Hierarchical combination of artificial intelligence and optimization for the operation of power systems

A computer-implemented method, executed on one or more processors, for real-time coordinated operation of power distribution systems, comprising:

• identifying a set of integrated hybrid resources (IHRs), wherein each IHR within the set of IHRs comprises one or more of:
• energy storage (ES) systems, solar generating units, electric vehicles (EVs), and/or inflexible loads, and is electrically coupled to one or more of the power distribution systems via one or more distribution buses;
• executing, at an IHR selected from the set of IHRs, a deep deterministic policy gradient (DDPG) algorithm, the DDPG algorithm utilizing a critic deep neural network and an actor deep neural network, wherein:
• the critic deep neural network estimates a Q-value of an action for a given state, and
• the actor deep neural network estimates a best action for the given state,
• wherein the given state is based on a state-of-charge of the ES systems, an inflexible load vector, a solar generation power vector, departure and remaining requested energy of the EVs, and a real-time electricity price;
• based upon an output of the DDPG algorithm, generating a charging schedule for the ES systems and the EVs within the IHR;
• and
• performing a power flow analysis to determine adjusted active power and reactive power setpoints while ensuring deliverability in real-time operation,
• wherein the charging schedule specifies active and reactive power dispatch values that control charging and discharging of the ES systems and charging of the EVs in response to the real-time electricity price, and
• wherein the adjusted active power and reactive power setpoints are distributed between distributed energy resources in each IHR.

A computer system for real-time coordinated operation of power distribution systems, comprising:

• a central controller configured to ensure a feasibility and deliverability of dispatched energy in a set of integrated hybrid resources (IHRs);
• the set of IHRs;
• one or more processors;
• and
• one or more computer-readable media having stored thereon executable instructions that when executed by the one or more processors configure the computer system to:
• identify the set of IHRs, wherein each IHR within the set of IHRs comprises one or more of:
• energy storage (ES) systems, solar generating units, electric vehicles (EVs), and/or inflexible loads, and is electrically coupled to one or more of the power distribution systems via one or more distribution buses;
• execute, at an IHR selected from the set of IHRs, a deep deterministic policy gradient (DDPG) algorithm, the DDPG algorithm utilizing a critic deep neural network and an actor deep neural network, wherein:
• the critic deep neural network estimates a Q-value of an action for a given state, and
• the actor deep neural network estimates a best action for the given state,
• wherein the given state is based on a state-of-charge of the ES systems, an inflexible load vector, a solar generation power vector, departure and remaining requested energy of the EVs, and a real-time electricity price;
• based upon an output of the DDPG algorithm, generate a charging schedule for the ES systems and the EVs within the IHR;
• and
• perform a power flow analysis to determine adjusted active power and reactive power setpoints while ensuring deliverability in real-time operation,
• wherein the charging schedule specifies active and reactive power dispatch values that control charging and discharging of the ES systems and charging of the EVs in response to the real-time electricity price, and
• wherein the adjusted active power and reactive power setpoints are distributed between distributed energy resources in each IHR.