Three-phase rectification control method based on improved adaptive fuzzy neural network

Abstract

The invention provides a three-phase rectification control method based on an improved adaptive fuzzy neural network, and relates to the technical field of power electronics. According to the invention, the defects of the traditional fuzzy neural network are improved; a mode of combining a type I function and a type II function is adopted; parameters are updated using an elliptical membership function through a gradient descent method; the output of the antecedent parameters with respect to the function is non-linear, and since there is a normalization layer in the structure of the fuzzy neural network, the parameter of the antecedent of each membership function exists at least in the denominator of the output of the normalization layer regardless of whether the parameter participates in the rule. Sliding mode control enables the whole system to be more stable, and can be ensure that the fuzzy neural network converges faster in the online adjustment process; and the PD controller and the improved self-adaptive fuzzy neural network controller are combined to achieve better control over three-phase rectification compared with a traditional PID controller.

Description

technical field [0001] The invention relates to the technical field of power electronics, in particular to an improved self-adaptive fuzzy neural network-based three-phase rectification control method. Background technique [0002] As a part of the power conversion circuit, the rectifier circuit occupies an indispensable position in the circuit conversion. Three-phase rectifiers are widely used in energy storage systems, electric vehicle charging systems, data and communication systems, microgrids and renewable energy systems. . The rectifier circuit needs to meet two conditions: first, to realize the synchronization of the AC side current and the grid voltage to reduce the damage to the grid. Second, it has a controllable DC output voltage. These two goals can be achieved through different control methods, which are mainly divided into voltage-oriented control and power control. Voltage-oriented control is to use a proportional-integral-derivative (PID) controller to cont...

AI Technical Summary

Problems solved by technology

However, the traditional adaptive fuzzy neural network has the following disadvantages: First, fuzzy systems are usually constructed using type-one fuzzy sets. Although they are popular, studies have shown that type-one fuzzy systems may have poor There are difficulties in deterministic influence. This is because the fuzzy function of type-1 fuzzy is clear, and it cannot handle some uncertain external disturbances better. Type-2 fuzzy system can solve this problem better. The powerful performance of type fuzzy systems is achieved at the cost of higher computational requirements

Second, fuzzy function selection, usually the fuzzy neural network chooses the Gaussian function as the fuzzy membership function, but the Gaussian function has some long-tail problems

Therefore, there is no closed form for calculating the partial derivative of the fuzzy neural network output with respect to its predecessor, and it is easy to fall into a local minimum. The biological evolution method can solve the above problems, but there are certain difficulties in the stability analysis

Images