CN-121999937-A - High-entropy alloy phase prediction method based on codon quantum neural network

Abstract

The invention discloses a high-entropy alloy phase prediction method based on a codon quantum neural network, which relates to the technical field of intersection of material science and quantum computation, and comprises the steps of obtaining high-entropy alloy data and preprocessing to generate a feature vector containing thermodynamic, geometric and electronic characteristics; the method comprises the steps of mapping classical features into quantum codons through quantum coding, realizing feature dimension amplification and nonlinear associated information injection, constructing an optimized quantum neural network containing an attention mechanism and a Dropout layer, carrying out classified prediction on high-entropy alloy phase after parameter training and super-parameter optimization, and carrying out experimental verification through an electron microscope. According to the invention, a quantum-classical mixed architecture is adopted, the characteristic distinction degree is effectively improved by quantum codons, the generalization capability of an optimized quantum neural network enhancement model is improved by 8% -12% compared with a traditional method, and a high-efficiency and reliable technical scheme is provided for high-entropy alloy phase prediction.

Inventors

• CHEN XIZHANG
• Jia Yanghao

Assignees

• 温州大学

Dates

Publication Date

20260508

Application Date

20260128

Claims (7)

1. 1. A high-entropy alloy phase prediction method based on a codon quantum neural network is characterized by comprising the following steps: Step 1, data acquisition and preprocessing, namely collecting components, a preparation process and phase data of a high-entropy alloy, constructing a 7-class mutual exclusion phase tag system comprising solid solution, intermetallic compounds and amorphous state, extracting 16 core features, performing skewness/kurtosis diagnosis, logarithmic compression, interval scaling and L 2 normalization, and removing erroneous judgment samples through multimode consistency voting; Step 2, quantum codon generation, namely embedding the preprocessed classical feature vector into a quantum circuit through amplitude coding or rotary interleaving coding, generating a quantum codon vector with the dimension of 2 n by using the superposition state of n quantum bits, and forming a composite input feature of classical feature and quantum codon after scaling with Min-Max through threshold filtering; Constructing a quantum neural network comprising a parameterized rotation-entanglement module, a quantum attention mechanism and a Dropout layer, wherein the rotation-entanglement module adopts a dynamic interweaving revolving door and linear/annular CNOT entanglement structure, and the attention mechanism strengthens key characteristic influence by trainable weights; Performing model training and optimization, namely performing Bayesian super-parameter search by adopting Optuna frames, optimizing the number of quantum bits, the number of circuit layers, the entanglement pattern and the learning rate, taking multiple kinds of range loss functions as targets, training the model by using an Adam optimizer, and introducing early stop system to prevent over fitting; And 5, phase prediction and verification, namely inputting the composite characteristics of the high-entropy alloy to be predicted into a trained model, outputting a phase classification result, and verifying the prediction accuracy through a Scanning Electron Microscope (SEM), a Transmission Electron Microscope (TEM) and an Electron Back Scattering Diffraction (EBSD) experiment.
2. 2. The method of claim 1, wherein the 16 core features in step 1 comprise thermodynamic, geometric and electronic characteristic parameters such as mixed entropy Δsmix, mixed enthalpy Δ Hmix, average valence electron concentration VEC, atomic radius difference δr, average electronegativity χ a ᵥ 9 , and the like, and there is no strong correlation between the features (pearson correlation coefficient absolute value < 0.8).
3. 3. The method of claim 1, wherein the quantum codon generation in step 2 is specifically implemented by converting classical eigenvectors into quantum states by controlled rotation gate (Ry/Rz) combined with CNOT entanglement gate using 4-8 qubits And the vector is decoded into a classical vector through Pauli-Z measurement or full probability measurement, and the dimension is amplified from 16 dimensions to 16-256 dimensions.
4. 4. The method of claim 1, wherein the parameterized spin-entanglement module of the quantum neural network in step 3 comprises 2-4 hidden layers, each layer comprises a cyclic rotation structure of 3 basic spin gates (R x , rgamma, rz), and quantum state association enhancement is achieved by dynamic CNOT entanglement between layers.
5. 5. The method of claim 1, wherein the super-parameter optimization range in step 4 is 4-8 qubits, 2-4 circuit layers, 0.001-0.01 learning rate, 0.1-0.3 Dropout probability, and a classification interval parameter delta=0.15-0.8.
6. 6. The method of claim 1, wherein the phase classification in step 5 comprises 9 types of single phase (FCC, BCC, IM, AM), two phase (FCC+IM, BCC+IM, FCC+BCC) and multiple phase (FCC+BCC+IM), and the prediction accuracy is greater than or equal to 91.7%.
7. 7. The method of claim 1, further comprising a multi-class expansion step of expanding the two-class quantum neural network into a multi-class model based on a decomposition strategy such as SVM, expansion OVO, DAG, RBF or ECOC, and adapting to 3-class to 8-class phase classification tasks.

