Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
2002-02-13
2004-12-07
Barlow, John (Department: 2863)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C320S132000
Reexamination Certificate
active
06829562
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a method for state sensing of a technical system, for example, an energy store, in which performance quantities are measured and supplied to a state estimation routine, which determines state variables characterizing the current system state using a model based on system-dependent model parameters and the measured performance quantities, the measured performance quantities and possibly the determined state variables additionally being supplied to a parameter estimation routine, which in turn, determines the model parameters depending on a use by estimation, to improve the state estimation. The present invention further relates to a corresponding device and a computer program for performing the method and a computer program.
BACKGROUND INFORMATION
A method for battery state identification is referred to in German Published Patent Application No. 199 59 019. Proceeding from measurable performance quantities, such as current, voltage, and temperature, state variables are determined by estimation using a model, which is implemented as a (Kalman) filter. Since the parameters of the model may change due to aging of the battery and by suddenly occurring defects, a parameter estimation routine tracks the parameter changes online and adjusts the parameters appropriately. The current parameters are then supplied to the state estimation routine, i.e., the filter. In this way, the model is continuously adapted to the actual state of the battery, and the filter does not estimate incorrect values for the state variables. The separation of the estimation of state variables and parameters by the filter and the parameter estimator, respectively, may cause biased estimations to be avoided and/or to become improbable.
This method for state identification, which uses estimation of both the state variables and the model parameters they are based on, may not be sufficient to guarantee a required accuracy of the estimated values and to avoid divergences in covariance matrices, which are used for estimation.
Therefore, it is believed that the state and parameter estimation should be more stable, with computing and memory needs being reduced as much as possible, allowing such an estimation for all conceivable system states.
SUMMARY OF THE INVENTION
In an exemplary method according to the present invention, only specific state variables and/or parameters are used at a time for estimation, the selection being performed on the basis of the dynamic response of the measured performance quantities. An exemplary device according to the present invention includes an arrangement to determine the dynamic response of the measured performance quantities, for example, an arrangement to produce temporal gradients of the respective performance quantities, and a selection arrangement, which determines specific state variables and/or parameters from the corresponding estimation routine. Such a selection arrangement may, for example, be implemented in tables, as stepped functions, or as threshold value functions, through which specific parameters and/or state variables may be assigned to specific dynamic ranges of the performance quantities.
Kalman filters, which operate with covariance matrices of the estimated quantities, may be used for state estimation. Covariance matrices represent a root-mean-square deviation of the estimated value from the measured value on their diagonals, and the remaining matrix elements represent the correlations between the individual state variables. Through an exemplary method according to the present invention, the order of the matrices is reduced, and thus the numerical outlay and the necessary storage requirements may be diminished. Those parameters that change in different time ranges and in the event of different excitations, i.e., in the event of the performance quantities present, may also be determined better.
It is believed to be advantageous to estimate the state variables and/or parameters that have small time constants at a high dynamic response of the measured performance quantities, and to estimate the state variables having large time constants at a low dynamic response of the measured performance quantities. At the same time, the respective other state variables and/or parameters are maintained or tracked using a predetermined model.
In an example application of battery state identification, a battery model uses various resistance and voltage quantities having different time constants. Ohmic values and charge-transfer overvoltage have small time constants and may be estimated when the measured performance quantities have a large dynamic response. In contrast, the concentration overvoltage, for example, may have a large time constant, so that it may be estimated at a low dynamic response. The respective other quantities are maintained during the estimation or changed according to a predetermined pattern.
It is believed to be advantageous to determine, before the estimation determination, whether the system is in a limit state and if the state variables and/or parameters are only estimated if the system is not in a limit state. These types of limit states may exist, for example, at the beginning and the end of the service life of a technical system. In the exemplary application of battery state identification, more accurate estimated values may be dispensed with, if the battery is almost fully charged, since the limit state is a desirable and non-critical area. Another case exists if the battery is in a very poor (charge) state. Since a poor charge state is normally recognized sufficiently early, therefore, avoiding complete failure of the battery, this limit state may not be relevant.
Through the masking out of the estimation in the boundary areas of model accuracy, divergences of the filter/estimator and poor qualities of the quantities determined may be avoided. If, for example, the charge state of a fully charged battery worsens again, the battery automatically enters an operating point, in which the model used is valid, and a Kalman filter may provide estimated values of greater quality. It is believed that advantages may result in regard to necessary hardware. Lesser numerical complexity may thereby be achieved, and thus a lesser utilization of the processor and lesser demands for the storage requirements in the RAM.
In an exemplary method according to the present invention, the quality of the estimation determined on the basis of a covariance matrix described above may be checked. Specifically, the smaller the value in the covariance matrix for the respective state variable, the more probable or more accurate the estimated value of the quantity. The same applies for parameter estimation, in which there are covariance matrices in the typical estimation theories (e.g., Bayes, maximum likelihood methods), which make an assertion about the quality of the parameter estimation and/or the accuracy of the estimated parameter. The approach used is approximately the same as for the state estimator. The convergence (to values near zero) of the matrix quantities assigned to the estimated quantities may be used to rate the quality of the estimation. In addition, with appropriate weighting of the results determined, the overall assertion, in regard to the state variables, such as charge and aging state of the battery, may be enhanced.
Fixing a threshold value for the matrix value assigned to the respective estimated quantity may permit the quality of the estimation to be determined. These threshold values may be determined by experimental values and may be close to zero.
If the estimated values only have low quality, other “backup” methods may also be included in the evaluation, and may be more strongly weighted. Using a “backup” method, the respective quantities may be maintained or adjusted according to simple models that do not cause divergences. Alternatively, certain parameters may not, momentarily, be accepted from the state estimation routine or certain states may not, momentarily, initiate the parameter estimation rou
Barlow John
Kenyon & Kenyon
Le Toan M.
Robert & Bosch GmbH
LandOfFree
Method and device for state sensing of technical systems... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and device for state sensing of technical systems..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and device for state sensing of technical systems... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3308876