jbae11 / ann_pwr

Cyclus batch-wise reactor module using neural network depletion model

ANN LWR

This Cyclus Reactor module is created to perform varying burnup and enrichment calculations for LWRs using a trained neural network model to predict UNF composition.

The reactor module will behave similarly to the Cycamore batch-recipe reactor, but it will deplete fuel with user-input burnup.

The user defines the enrichment and burnup matrix, as a list of strings separated by spaces, where each entry is a enrichment_burnup format. For example, for a core with 24 assemblies and 8 assemblies per batch, the input is the following:

<enr_bu_matrix>
    <val>2.1_30000 3.1_30000 3.1_30000 3.1_30000 3.1_30000 4.1_30000 4.1_30000 3.1_30000</val>
    <val>3.1_32000 3.1_34000 3.1_40000 3.1_36000 3.1_39000 4.1_32000 3.5_40000 3.2_36000</val>
    <val>2.1_33000 3.5_34000 3.2_43000 2.8_43100 3.1_28000 4.5_42000 3.1_23555 3.1_30000</val>
</enr_bu_matrix>

The numbers could also be equations. For example, if the user wanted the burnup of the equilibrium fuel to increase by 1% every month:

<enr_bu_matrix>
    <val>2.1_30000 3.1_30000 3.1_30000 3.1_30000 3.1_30000 4.1_30000 4.1_30000 3.1_30000</val>
    <val>3.1_32000 3.1_34000 3.1_40000 3.1_36000 3.1_39000 4.1_32000 3.5_40000 3.2_36000</val>
    <val>2.1_33000*(1.01)**t 3.5_34000*(1.01)**t 3.2_43000*(1.01)**t 2.8_43100*(1.01)**t 3.1_28000*(1.01)**t 4.5_42000*(1.01)**t 3.1_23555*(1.01)**t 3.1_30000*(1.01)**t</val>
</enr_bu_matrix>

Currently, only integer batches are accepted.

When decommissioning, the reactor will asses how long the fuel has been in the reactor and will reduce the burnup accordingly.

Also, the user can choose to define the cycle time and refuel time as a function of simulation time by writing the cycle_time_eq and refuel_time_eq variable as a function of t. For example, if the user wants the cycle time to linearly increase with time,

<cycle_time_eq>18+t/100</cycle_time_eq>
<refuel_time_eq>1+t/100</refuel_time_eq>

This input will increase the cycle time and refuel time by 1 every 100 timesteps. Note that the t here is simulation time, not reactor age.

This module requires a pickled file with the trained ann model. The pickled file, when imported, is a dictionary with keys:

• model: Keras trained model object
• xscaler: sklearn.preprocessing.MinMaxScaler object to transform [enrichment, burnup] to standard values for model
• yscaler: sklearn.preprocessing.MinMaxScaler object to transform model prediction results to standard value [% of fuel]
• iso_list: list of isotope names for model prediction (index matches y indexes)

Inputs:

• fuel_incommod: commodity name for incoming fuel
• fuel_outcommod: commodity name for outgoing fuel
• pickle_path: absolute path of the pickle file- n_batch: number of batches for reactor
• enr_bu_matrix: vector of strings separated by space. Shape must be n_assem_core / n_assem_batch X n_assem_batch
• assem_size: fuel mass per assembly in kg
• power_cap: power produced by reactor when operational (units arbitrary)
• cycle_time_eq: operational cycle time of reactor equation
• refuel_time_eq: time equation for reactor to refuel, reactor is not operational during this time.

Dependencies:

• keras
• numpy
• cyclus
• sklearn
• pickle

ann_pwr's Issues

unf assemblies are lumped into one material

Currently, the used fuel are lumped together at get_material_trades.

There has to be a way to transact discrete assemblies instead of lumping all of the assemblies into one