Working with Multi-Network Data

There are occasions when it is desirable to co-optimize multiple networks, these networks might encode time for dynamic network optimization, or scenarios for stochastic optimization.

To distinguish between network data (see The Network Data Dictionary) that correspond to a single network or to multiple networks, PowerModels.jl provides the function ismultinetwork(). For example, we can do the following:

network_data = PowerModels.parse_file(case3file)
pm = instantiate_model(network_data, ACPPowerModel, PowerModels.build_opf)


PowerModels.jl would generally not read in network data as multi-networks. To generate multiple networks from the same network data, we use the following method


Turns in given single network data in multinetwork data with a count replicate of the given network. Note that this function performs a deepcopy of the network data. Significant multinetwork space savings can often be achieved by building application specific methods of building multinetwork with minimal data replication.


For example, we can make three replicates by calling

network_data3 = PowerModels.replicate(network_data, 3)

Observe that the structure of network_data3 is different from that of network_data, since it is a multi-network. The user can then modify each replicate of the network to vary in the corresponding parameter of interest. See test/common.jl for examples on setting up valid Multi-Network data.

To build a PowerModel from a multinetwork data dictionary (see Building PowerModels from Network Data Dictionaries), we supply multinetwork=true during the call to build_generic_model and replace build_opf with build_mn_opf,

pm3 = PowerModels.instantiate_model(network_data3, ACPPowerModel, PowerModels.build_mn_opf, multinetwork=true)


The replicate() method only works on single networks. So

data33 = PowerModels.replicate(data3, 3)

will result in an error. Moreover, instantiate_model() (see )

Because this is a common pattern of usage, we provide corresponding calls to solve_mn_opf (which behaves analogously to solve_opf, but for multinetwork data).


Working with Multi-Networks is for advanced users, and those who are interested should refer to src/prob/test.jl for toy problem formulations for multi-network and multi-conductor models.