Risk-sensitive reinforcement learning algorithms with generalized average criterion

A new algorithm is proposed, which immolates the optimality of control policies potentially to obtain the robnsticity of solutions. The robnsticity of solutions maybe becomes a very important property for a learning system when there exists non-matching between theory models and practical physical system, or the practical system is not static, or the availability of a control action changes along with the variety of time. The main contribution is that a set of approximation algorithms and their convergence results are given. A generalized average operator instead of the general optimal operator max （or rain） is applied to study a class of important learning algorithms, dynamic prOgramming algorithms, and discuss their convergences from theoretic point of view. The purpose for this research is to improve the robnsticity of reinforcement learning algorithms theoretically.

A Generalized State Space Average Model for Parallel DC-to-DC Converters

The high potentiality of integrating renewable energies,such as photovoltaic,into a modern electrical microgrid system,using DC-to-DC converters,raises some issues associated with controller loop design and system stability.The generalized state space average model(GSSAM)concept was consequently introduced to design a DC-to-DC converter controller in order to evaluate DC-to-DC converter performance and to conduct stability studies.This paper presents a GSSAM for parallel DC-to-DC converters,namely:buck,boost,and buck-boost converters.The rationale of this study is that modern electrical systems,such as DC networks,hybrid microgrids,and electric ships,are formed by parallel DC-to-DC converters with separate DC input sources.Therefore,this paper proposes a GSSAM for any number of parallel DC-to-DC converters.The proposed GSSAM is validated and investigated in a time-domain simulation environment,namely a MATLAB/SIMULINK.The study compares the steady-state,transient,and oscillatory performance of the state-space average model with a fully detailed switching model.

Impedance Modeling and Stability Analysis of Dual Active Bridge Converter with LC Input Filter

The dual active bridge(DAB)converter is gaining more and more attention in various applications such as energy storage systems,electric vehicles and smart grids.To improve the quality of the input current,a LC filter is often cascaded at the input side of the DAB converter.However,there are instable problems of this cascaded system due to the impedance interactions of the DAB converter and the LC filter,although the DAB converter is stable at the individual operation mode.To assess the stability of the cascaded system of the DAB converter and the LC filter,the impedance model of the DAB converter is firstly developed based on generalized state-space averaging method.The developed impedance model can be used to accurately predict the stability of the DAB converter with its LC input filter.Based on the stability analysis,the optimum filter parameter design guideline is determined.The impedance model and stability analysis are validated by the simulation and experimental results.

EFFECTS OF MICROSTRUCTURE ON THE MACROSCOPIC ELASTO-PLASTIC PROPERTIES OF METAL MATRIX COMPOSITES

The generalized self-consistent finite-element iterative averaging method was adopted to analyze the elasto-plastic tensile properties of SiC whiskers reinforced aluminum matrix composites. The effects of varying fiber's aspect ratio and volume fraction on the macroscopic elasto-plastic deformation of the composites were studied. By the analysis of microscopic stress fields, the relation between the propagation of the elasto-plastic region in the matrix and the macroscopic elasto-plastic deformation of composites was discussed. It was found that the propagation of the plastic region in the matrix between the fiber's ends would affect prominently the elasto-plastic tensile behaviour of the composites. It was shown that the characterization of the stress-strain response in terms of the 0.2% offset yield strength is incomplete.

A System Identification-based Modeling Framework of Bidirectional DC-DC Converters for Power Grids

This paper proposes a system identification framework based on eigensystem realization to accurately model power electronic converters.The proposed framework affords an energy-based optimal reduction method to precisely identify the dynamics of power electronic converters from simulated or actual raw data measured at the converter’s ports.This method does not require any prior knowledge of the topology or internal parameters of the converter to derive the system modal information.The accuracy and feasibility of the proposed method are exhaustively evaluated via simulations and practical tests on a software-simulated and hardware-implemented dual active bridge(DAB)converter under steady-state and transient conditions.After various comparisons with the Fourier series-based generalized average model,switching model,and experimental measurements,the proposed method attains a root mean square error(RMSE)of less than 1%with respect to the actual raw data.Moreover,the computational effort is reduced to 1/8.6 of the Fourier series-based model.

Review of General Modeling Approaches of Power Converters

The modeling approaches of power converters occupy an important position in power electronic systems and have made considerable progress over the past years.Continuous modeling approaches and linearization techniques are reviewed,including the state-space average model,generalized average model,averaged small-signal model,and describing function method.A Buck converter with PWM modulation and voltage-mode control is taken as an example to compare the advantages and disadvantages of different methods through simulation analysis.Moreover,the corresponding equivalent circuit with an intuitive physical meaning of state-space average model,generalized average model,and averaged small-signal model is given.The results point out that the generalized average model can improve the modeling accuracy based on the state-space average model.In the linearization techniques,the averaged small-signal model reflects accuracy at low frequencies,but introduces phase lag in the high-frequency region.The describing function method is derived from harmonic linearization,which takes into account the sideband effect and improves the modeling accuracy at high frequencies.