Control data in tilsigsddp.CPAR#
Control data for the stochastic inflow model used in the backward recursion in Prodrisk is specified in the file tilsigsddp.CPAR. If this file does not exist, a file containing the default values named Std_tilsigsddp.CPAR is created. The user can make adjustments to the default parameters in Std_tilsigsddp.CPAR and rename the file to tilsigsddp.CPAR so that the new values are read by Prodrisk.
Prodrisk can use three different types of inflow models in the backward recursion, based on either:
Lognormally distributed noise
Principal component analysis
Sampling from residuals (“residual model”)
The lognormal model is the default choice, see Control data in prodrisk.CPAR.
3, NPRBRUK – no. of principal components(1 < NPRBRUK < number of inflow series)
3, 2, 2, NRSTKOM(1:NPRBRUK) – no. of discrete values per principal component. The number of noise alternatives for the lognormal and residual inflow models is also decided by these numbers: The number of noise alternative is the same as the number of discrete values per principal componets multiplied with each other. E.g: If NPRBRUK == 3, then the number of noise alternatives is 322. 3 alternatives for the first principal component, 2 for the second and 2 for the third. Although the lognormal model and residual model do not use principal components, they will also use 12 noise alternatives.
1, NSESONG – no. of seasons (1 < NSESONG < no. of weeks)
0 0, SESONG(0:NSESONG) – first season start and all season endings
0, 0, 1, -1 NAARSIM, ITILFSTART, IKLIPPNEG, I
If NSESONG = 1, line number 4 can have the values 0 0. A negative seasonal ending indicates that the correlation matrix should be zeroed for that season.
The last line (defining NAARSIM, ITILFSTART, IKLIPNEG) defines parameters used in the inflow model which is based on sampling from residuals. The last value I should always be -1 indicating end-of file.
The parameters specified in tilsigsddp.CPAR are specified below, with default values in parenthesis.
NPRBRUK (3)
The number of axis for which the inflow model is developed. The default value is 3. The default value is automatically reduced to NSER (number of inflow series) if NSER < 3.
NRSTKOM (3,2,2)
Number of discrete noise values along each principal component. Prodrisk expects NPRBRUK values on this line, i.e., one for each principal component. If the values 2 or 3 are used, the discretization is done to optimally retain the statistical moments. If the value is 4 or higher, a simpler discretization is used.
The number of discrete noise values used in the backward recursion equals the product of the number of discrete noise values for each principal component. The computation time therefore depends heavily on the discretization.
The default values are (3,2,2). Three discrete noise values are then found for the first (and most important) principal component, and two discrete values are used for the last two components. The user is encouraged to experiment with this setup to find a balance between inflow model accuracy and computation time.
3 discrete values will give a rather high weight on the mean value, while the two other values are far from the mean with low probabilities.
2 discrete values will give realizations distant from the mean value, but not as distant as the two extremes when using 3 discrete values.
Note that in the NPRBRUK and NRSTKOM are not directly relevant for the residual model. However, they together define the number NSTOY of residuals being sampled. NSTOY is not specified directly, but as the product of all NPRBRUK values in NRSTKOM. As an example, we can specify NSTOY=15 by setting NPRBRUK=2, NRSTKOM(1)=5, and NRSTKOM(2)=3.
NSESONG (1)
Both auto- and cross-correlations in inflow can vary significantly over a whole year. Thus, the period of analysis can be divided into seasons, where each season will have a separately computed correlation matrix. Be aware that increasing NSESONG results in less data per season to estimate the parameters from.
SESONG (0 0)
This line should read NSESONG+1 values, specifying the first season start and all subsequent season endings. If NSESONG=1, the values 0 0 can be given, indicating that the start and end of season will be defined by the inflow series. Typically, seasons are defined for one year (until week 52), but can be set until the last week in the simulation period. If the residual model is used with versions until 10.3.0, it is mandatory that the end of the last season equals NUKE, the total number of weeks in the simulation. For version 10.3.1 and later, the seasons of the first year are repeated if seasons are defined until a week smaller than or equal to 52. If the last defined week ends in a week larger than 52 but smaller than NUKE, one final season is added that spans from the end of the last defined season until the end of the simulation (no repetition of seasons).
Example:
3, NSESONG
1, 17, 30, 52
A negative season ending indicates that the correlation matrix should be zeroed for that season. If so, the serial- and cross-correlations are not considered. In the example below, the correlation matrix is zeroed during the snow melting.
Example:
3, NSESONG
1, 17, -30, 52
NAARSIM (0)
Prodrisk allows generation of simulated inflows using both types of inflow models (principal components and residual). If NAARSIM > 0, NAARSIM inflow years will be generated, using the mean weekly value of all historical inflow years as a starting point.
ITILFSTART (0)
If ITILFSTART = 0, Prodrisk will generate the same inflow sequence every time the program is run, provided that all relevant data are unchanged. If ITILFSTART = 1, the system clock provides the seed for the random generator, so that the sequences will vary. If ITILFSTART > 1, ITILFSTART itself provided the seed to the random generator, and thus allow the user to generate different series that are reproducible.
IKLIPPNEG (1)
This parameter is only relevant for the residual model, and indicates if large negative inflows should be omitted. The default (IKLIPPNEG = 1) is to omit large negative inflows in the residual model.