CN-121215099-B - Knowledge-driven neural network-based intelligent driving material hysteresis modeling method

CN121215099BCN 121215099 BCN121215099 BCN 121215099BCN-121215099-B

Abstract

The invention discloses an intelligent driving material hysteresis modeling method based on a knowledge driving type neural network, which relates to the technical field of intelligent manufacturing and driving control and comprises the steps of constructing a fractional-order backhaul-GRUNN hybrid model of a fusion material physical mechanism, inputting multi-scale physical characteristics of an intelligent driving material into the fractional-order backhaul-GRUNN hybrid model, carrying out model training on the fractional-order backhaul-GRUNN hybrid model to obtain a trained fractional-order backhaul-GRUNN hybrid model, and taking the trained fractional-order backhaul-GRUNN hybrid model as a hysteresis model of the intelligent driving material for describing hysteresis characteristics of the intelligent driving material under different driving signals. The method solves the problems of high modeling complexity, strong dependence on parameters, lower model precision and poor interpretability of the existing method.

Inventors

WANG GENG
HUANG LINHAI
WANG JIANXIN
WANG YING
LIAO XUAN
YU JIAXIN

Assignees

西南科技大学

Dates

Publication Date: 20260508
Application Date: 20250925

Claims (5)

1. The intelligent driving material hysteresis modeling method based on the knowledge driving type neural network is characterized by comprising the following steps of: S1, constructing a fractional order backhaul-GRUNN mixed model of a fusion material physical mechanism; the fractional order backlight-GRUNN hybrid model is expressed as: Wherein, the As an output variable of the fractional order backhaul neural network submodel, In order to be able to take time, As the sign of the fractional calculus operator, For the order of the number of steps, As an output variable of the fractional order backlight-GRUNN hybrid model, In order for the excitation signal to be present, For the derivative of the excitation signal, 、、、、、 Is a corresponding equation parameter; discretizing according to the definition of the backward difference to obtain: Wherein, the And Respectively the first And The output variable of the time fractional order backhaul neural network submodel, In order to provide a sampling interval, Is the first The time-of-day excitation signal, Is the first The output variable of the moment fractional order backlight-GRUNN hybrid model, Is the first The time-of-day excitation signal, Is the first Derivative of the time-of-day excitation signal; S2, inputting the multi-scale physical characteristics of the intelligent driving material into the fractional-order backhaul-GRUNN hybrid model, and performing model training on the multi-scale physical characteristics to obtain a trained fractional-order backhaul-GRUNN hybrid model; And S3, taking the trained fractional-order backlight-GRUNN mixed model as a hysteresis model of the intelligent driving material, and describing hysteresis characteristics of the intelligent driving material under different driving signals.
2. The knowledge-driven neural network-based intelligent driving material hysteresis modeling method according to claim 1, wherein the fractional-order backhaul-GRUNN hybrid model in S1 is formed by interconnecting a fractional-order backhaul neural network sub-model and a GRUNN neural network sub-model; the fractional order backhaul neural network sub-model comprises 9 input layers, 20 hidden layers and 1 output layer; the input layer comprises the excitation signal of the real system Derivative of excitation signal And 7 input sequences of constant 1 , Wherein, when the signal is excited When connected with the absolute value activation function layer, the connection weight is set to be a constant value of 1, and when an excitation signal is generated When connected with the addition layer, the connection weights are respectively set to be 1, 0.8, 0.08, 0.032 and 0.0176; Input sequence The output of the connected neurons is 7 parameters of a hysteresis model, and the input weights of the neurons are respectively The corresponding custom activation function is: Wherein, the Corresponding equation parameters are respectively 、、、、、、 , For the corresponding neuron(s), For the weights corresponding to the respective neurons, For the corresponding weight sequence number, , An input for the neuron; the expression of the absolute value activation function layer is as follows: Wherein, the The function is activated for the absolute value, An operation for absolute value determination of the input signal of the neuron; each hidden layer contains a neuron; the output layer is used for generating an output signal of the network.
3. The knowledge-driven neural network-based intelligent driving material hysteresis modeling method according to claim 2, wherein the GRUNN neural network sub-model uses an output including a fractional-order backhaul neural network sub-model as an extended input of an GRUNN neural network sub-model input sequence, and an excitation signal And the derivative of the excitation signal Acting together on GRUNN neural network submodels; the calculation process of the GRUNN neural network submodel is as follows: Wherein, the To reset the door The output of the moment of time, To update the door The output of the moment of time, For the sigmoid activation function, In order to reset the weight matrix of the gate, In order to update the weight matrix of the gate, Is that The output of the time of day circulation unit, Is that The input of the model at time GRUNN, In order to reset the bias vector of the gate, To update the bias vector of the gate; Wherein, the Is that The output of the time of day circulation unit, As a candidate for the current hidden state, For the hyperbolic tangent activation function, Is a matrix of weights for the candidate hidden states, Is the bias vector of the candidate hidden state, Representing the element-by-element product.
4. The knowledge-driven neural network-based intelligent driving material hysteresis modeling method according to claim 2, wherein the LM algorithm is adopted to train a fractional order backhaul neural network sub-model.
5. An intelligent driven material lag modeling system based on a knowledge driven neural network, comprising a processor and a memory, the memory storing a computer program which when executed by the processor performs the steps of the intelligent driven material lag modeling method based on a knowledge driven neural network of any of claims 1-4.

