Numerical analysis of modeling based on improved Elman neural network

Area of Science:

• Electrical Engineering
• Computational Intelligence
• Nonlinear Circuit Analysis

Background:

• Nonlinear circuits with memory effects pose significant challenges for accurate behavioral modeling.
• Existing models like Volterra-Laguerre (VL), Chebyshev neural network (CNN), and basic Elman neural network (BENN) have limitations in capturing complex dynamics.

Purpose of the Study:

• To propose and evaluate an improved Elman neural network (IENN) model for analyzing nonlinear circuits with memory effects.
• To enhance the accuracy and performance of behavioral modeling for systems like Class-D power amplifiers.

Main Methods:

• A novel IENN model utilizing Chebyshev orthogonal basis functions in the hidden layer instead of traditional sigmoid functions.
• Analysis of error curves (Sum of Squared Error - SSE) to determine optimal hidden layer neuron count and iteration steps.
• Simulation and comparison with VL, CNN, and BENN models using a half-bridge Class-D power amplifier (CDPA) with two-tone and broadband input signals.

Main Results:

• The proposed IENN model accurately reconstructs the behavior of Class-D power amplifiers.
• The model effectively depicts the memory effect inherent in these nonlinear circuits.
• Simulation results indicate superior performance of the IENN model compared to VL, CNN, and BENN models.

Conclusions:

• The IENN model offers a robust and accurate approach for behavioral modeling of nonlinear circuits with memory effects.
• The use of Chebyshev basis functions enhances the model's ability to capture complex nonlinear dynamics.
• This advanced modeling technique provides improved performance for power amplifier analysis.