High-entropy alloy performance prediction method and system based on TME composite algorithm

Abstract

The invention is suitable for the technical field of alloy material analysis, and provides a TME composite algorithm-based high-entropy alloy performance prediction method and a TME composite algorithm-based high-entropy alloy performance prediction system, and the method comprises the steps of firstly obtaining hardness characteristic data set information of a target alloy, and then inputting the hardness characteristic data set information into a trained multi-layer perceptron model, and performance prediction result information can be quickly and accurately generated. According to the method, the advantages of the converter model, the multi-layer perceptron model and the egret swarm optimization algorithm can be combined, the precision of high-entropy alloy performance prediction is effectively improved, the prediction cost and the calculation amount are reduced, the prediction precision and the generalization ability of the model are enhanced, and the method is suitable for performance prediction of various high-entropy alloys and has wide application prospects. Efficient application of the prediction model under different alloy components and process conditions is ensured, and reliable technical support and reference are provided for research, development and application of the high-entropy alloy.

Description

Technical Field

[0001] The present application relates to the technical field of alloy material analysis, and in particular to a high entropy alloy performance prediction method and system based on a TME composite algorithm. Background Art

[0002] High Entropy Alloys (HEAs) are a new type of multi-principal alloys, usually composed of five or more elements in equiatomic or nearly equiatomic ratios. Due to their unique diversified composition, high entropy alloys show remarkable excellence in structure and performance, such as high hardness, high strength, excellent corrosion resistance and oxidation resistance, etc., so they have broad application prospects in aerospace, defense industry, nuclear energy, electronic devices and other fields.

[0003] At present, molecular dynamics methods are usually used to simulate the microstructure and mechanical behavior of high-entropy alloys. However, the calculation accuracy of this method depends on the choice of potential function, which makes it difficult to accurately predict the actual properties of high-entropy alloys. There is a problem of low accuracy, which needs further improvement. Summary of the invention

[0004] Based on this, the embodiments of the present application provide a high entropy alloy performance prediction method and system based on the TME composite algorithm to solve the problem of low accuracy in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting high entropy alloy properties based on a TME composite algorithm, the method comprising:

[0006] Acquire hardness characteristic data set information of a target alloy, wherein the target alloy includes a plurality of metal elements, and the hardness characteristic data set information includes component specific gravity information corresponding to each of the metal elements;

[0007] The hardness feature data set information is input into a trained multi-layer perceptron model to generate performance prediction result information.

[0008] Compared with the prior art, the beneficial effect is that in the high-entropy alloy performance prediction method based on the TME composite algorithm provided in the embodiment of the present application, the terminal device can first obtain the hardness characteristic data set information of the target alloy, and then input the hardness characteristic data set information into the trained multi-layer perceptron model to quickly and accurately generate performance prediction result information, thereby effectively improving the accuracy of high-entropy alloy performance prediction, accurately predicting the actual performance of high-entropy alloys, with high accuracy, and to a certain extent solving the current problem of low accuracy.

[0009] In a second aspect, an embodiment of the present application provides a high entropy alloy performance prediction system based on a TME composite algorithm, the system comprising:

[0010] A hardness characteristic data set information acquisition module is used to acquire hardness characteristic data set information of a target alloy, wherein the target alloy includes a plurality of metal elements, and the hardness characteristic data set information includes component specific gravity information corresponding to each of the metal elements;

[0011] Performance prediction result information generation module: used to input the hardness feature data set information into the trained multi-layer perceptron model to generate performance prediction result information.

[0012] In a third aspect, an embodiment of the present application provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of the first aspect described above when executing the computer program.

[0013] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method of the first aspect described above are implemented.

[0014] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0016] Figure 1 It is a schematic flow chart of a method for predicting high entropy alloy properties provided in one embodiment of the present application;

[0017] Figure 2 It is a schematic diagram of the process after step S100 in the high entropy alloy performance prediction method provided in one embodiment of the present application;

[0018] Figure 3 It is a flow chart of step S1021 in the high entropy alloy performance prediction method provided in one embodiment of the present application;

[0019] Figure 4 It is a schematic flow chart of step S103 in the high entropy alloy performance prediction method provided in one embodiment of the present application;

[0020] Figure 5 It is a flow chart of step S10391 in the high entropy alloy performance prediction method provided in one embodiment of the present application;

