pre-allocation of appliances in grid [performance]

[Bramtimm]

In power-grid-model, if we are extending the loads of the grid in the future, is it more performant to include a SymLoad (or AsymLoad) object for every node in the topology with zero-valued p_specified/q_specified for inactive nodes, or to only include load objects for nodes that carry an actual load?

Specifically: does PGM benefit from a static, pre-allocated input array of uniform size for appliances (all nodes, some zeroed), or does reducing the number of active load objects for the solver outweigh the overhead of dynamically copying/resizing/reindexing the input between calculations?

Hence, we are looking into setting loads for future electric vehicle charging points on the junctions in a grid, such that we do not have to dynamically grow the grid, but set the load relative to where it will probably will land.

[nitbharambe]

Hi @Bramtimm ,

is it more performant to include a SymLoad (or AsymLoad) object for every node in the topology with zero-valued p_specified/q_specified for inactive nodes, or to only include load objects for nodes that carry an actual load?
Specifically: does PGM benefit from a static, pre-allocated input array of uniform size for appliances (all nodes, some zeroed), or does reducing the number of active load objects for the solver outweigh the overhead of dynamically copying/resizing/reindexing the input between calculations?

Having additional inactive loads leads to almost no loss in performance.
And as you mention, there is indeed cost involved in first of all changing the input array and then later creating new PowerGridModel instance in between calculations.
Hence, its a good strategy to create as many empty loads as needed and mark them inactive. Then in batch dataset, make them active when needed.
(Edit: Corresponding topic in documentation)

Lmk if that addresses your question.

[Bramtimm]

thanks, I think we know how to proceed then 👍🏼