Description

High-entropy alloy phase prediction method based on codon quantum neural network Technical Field The invention relates to the technical field of intersection of material science, quantum computation and machine learning, in particular to a high-entropy alloy phase prediction method based on a codon quantum neural network. Background The high-entropy alloy is a novel metal material containing a plurality of main elements (the atomic ratio is 5% -35%), and the high-entropy alloy (HEAs) is used as a novel alloy material consisting of a plurality of main elements, and has wide application prospect in the fields of aerospace, energy equipment and the like due to excellent mechanical, physical and chemical properties. The phase structure of the high-entropy alloy directly determines the core performance of the high-entropy alloy, however, the element composition of the high-entropy alloy is complex, the phase state is various and is easily influenced by a preparation process, so that the phase structure is extremely difficult to predict. The classical machine learning method improves the prediction effect to a certain extent, but has the defects of insufficient generalization capability and higher misjudgment rate when facing complex scenes such as multi-phase coexistence of high-entropy alloy, nonlinear association of characteristics and the like. The integration of quantum computing and machine learning provides a new path for solving the complex classification problem, and Quantum Machine Learning (QML) can efficiently process high-dimensional nonlinear data by utilizing the superposition and entanglement of quantum states. In the high-entropy alloy phase prediction, the existing quantum machine learning model as shown in fig. 2 has the problems of low feature coding efficiency, poor adaptability of a model structure, insufficient multi-classification processing capability and the like, and the advantage of the characteristic dimension of the quantum state amplification is not fully utilized, so that the prediction performance is difficult to meet the actual application requirements. Therefore, a method for realizing accurate prediction of high-entropy alloy phase through feature dimension amplification and model structure optimization by effectively integrating quantum computing advantages and classical machine learning stability is needed. Disclosure of Invention Aiming at the problems of low accuracy, weak generalization capability, poor adaptability of complex multi-classification scenes, capability of only carrying out target prediction below 3 types at present and the like in the existing high-entropy alloy phase prediction method, the invention provides a codon quantum neural network multiphase prediction method. The method is suitable for classifying and predicting the multi-class phase structures such as high-entropy alloy solid solutions, intermetallic compounds, amorphous states and the like, and can be applied to the design, preparation and performance optimization scenes of novel high-entropy alloy materials. Furthermore, the method realizes high-dimensional amplification of classical features through quantum codons, combines an optimized quantum neural network structure and a multi-classification strategy, remarkably improves the accuracy, stability and generalization capability of high-entropy alloy phase prediction, and provides technical support for efficient design of novel high-entropy alloy materials. In order to achieve the above purpose, the invention provides a machine learning high entropy alloy phase prediction method based on semi-quantum neural network codons, which has the following core technical scheme: The method comprises the following steps: Step1, data acquisition and preprocessing, collecting high-entropy alloy related data, classifying and labeling, extracting characteristic parameters, performing standardization processing, and filtering erroneous judgment samples. Step2, quantum codon generation, namely mapping the classical feature after pretreatment into a high-dimensional quantum state through quantum coding, and obtaining a quantum codon feature vector through measurement and decoding. Step3, constructing a codon quantum neural network, designing a feature embedding module, a parameterized rotation-entanglement module and a multi-classification discrimination module, and forming a complete prediction model. Step4, model training and optimization, namely training the model by adopting a super-parameter searching and regularization strategy and optimizing parameters. Step5, prediction and experimental verification, namely performing high-entropy alloy phase prediction by using a trained model, and verifying the accuracy of a prediction result through an electron microscope experiment. Further, the data acquisition and preprocessing in step 1 specifically includes: collecting 456 groups of high-entropy alloy data which are prepared by additive manufacturing and cover 42 e