Description

Knowledge-driven neural network-based intelligent driving material hysteresis modeling method Technical Field The invention relates to the technical field of intelligent manufacturing and driving control, in particular to an intelligent driving material hysteresis modeling method based on a knowledge driving type neural network. Background The piezoelectric material is an intelligent driving material which generates mechanical strain under the action of an electric field and is applied to a sensor and a driver, and the other intelligent driving material, namely a dielectric elastomer (DIELECTRIC ELASTOMER), is used as a novel soft intelligent material and is used for driving a dielectric film under the action of electric polarization based on Maxwell stress effect induced by the electric field so as to compress the thickness and expand the plane of the dielectric film, thereby realizing large deformation output, showing excellent driving strain, high energy density and excellent energy efficiency performance and providing a brand new path for breaking through the kinematic and dynamic bottlenecks of the traditional electromechanical system. The intelligent driving materials can realize efficient, accurate and flexible control and operation, and have unique physical characteristics and engineering application in the technical field of intelligent manufacturing and driving control. But their response mechanisms to external stimuli (e.g., electric fields, magnetic fields, temperature, or stress) are complex and tend to exhibit hysteresis and non-linear characteristics. For example, the mechanical strain or charge output of the piezoelectric material under the action of an electric field is influenced by historical excitation, and the deformation of the dielectric elastomer under the action of the electric field also shows hysteresis and nonlinear characteristics. Common mathematical modeling describing the hysteresis nonlinear characteristics of intelligent materials can be classified into differential equation models, integral operator models, artificial intelligent models and the like. The differential equation model and the integral operator model belong to a physical knowledge model, have good physical and knowledge interpretability, are suitable for describing dynamic evolution and a physical mechanism of a system, but have high modeling complexity and strong dependence on parameters, so that the built model has low precision. The artificial intelligent model has strong nonlinear modeling capability and adaptability, can process big data and complex systems, but has the problem of poor interpretability, and has high requirements on data quantity and computing resources. In particular, when the data amount is small, the model performance is poor. Disclosure of Invention Aiming at the defects in the prior art, the intelligent driving material hysteresis modeling method based on the knowledge driving type neural network provided by the invention solves the problems of high modeling complexity, strong dependence on parameters, lower model precision and poor interpretability of the existing method. In order to achieve the aim of the invention, the technical scheme adopted by the invention is that the intelligent driving material hysteresis modeling method based on the knowledge driving type neural network comprises the following steps: S1, constructing a fractional order backhaul-GRUNN mixed model of a fusion material physical mechanism; S2, inputting the multi-scale physical characteristics of the intelligent driving material into the fractional-order backhaul-GRUNN hybrid model, and performing model training on the multi-scale physical characteristics to obtain a trained fractional-order backhaul-GRUNN hybrid model; And S3, taking the trained fractional-order backlight-GRUNN mixed model as a hysteresis model of the intelligent driving material, and describing hysteresis characteristics of the intelligent driving material under different driving signals. Further, the fractional-order backhaul-GRUNN hybrid model in the S1 is formed by interconnecting a fractional-order backhaul neural network sub-model and a GRUNN neural network sub-model; the fractional order backhaul neural network sub-model comprises 9 input layers, 20 hidden layers and 1 output layer; the input layer comprises the excitation signal of the real system Derivative of excitation signalAnd 7 input sequences of constant 1,Wherein, when the signal is excitedWhen connected with the absolute value activation function layer, the connection weight is set to be a constant value of 1, and when an excitation signal is generatedWhen connected with the addition layer, the connection weights are respectively set to be 1, 0.8, 0.08, 0.032 and 0.0176; Input sequence The output of the connected neurons is 7 parameters of a hysteresis model, and the input weights of the neurons are respectivelyThe corresponding custom activation function is: Wherein, the Corr