[0021] Figure 6It is a schematic diagram of the process after step S10395 in the high entropy alloy performance prediction method provided in one embodiment of the present application;

[0022] Figure 7 is a schematic diagram of a change curve of the mean square error provided by an embodiment of the present application;

[0023] Figure 8 is a schematic diagram of a variation curve of a coefficient of determination provided by an embodiment of the present application;

[0024] Fig. 9 is a schematic diagram of an iterative learning curve provided by an embodiment of the present application;

[0025] Fig.10 This is a module block diagram of a high entropy alloy performance prediction system provided in one embodiment of the present application;

[0026] Fig.11 It is a schematic diagram of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0028] In the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0029] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0030] In order to illustrate the technical solution described in this application, a specific embodiment is provided below for illustration.

[0031] See also Figure 1 , Figure 1It is a schematic flow diagram of a high entropy alloy performance prediction method based on a TME composite algorithm provided in an embodiment of the present application. In the present embodiment, the execution subject of the high entropy alloy performance prediction method is a terminal device. It is understandable that the types of terminal devices include but are not limited to mobile phones, tablet computers, laptop computers, ultra-mobile personal computers (Ultra-Mobile Personal Computer, UMPC), netbooks, personal digital assistants (Personal Digital Assistant, PDA), etc., and the present embodiment of the application does not impose any restrictions on the specific types of terminal devices.

[0032] See also Figure 1 The high entropy alloy performance prediction method provided in the embodiment of the present application includes but is not limited to the following steps:

[0033] In S100, hardness characteristic data set information of the target alloy is obtained.

[0034] Specifically, the terminal device can first obtain the hardness characteristic data set information of the target alloy, wherein the target alloy includes multiple metal elements. For example, when the high entropy alloy system of the target alloy is the Al-Co-Cr-Cu-Fe-Ni system, the multiple metal elements can be aluminum (Al), cobalt (Co), chromium (Cr), copper (Cu), iron (Fe) and nickel (Ni); the hardness characteristic data set information includes the composition ratio information corresponding to each metal element, and the composition ratio information is used to describe the composition ratio of the metal element, and the composition ratio information can be a molar coefficient. In a possible implementation, the terminal device can obtain the Vickers hardness corresponding to each metal element as a label; the terminal device can pre-process the hardness characteristic data set information by cleaning and screening, so as to remove noise data and outliers and ensure the quality of the data.

[0035] In order to train an effective multilayer perceptron model, see Figure 2 After step S100, the method further includes but is not limited to the following steps:

[0036] In S101, an untrained multilayer perceptron model and preset alloy training data set information are obtained.

[0037] Specifically, the terminal device may first obtain an untrained multi-layer perceptron (MLP) model and preset alloy training data set information, wherein the alloy training data set information is used to describe the portion of the hardness characteristic data set information used as a training set.

[0038] In S102, based on a preset converter model, data enhancement processing is performed on the alloy training data set information to generate enhanced data set information.

[0039] Specifically, after the terminal device obtains the multi-layer perceptron model and the alloy training data set information, the terminal device can perform data enhancement processing on the alloy training data set information based on a preset transformer model to generate enhanced data set information.

[0040] In some possible implementations, in order to facilitate the construction of an effective multilayer perceptron model, after step S101, the method further includes but is not limited to the following steps:

[0041] In S1011, based on a preset division ratio, the alloy training data set information is divided and processed to generate test set information and training set information.

[0042] Specifically, the terminal device may first divide and process the alloy training data set information based on a preset division ratio to generate test set information and training set information, wherein the test set information may account for 20% and the training set information may account for 80%.

[0043] Accordingly, the above step S102 includes but is not limited to the following steps:

[0044] In S1021, based on the preset converter model, data enhancement processing is performed on the training set information to generate enhanced data set information.

[0045] Specifically, after the terminal device generates test set information and training set information, the terminal device can perform data enhancement processing on the training set information based on a preset converter model to quickly generate enhanced data set information.

[0046] In some possible implementations, in order to facilitate the training of effective multilayer perceptron models, see Figure 3 Step S1021 includes but is not limited to the following steps:

[0047] In S10211, the training set information is linearly transformed and mapped to a high-dimensional space to generate transformed data set information.

