Method and a system for non-linear optimal estimation of dynamic processes in real time

• US 5,933,352 A
• Filed: 11/26/1996
• Issued: 08/03/1999
• Est. Priority Date: 06/08/1994
• Status: Expired due to Fees

• Current Assignee: Gerard Salut
• Original Assignee: Gerard Salut
• Inventors: Salut, Gerard
• Primary Examiner(s): Grant, William
• Assistant Examiner(s): THOMAS, CAROLYN D
• Application Number:
  US08/750,313
• Time in Patent Office: 980 Days
• Field of Search: 364/148, 364/413, 364/724.01, 364/551.02, 364/151, 364/131, 364/474.01, 364/474.11, 364/474.16, 395/13, 395/23, 358/209, 702/185
• US Class Current: 700/174
• CPC Class Codes: G05B 21/02 electric
• CPC Class Codes: G05D 1/0825 using mathematical models
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30

Abstract

A system for non-linear optimal estimation of a state of a dynamic process in real time includes several measurement sensors for delivering sampled data associated with the state of the process at the instant under consideration t, and a computation unit associated with memories and loaded with computation programs suitable for delivering in recurrent manner the estimated components of the state of the process at the instant under consideration t. The computation unit includes N particular processors disposed in parallel and each receiving the measurement data. Each of the N processors includes at least one behavior random generator which, by taking account of random disturbances in said behavior, delivers possible values of the state of the dynamic process at the instant t, together with a specialized operator which uses possible values for the state of the process and the measurement data to compute a scalar magnitude representing the probability that the state is real at instant t of the process.

Claims

1. A method for non-linear optimal estimation of dynamic processes in real time, the method comprising

providing measurement sensors arranged to measure sampled data associated with a state of a dynamic process at an instant under consideration t;

providing a computation unit associated with memories;

said data measured by the sensors is then used in application of a computation program stored in the computation unit to deduce the estimated components of the state of the dynamic process at said instant t, and the operation is repeated recurrently to estimate the state at the instant following t+1 on arrival of new measurements;

using N identical particular processors disposed in parallel in said computation unit and each comprising at least two associated elementary operators, one of which is a behavior random generator which delivers components of a possible state of the dynamic process at the current instant t, and the other of which is a weighting unit from which said components delivered by the random generator are associated with a scalar magnitude referred to as a weight, representing the probability that said components are those of the current state of the dynamic process to be estimated;

producing by said behavior random generator a possible state for the process to be estimated selected from the set constituting state space, and taking into account the components of said possible state computed at instant t-1 and the probability of transition in the state of the dynamic process between instants t-1 and t;

computing by said weighting unit the weight of the possible state on the basis of the value of said weight at the preceding instant t-1, on the basis of the values of the components of said possible state, and on the basis of measurement data picked up by the sensors at the current instant t, with all the above taking into account the probability of noise disturbing said measurements;

initializing each processor by randomly drawing the initial components of said possible state and the associated initial weight by means of an initial state generator which applies its own specific a priori probability relationship representative of knowledge about the initial state of the dynamic process; and

delivering, at each instant t, a probabilistic distribution of the state of the dynamic process conditional on the data measured by the sensors up to the instant t, said distribution having as its support the set of possible states and as its mass point distribution the weights associated with each of said possible states.

2. A method according to claim 1, wherein an optimal estimated state of the dynamic process is determined on the basis of said conditional probabilistic distribution by selecting a particular indicator.

3. A method according to claim 1, wherein the influence of past data measured by the sensors on the weights associated with the components of a possible state is strictly attenuated in time.

4. A method according to claim 1, wherein said behavior random generator is conditioned by the n most recent measurements taken by the sensors.

5. A method according to claim 1, wherein at least a portion of the N possible states depending on said conditional probabilistic distribution of the supports of said possible states delivered by the N particular processors are randomly redistributed by distribution members in uniformly weighted manner.

9. A method according to claim 1, wherein the influence of past data measured by the sensors on the weights associated with the components of a possible state is attenuated asymptotically.

6. A system for non-linear optimal estimation of a state of a dynamic process in real time, the system comprising

measurement sensors constructed to deliver sampled data associated with a state of a dynamic process at an instant under consideration t,

a computation unit associated with memories and loaded with computation programs suitable for delivering in recurrent manner estimated components of the state of the dynamic process at the instant under consideration t, said computation unit comprising N particular processors disposed in parallel and each receiving said measurement data,

each of said N processors including at least one behavior random generator which, by taking account of random disturbances in said behavior, delivers possible values of the state of the dynamic process at the instant t, and each of said N processors also including a weighting unit that uses said possible values for the state of the dynamic process and said measurement data to compute a scalar magnitude representing the probability that said state is real at the instant t of the process.

7. A system according to claim 6, wherein said computation unit includes a particular indicator which receives the set of said possible states weighted by their weights and delivering the estimate of the real state of the process depending on its probable behavior.

8. A system according to claim 6, wherein said computation unit is further arranged to be loaded with different programs.

10. A system for non-linear optimal estimation of a state of a dynamic process in real time, the system comprising

at least one measurement sensor for delivering sampled data associated with a state of a dynamic process at an instant under consideration t,

a computation unit means loaded with computation programs for delivering in recurrent manner estimated components of the state of the dynamic process at the instant under consideration t,

said computation unit means comprising N parallel processor means arranged to receive said measurement data, each said processor means including

behavior random generator means for delivering possible values of the state of the dynamic process at the instant t by taking account of random disturbances in said behavior, and

specialized operator means for computing a scalar magnitude, representing the probability that said state is real at the instant t of the process, by using the possible values for the state of the dynamic process and the measurement data.