mean**: X1/m |
| - | * Unnormalized = D bar, Y bar | + | *__ Correlation(R)__ : X*X.T/m |
| - | * Statistical: | + | * Samples along diagonal of correlation matrix should be similar |
| - | * Y - sig*Y + m | + | * Like toeplitz matrix |
| - | * D - sig*D + m | + | * Right-hand bottom corner needs to be 1 |
| - | * Zenith: | + | * Number of columns (m) in array rows (n) |
| - | * Y*cos(theta) | + | * Number of times you are doing prediction |
| - | * D*cos(theta) | + | * Estimate 9-5 (8 hours, every 5 mins, 96 solar irradiance times/day *365 days/year= 35040 solar irradiance data points/year) |
| - | * W = (XX.T)^(-1)*XD <= XX.T*W = XD | + | * Diagonals same and symmetrical |
| - | * E^2 Averaged = (D-Y)^2 for every time | + | * 1's on bottom are for the zenith normalized data |
| - | * MSE = 1/m (||D-Y||)^2 | + | * Do not need for the statistically normalized data |
| - | * Don't need r bc cancel | + | |
| - | * Ask Seyed what he did for decimation/downsample | + | |
| - | * 1) Average and down sample | + | * Types of Prediction: |
| - | * 2) Take moving average | + | * Average |
| - | * Overlapping data points: | + | * Persistence: Most recent value for prediction (tap=1) |
| - | * 8:01-8:10 moving average | + | |
| - | * Then downsample | + | * Python programming help from Sharif tomorrow |
| - | * Filter then normalize | + | * MSE value: D-Y=E(365*96) |
| - | * <del>Normalize then filter</del> | + | * Inner product of E/entries = 2D array |
| - | * **Normalization: mean then deviation** | + | * (Sum |E|^2)^.5 |
| - | * Zenith Angle Filter: Zenith angle is from the 5th minute, 5 irradiance is averaged | + | * Root Mean Squared Working this week |
| - | * Bias term | + | * Training to use same data for RMS |
| - | * Number of taps is # present + number before | + | * Go down with number of taps increasing |
| - | * 1 tap = present | + | * Take square root of the error |
| - | * 2 tap = present + future (5 mins in our case) | + | * (90,000)^.5 = 300 |
| - | * In weight matrix | + | * Weight for taps |
| - | * Row of 1's to the matrix to get bias term | + | * Positive values |
| - | * Statistical normalization: -mean/deviation | + | * Decrease in value |
| - | * Standard normalization 0 mean | + | * Train on 2011 and test on 2012 |
| - | * Estimate will need to be given back after the normalization | + | * Train on 2012 and test on 2011 |
| - | * 3D plot for both zenith and stat normalization | + | * RMS graphs for the 2011 (Test and train) and 2012 (test and train) |
| - | * Plot MSE | + | * Higher on test than training |
| - | * y = MSE, x = # taps | + | * Can become higher though |
| - | * Train error goes down, test error decreases until flattens or increases | + | * More time then train based on seasons |
| - | * Check 2012 MSE test against 2011 MSE train | + | * Winter 2011/2012, Spring 2011/2012, Summer 2011/2012, and Fall 2011/2012 |
| - | * Keep 2011 W | + | * Interested in power solar panels will be making |
| - | * Get new X, D to compute new Y | + | |
| - | * Flip: 2011 MSE test against 2012 MSE train | + | |
| - | * Correlation is 3D plot | + | |
| - | * R=(1/m)XX.T= 10x10 matrix that is symmetric | + | |