[0048] Specifically, the terminal device can linearly transform the training set information and map it to a high-dimensional space, that is, take the percentage content of the six metal elements Al, Co, Cr, Cu, Fe and Ni as input feature data, and perform linear transformation on these data, map them to a high-dimensional space, and generate transformed data set information, wherein the transformed data set information is used to describe the training set information after being mapped to the high-dimensional space.

[0049] In S10212, sequence information set information is generated based on a preset position coding formula set and transformation data set information.

[0050] Specifically, after the terminal device generates the transformation data set information, the terminal device may generate the sequence information set information based on the preset position coding formula set and the transformation data set information, thereby adding the position coding to retain the sequence information.

[0051] In a possible implementation, the above position encoding formula set may be:

[0052]

[0053] In S10213, data similarity information is generated based on the core formula of the attention mechanism corresponding to the preset converter model and the transformed data set information.

[0054] Specifically, after the terminal device generates sequence information set information, the terminal device can generate data similarity information based on the core formula of the attention mechanism and the transformation data set information corresponding to the preset converter model, thereby calculating the similarity between input data through a multi-head self-attention mechanism and capturing global dependencies.

[0055] In one possible implementation, the core formula of the attention mechanism can be:

[0056]

[0057] In the formula, Q is the abbreviation of Query, which means query vector; K is the abbreviation of Key, which means key vector; V is the abbreviation of Value, which means value vector; dk represents the dimension of key vector.

[0058] In S10214, based on the preset multi-head self-attention mechanism, multi-head expansion processing is performed on the data similarity information to generate extended data information.

[0059] Specifically, after the terminal device generates data similarity information, the terminal device can perform multi-head expansion processing on the data similarity information based on a preset multi-head self-attention mechanism to generate extended data information, thereby realizing the use of the multi-head self-attention mechanism to expand the above calculation into multiple heads to capture information in different subspaces.

[0060] In a possible implementation, the multi-head self-attention mechanism can be:

[0061] MultiHead(Q,K,V)=Concat(head1,head2,…,head h )W O .

[0062] In S10215, the multi-head output data information is determined based on the extended data information.

[0063] Specifically, after the terminal device generates the extended data information, the terminal device can determine the multi-head output data information according to the extended data information.

[0064] In a possible implementation, the multi-head output data information may be:

[0065] head i =Attention(QW i Q ,KW i K ,VW i V ).

[0066] In S10216, residual connection processing and layer normalization processing are performed on the multi-head output data information to generate first optimized data information.

[0067] Specifically, after the terminal device determines the multi-head output data information, the terminal device can perform residual connection processing and layer normalization processing on the multi-head output data information to generate first optimized data information, thereby preventing gradient disappearance and explosion, wherein the first optimized data information is used to describe the multi-head output data information after residual connection processing and layer normalization processing.

[0068] Exemplarily, the terminal device may first add the input features to the output of the multi-head attention mechanism, and then perform layer normalization on the addition result.

[0069] In S10217, based on a preset feedforward neural network, feature enhancement processing is performed on the optimized data information to generate second optimized data information.

[0070] Specifically, after the terminal device generates the first optimized data information, the terminal device can perform feature enhancement processing on the optimized data information based on a preset feedforward neural network to generate second optimized data information, thereby achieving enhanced feature representation, wherein the second optimized data information is used to describe the optimized data information after feature enhancement processing.

[0071] In a possible implementation, the feedforward neural network may be composed of two linear transformations and an activation function. The feedforward neural network may be:

[0072] FFN(x)=max(0,xW1+b1)W2+b2.

[0073] In S10218, the second optimized data information is subjected to secondary residual connection processing and secondary layer normalization processing to generate enhanced data set information.

[0074] Specifically, after the terminal device generates the second optimized data information, the terminal device can perform secondary residual connection processing and secondary layer normalization processing on the second optimized data information, that is, perform residual connection processing and layer normalization processing on the second optimized data information again to generate enhanced data set information, thereby ensuring data stability, so that the subsequent MLP model provides screened and enhanced feature data, and at the same time, it can effectively extract important information from the input feature data, reduce noise and redundant features, and improve the prediction accuracy and robustness of the subsequent model.

[0075] In S103, the untrained multilayer perceptron model is trained according to the enhanced data set information to generate a trained multilayer perceptron model.

[0076] Specifically, after the terminal device generates the enhanced data set information, the terminal device may train an untrained multi-layer perceptron model according to the enhanced data set information to generate a trained multi-layer perceptron model.

