Setting up the autoregressive processes

For both, log-linear and linearized AR processes, the processes are set up using functions in set_up_ar_process.jl.

As inputs for this set-up three sources are used:

• For each reservoir system (N, S, NE, SE) and for each month, a log-linear AR model was fitted using the historical data (for more details, see model fitting). The estimates are stored in folder AutoregressivePreparation and then bic_model (or custom_model) in some txt-files (for the linear case it is LinearizedAutoregressivePreparation). Each txt-file contains the months, the lag order, the process intercept, the process coefficients, the factor multiplied with the error term and the standard error from the estimation.
• For each reservoir system, some historical inflow values are provided in the same folder in the txt-file history_nonlinear.txt. It contains a sufficient number of time steps for the lag order of the processes.
• For each reservoir system and each stage, 100 realizations of the stagewise independent noise term $\eta_t$ are provided in the same folder in the txt-file scenarios_nonlinear.txt.

In the set-up process, first the above data is read from the source files. Then it is used to create structs of type AutoregressiveProcessStage and AutoregressiveProcess.

As loglinearSDDP requires a constant lag order, the maximum over all reservoirs and months is used.

This page was generated using Literate.jl.