System and Method for Controlling System

Patent metadata

Field	Value
Publication number	US-2016246266-A1
Application number	US-201514631159-A
Country	US
Kind code	A1
Filing date	Feb 25, 2015
Priority date	Feb 25, 2015
Publication date	Aug 25, 2016
Grant date	—

Abstract

A controller for controlling a system includes a non-transitory computer-readable memory storing data for an operation and a control of the system and at least one processor operatively connected to the memory for determining a control signal transitioning a state of the system from a current state to a next state. At least two instances of the data are stored in the memory with different precisions defined by numbers of bits storing the instance in the memory. The processor determines the control signal using the instances of the data with the different precisions.

First claim

We claim:

1 . A controller for controlling a system, comprising:
a non-transitory computer-readable memory storing data for an operation and a control of the system, wherein at least two instances of the data are stored in the memory with different precisions, wherein the precision of the instance is defined by a number of bits storing the instance in the memory; and at least one processor operatively connected to the memory for determining, in response to receiving a current state of the system, a control signal transitioning a state of the system from the current state to a next state using the instances of the data with the different precisions.

2 . The controller of claim 1 , wherein the control of the system is a model predictive control (MPC), such that the controller determines the next state of the system using a model of the system, wherein the data includes the current state of the system, parameters of the model of the system, and parameters of the control, wherein the different precisions are determined such that an accuracy of the controller is greater than a target accuracy of the controller, wherein the accuracy of the controller is an inverse of one or a combination of an absolute and relative deviation of a true next state of the system determined with a true control using true data having a true precision from the next state of the system determined with the control using the data including the instances with the different precisions lower than the true precision.

3 . The controller of claim 2 , further comprising of one or a combination of:
a sensor for sensing the current state of the system, wherein an accuracy of the sensor defines the target accuracy of the controller; an estimator for estimating the current state of the system, wherein an accuracy of the estimator defines the target accuracy of the controller.

4 . The controller of claim 1 , wherein the control of the system is an explicit model predictive control (EMPC), wherein the data includes parameters of facets partitioning a state space of the system into a set of convex state regions and parameters of the control for each state region, wherein the processor determines the control signal using the parameters of the control for a state region including the current state of the system, and wherein the instances of the data with the different precisions include at least one of the parameters of the facets and the parameters of the control for various state regions.

5 . The controller of claim 4 , wherein the parameters of the control include a gain and an offset of the control for each state region, wherein the data includes a first gain and a first offset of the control for a first state region quantized with a first precision, and includes a second gain and a second offset of the control for a second state region quantized with a second precision, and wherein the first and the second precisions are such that a first accuracy of the controller with the current state of the system in the first region substantially equals a second accuracy of the controller with the current state of the system in the second region.

6 . The controller of claim 4 , wherein the parameters of the facets include a parameter for a first facet quantized with a first precision and a parameter for a second facet quantized with a second precision, wherein the first and the second precisions are lower than a true precision, and wherein the first precision is different than the second precision.

7 . The controller of claim 6 , wherein the first facet boarders at least a first state region and the second facet boarders at least a second region, and wherein the first precision and the second precision are such that a first accuracy of the controller with the state of the system in the first region substantially equals a second accuracy of the controller with the state of the system in the second region.

8 . The controller of claim 7 , wherein the first accuracy of the controller is an inverse of deviations of values of a first next state of the system determined by the controller from a true first next state at a first point of a maximal deviation of a first current state of the system from the first facet such that a quantized current state remains in the quantized first region bordered by the first facet quantized with the first precision, and wherein the second accuracy of the controller is an inverse of the deviations of values of a second next state of the system determined by the controller from a second true next state at a second point of a maximal deviation of the second current state of the system from the second facet such that the quantized current state remains in the quantized second region bordered by the second facet quantized with the second precision.

9 . The controller of claim 1 , wherein the different precisions of the instances of the data during the operation of the system are determined using one or a combination of a known probability distribution of the current state in the state space and subset of the state space that is more likely to include the current state of the system than the rest of the state space, and wherein the precision for the instances of the data for subset of the state space is greater than the precision for the instances of the data for the rest of the state space.

10 . The controller of claim 6 , wherein the first precision and the second precision are such that a first accuracy of the controller with the state of the system in a first vicinity of the first facet substantially equals a second accuracy of the controller with the state of the system in a second vicinity of the second facet, wherein the vicinity of the facet is a largest set of the state space such that the current state of the system in the set is on the opposite side from the facet compared to the quantized current state from the quantized facet.

11 . The controller of claim 6 , wherein the first precision is greater than the second precision and the data are ordered in the memory such that the second facet is evaluated with respect to the current state before the first facet.

12 . The controller of claim 6 , wherein the facets are clustered in the memory according to values of the different precisions into a plurality of clusters, further comprising:
a plurality of processors connected to corresponding clusters for evaluating concurrently facets from different clusters.

13 . The controller of claim 1 , wherein the processor determines and updates the different precisions of the instances of the data during an operation of the system.

14 . A method for controlling a system, comprising:
determining, in response to receiving a current state of the system, a control signal transitioning a state of the system from the current state to a next state using the data representing an operation and a control of the system, wherein at least two instances of the data are quantized with different precisions, wherein the precision of the instance is defined by a number of bits required to store the instance in a memory; and controlling the system using the control signal, wherein steps of the method are performed by at least one processor of a controller.

15 . The method of claim 14 , wherein the different precisions are determined according to a function bounding an accuracy of the controller for the different precisions of the instances of the data, wherein the accuracy of the controller is an inverse of a deviation of a true control determined with true data having a true precision from a control determined with the data having a precision lower than t

Assignees

• Mitsubishi Electric Res Laboratories Inc

Inventors

• Kniazev Andrei
• Zhu Peizhen
• Di Cairano Stefano

Classifications

• H04N19/124

  Quantisation · CPC title

• H04N19/132

  Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking · CPC title

• G06F9/30025

  Format conversion instructions, e.g. Floating-Point to Integer, decimal conversion · CPC title

• G06F9/30014

  with variable precision · CPC title

• G05B13/048 Primary

  using a predictor · CPC title