[0077] For some possible implementations, see Figure 4 Step S103 includes but is not limited to the following steps:

[0078] In S1031, the preset egret flock optimization algorithm is initialized to generate initial position information and initial speed information of a plurality of egrets.

[0079] Specifically, the terminal device can initialize the preset egret swarm optimization algorithm (ESOA) to generate initial position information and initial speed information of multiple egrets, wherein the initial position information is used to describe the initial position of the egret; the initial speed information is used to describe the initial speed of the egret.

[0080] In S1032, initial fitness information of a plurality of egrets is generated according to a preset objective function.

[0081] Specifically, after the terminal device generates initial position information and initial speed information, the terminal device can generate initial fitness information of each egret according to a preset objective function, thereby determining the fitness value of each egret through the objective function, wherein the initial fitness information is used to describe the initial fitness of the egret.

[0082] In S1033, the egret with the best initial fitness information is determined as the optimal egret in the current population.

[0083] Specifically, after the terminal device generates the initial fitness information, the terminal device can evaluate the fitness of each egret and determine the egret with the best initial fitness information as the optimal egret of the current population, where the optimal egret of the current population is used to describe the optimal egret of the current population.

[0084] In S1034, based on a preset position update formula, the updated position information of the optimal egret in the current population is determined.

[0085] Specifically, after the terminal device determines the optimal egret in the current population, the terminal device can determine the updated position information of the optimal egret in the current population based on a preset position update formula, thereby updating the position of the egret population based on determining the optimal egret in the current population.

[0086] In a possible implementation, the above position update formula may be:

[0087] x i (t+1)=x i (t)+v i (t+1).

[0088] In the formula, x i (t) represents the position of the ith egret at time t; v j (t+1) represents the speed of the ith egret at time t+1; x i (t+1) represents the position of the i-th egret at time t+1, which is the global optimal position.

[0089] In S1035, based on a preset speed update formula, the update speed information of the optimal egret in the current population is determined.

[0090] Specifically, after the terminal device determines to update the position information, the terminal device can determine the update speed information of the optimal egret in the current population based on a preset speed update formula, thereby updating the speed of the egret population based on determining the optimal egret in the current population.

[0091] In a possible implementation, the speed update formula may be:

[0092] v i (t+1)=ωv i (t)+c1r1(p i -x i (t))+c2r2(gx i (t)).

[0093] In the formula, v j(t) represents the speed of the ith egret at time t; ω is the preset inertia weight; c1 and c2 are the preset acceleration coefficients; r1 and r2 are any random numbers in the range of 0 to 1, r1 and r2 can be 0 or 1; p i Describe the optimal individual position of the i-th egret.

[0094] In S1036, the current iteration number information is obtained.

[0095] Specifically, after the terminal device determines the update speed information, the terminal device may obtain current iteration number information, wherein the current iteration number information is used to describe the number of current iterations.

[0096] In S1037, the current iteration number information is compared with the preset iteration number upper limit information.

[0097] Specifically, after the terminal device obtains the current iteration number information, the terminal device may compare the current iteration number information with the preset iteration number upper limit information.

[0098] In S1038, if the current number of iterations is less than the upper limit of the number of iterations, the current fitness information of the optimal egret in the current population is generated according to the objective function, and the preset position update formula is executed again to determine the updated position information of the optimal egret in the current population to compare the current number of iterations with the preset upper limit of the number of iterations, until the current number of iterations reaches the preset upper limit of the number of iterations, or meets the preset convergence conditions.

[0099] Specifically, if the current number of iterations is less than the upper limit of the number of iterations, the terminal device can generate the current fitness information of the optimal egret in the current population based on the objective function, and execute steps S1034 to S1037 again until the current number of iterations reaches the preset upper limit of the number of iterations, or satisfies the preset convergence conditions, thereby evaluating the fitness of the new position after each iteration and updating the optimal position. When the predetermined number of iterations is reached or the convergence conditions are met, the iteration is terminated, and the position of the optimal egret is effectively output as the final result.

[0100] In S1039, the final updated position information is determined to be the optimal hyperparameter combination information.

[0101] Specifically, after the terminal device terminates the iteration, the terminal device can determine that the final updated position information is the optimal hyperparameter combination information, thereby finding the optimal hyperparameter combination of the MLP model.

