Systems and methods of composite load modeling for electric power systems

An electric power load modeling system comprising:

• a load;
• a bus connected to the load;
• a measurement device coupled to the bus for measuring electric state at the bus;
• a processor;
• a database storing a training event record;
• and
• a computer-readable storage medium, comprising:
• software instructions executable on the processor to perform operations, including:
• acquiring state information at the bus from the measurement device;
• acquiring the training event record from the database;
• generating a composite load model by sequentially executing steps of:
• determining a final load component composition for a predetermined composite load model structure referencing to both the state information and the training event record;
• and
• determining a plurality of final load parameters corresponding to the final load component composition and the predetermined load model structure also referencing to both the state information and the training event record,
• wherein the determining the plurality of final load parameters by the software instructions executable on the processor comprises:
• fixing the final load component composition;
• generating a plurality of sets of parameter values from predetermined parameter ranges;
• generating a dynamic response curve for each of the plurality of sets of parameter values;
• evaluating the generated dynamic response curves by calculating root mean squared error (RMSE) of the dynamic response curves in reference to the training event record;
• and
• selecting one of the plurality of sets of parameter values with a dynamic response curve having the lowest RMSE for the plurality of final load parameters.

A method for generating a load model having composite load components for an electric power system, the method comprising:

• acquiring state information at a bus of the electric power system;
• acquiring a training event record;
• and
• generating the load model by sequentially executing steps of:
• determining a final load component composition for a predetermined composite load model structure referencing to the state information and the training event record;
• and
• determining a plurality of final load parameters corresponding to the final load component composition and the predetermined load model structure using a curve fitting process approximating a modeled curve based on the state information to a reference curve generated by the training event record,
• wherein the determining the final load component composition comprises:
• initializing a plurality of load component compositions;
• optimizing the plurality of the initial load component compositions with a deep reinforcement learning (DRL) agent to generate a predetermined number of top load component compositions;
• calculating quantile loss for each of the top load component compositions with a predefined quantile level;
• and
• selecting one of the top load component compositions that has the lowest quantile loss as the final load component composition.