Multiple-agent hybrid control architecture for intelligent real-time control of distributed nonlinear processes

An agent for controlling a local process of a distributed hybrid system comprising:

• a knowledge base for storing requirements of a local process;
• a knowledge base builder within the agent and connected to the knowledge base for acquiring system constraint information, the knowledge base builder having an interface to a user for receiving system requirement information;
• a knowledge decoder for inserting information from other agents into the knowledge base;
• a planner for generating a statement representing a desired behavior of the system as an existentially quantified logic expression;
• an inferencer for determining whether the statement is currently active in the knowledge base;
• and
• an adapter connected to the inferencer for modifying the statement if the statement is not currently active in the knowledge base.

An architecture for implementation of control of a distributed, real-time computing system comprising:

• a generic software agent process for controlling a physical or logical device;
• a subprocess of the agent for extracting the agent's current optimal control by computing an affine connection for a manifold associated with the system;
• a feedback loop in the agent for using violations of conservation laws to compute corrections to a Lagrangian used by the agent;
• and
• a memory map design for enabling the affine connection to be implemented for real time processes by arranging adjacent charts of an atlas of the manifold, through which a geodesic carrying the connection passes, to be close in memory access.

A method for controlling a distributed real-time computing system comprising the steps of:

• creating a description of the distributed, real-time computing system;
• proving that a solution to the system exists;
• enabling an affine connection to be implemented for real time processes by using a memory map process to arrange adjacent charts of an atlas of a manifold to be close in memory access;
• extracting a near optimal path to meet performance criteria for the system by calculating the affine connection;
• and
• generating automata that take actions according to a current system state to cause a compensated system state to follow a near-optimal path to meet performance criteria for the system.

A method for transforming a software procedure comprising the steps of:

• creating a discrete software procedure representing physical or functional conditions;
• collecting data from at least one sensor;
• transforming the discrete software procedure into a differential equation whose solution contains values computed by the software procedure using data from the at least one sensor;
• converting the differential equation into a Hamilton-Jacobi-Bellman equation;
• coding recursion into an evolution of the Hamilton-Jacobi-Bellman equation;
• solving on-line the Hamilton-Jacobi-Bellman equation;
• and
• producing signals for controlling at least one action desired to be taken as a result of solving the Hamilton-Jacobi-Bellman equation.