Scalably generating distribution grid topology

A system comprising one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to perform operations for generating a composite representation of an electric power grid, the operations comprising:

• obtaining two or more datasets comprising data indicating a presence of or operation of assets of the electric power grid in a geographical region that has the electric power grid, each dataset including data of a particular kind, wherein the assets comprise utility poles and connecting lines between the utility poles;
• identifying one or more respective predictive models for each dataset, each predictive model being configured to process the respective kind of data of the respective dataset to generate a corresponding output that specifies a respective partial representation of the electric power grid;
• generating, using each of the predictive models that have been identified for the two or more datasets, respective partial representations of the electric power grid that each characterize a different electrical aspect of the electric power grid in the geographical region, wherein the generating comprises:
• generating, using a first predictive model that is configured as a convolutional neural network, predicted locations of the utility poles included in the assets based on processing a first dataset that comprises imagery data of the geographical region, and
• assigning the predicted locations of the utility poles as locations of anchor points for a composite representation of the electric power grid;
• and
• aggregating, in accordance with a set of aggregation rules, the respective partial representations to generate the composite representation of the electric power grid, wherein the composite representation of the electric power grid comprises information about the locations of the anchor points and information about connections between the anchor points, wherein aggregating comprises removing a particular output of one predictive model based on output generated by at least one other predictive model such that the particular output is excluded from the composite representation of the electric power grid.

One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for generating a composite representation of an electric power grid, the operations comprising:

• obtaining two or more datasets comprising data indicating a presence of or operation of assets of the electric power grid in a geographical region that has the electric power grid, each dataset including data of a particular kind, wherein the assets comprise utility poles and connecting lines between the utility poles;
• identifying one or more respective predictive models for each dataset, each predictive model being configured to process the respective kind of data of the respective dataset to generate a corresponding output that specifies a respective partial representation of the electric power grid;
• generating, using each of the predictive models that have been identified for the two or more datasets, respective partial representations of the electric power grid that each characterize a different electrical aspect of the electric power grid in the geographical region, wherein the generating comprises:
• generating, using a first predictive model that is configured as a convolutional neural network, predicted locations of the utility poles included in the assets based on processing a first dataset that comprises imagery data of the geographical region, and
• assigning the predicted locations of the utility poles as locations of anchor points for a composite representation of the electric power grid;
• and
• aggregating, in accordance with a set of aggregation rules, the respective partial representations to generate the composite representation of the electric power grid, wherein the composite representation of the electric power grid comprises information about the locations of the anchor points and information about connections between the anchor points, wherein aggregating comprises removing a particular output of one predictive model based on output generated by at least one other predictive model such that the particular output is excluded from the composite representation of the electric power grid.