Robust state estimation using desensitized Divided Difference Filter

Area of Science:

• Control Systems Engineering
• Signal Processing
• Electrical Engineering

Background:

• Kalman filtering is widely used for state estimation but sensitive to model parameter errors.
• Parameter uncertainties in dynamic systems can degrade filter performance and lead to inaccurate state estimates.
• Developing robust filtering techniques is crucial for reliable system operation.

Purpose of the Study:

• To develop a robust divided difference filtering approach.
• To enhance Desensitized Kalman Filtering by incorporating state sensitivity penalties.
• To provide solutions for first and second-order Divided Difference Filters.

Main Methods:

• Formulating filters using a minimum variance cost function.
• Augmenting the cost function with a penalty for state sensitivities.
• Developing first and second-order Divided Difference Filters.
• Utilizing Monte Carlo simulations for algorithm validation.

Main Results:

• The developed filters are non-minimum variance.
• Filters demonstrate reduced sensitivity to deviations in plant model parameters.
• Successful application to induction motor state estimation with parameter uncertainties.

Conclusions:

• The proposed divided difference filtering approach effectively reduces sensitivity to parameter uncertainties.
• The method offers improved robustness for state estimation in systems with model inaccuracies.
• The technique is validated for practical applications like induction motor analysis.