[0102] In S10391, based on the optimal hyperparameter combination information, the untrained multilayer perceptron model is trained to generate a trained multilayer perceptron model.

[0103] Specifically, after the terminal device determines the optimal hyperparameter combination information, the terminal device can train an untrained multilayer perceptron model based on the optimal hyperparameter combination information to generate a trained multilayer perceptron model.

[0104] Among some possible implementations, to effectively train a multilayer perceptron model, see Figure 5 Step S10391 includes but is not limited to the following steps:

[0105] In S10392, based on the input layer of the multilayer perceptron model, the optimal hyperparameter combination information is mapped to the hidden layer of the multilayer perceptron model to generate neuron output information.

[0106] Specifically, the terminal device can first map the optimal hyperparameter combination information to the hidden layer of the multilayer perceptron model based on the input layer of the multilayer perceptron model to generate neuron output information, thereby mapping the feature data to the hidden layer through the input layer, so as to input the feature data filtered and enhanced by the Transformer attention mechanism module into the MLP model, where the neuron output information is used to describe the output of the neuron.

[0107] In one possible implementation, the output of each neuron can be:

[0108]

[0109] In the formula, x i represents the input feature; w ij represents weight; b j represents the bias; σ represents the activation function, such as the ReLU function or the Sigmoid function.

[0110] In S10393, based on the preset output layer calculation formula and neuron output information, the predicted hardness value information is generated.

[0111] Specifically, after the terminal device generates neuron output information, the terminal device can generate predicted hardness value information based on a preset output layer calculation formula and neuron output information, thereby repeating the above calculation in the hidden layer and passing it to the output layer layer by layer.

[0112] In a possible implementation, the calculation formula of the output layer may be:

[0113]

[0114] In the formula, y k Indicates the predicted hardness value information.

[0115] In S10394, based on the preset minimization loss function and the predicted hardness value information, it is determined whether the loss meets the requirements.

[0116] Specifically, after the terminal device generates the predicted hardness value information, the terminal device can determine whether the loss meets the requirements based on the preset minimization loss function and the predicted hardness value information.

[0117] In a possible implementation, the above minimization loss function may be:

[0118]

[0119] In the formula, y i Indicates the actual hardness value, represents the predicted hardness value, and N represents the number of samples.

[0120] In S10395, if the loss meets the requirements, the multilayer perceptron model is determined to be a trained multilayer perceptron model. Otherwise, based on the back propagation algorithm, the preset weight values ​​and bias values ​​are adjusted and processed, and the preset output layer calculation formula and neuron output information are executed again to generate predicted hardness value information based on the preset minimization loss function and predicted hardness value information, and judge whether the loss meets the requirements until the loss meets the requirements.

[0121] Specifically, if the loss meets the requirements, the terminal device can determine that the multilayer perceptron model is a trained multilayer perceptron model, otherwise, based on the back propagation algorithm, the preset weight values ​​and bias values ​​are adjusted to minimize the loss function, and the above steps S10393 to S10394 are performed again until the loss meets the requirements. It should be noted that during the training process, the training set data and labels can be combined.

[0122] In order to quantify the prediction accuracy of the model, see Figure 6 After step S10395, the method further includes but is not limited to the following steps:

[0123] In S104, based on a preset mean square error formula and a multi-layer perceptron model, mean square error value information is generated.

[0124] Specifically, the terminal device can generate mean square error value information based on a preset mean square error formula and a multi-layer perceptron model, wherein the mean square error value information is used to describe the mean square error (MSE), and the mean square error value information can be used as an evaluation indicator for quantifying the prediction accuracy of the model.

[0125] In a possible implementation, the above mean square error formula may be:

[0126]

[0127] In the formula, y i Indicates the actual hardness value, represents the predicted hardness value, and N represents the number of samples. For example, see Figure 7 , the lower the MSE value, the smaller the error between the predicted hardness value and the actual value.

[0128] In S105, determination coefficient value information is generated based on a preset determination coefficient formula and a multi-layer perceptron model.

[0129] Specifically, after the terminal device generates the mean square error value information, the terminal device can generate the determination coefficient value information based on the preset determination coefficient formula and the multi-layer perceptron model, wherein the determination coefficient value information is used to describe the determination coefficient R 2 ,The determination coefficient value information can also be used as an evaluation indicator for ,quantifying the prediction accuracy of the model.

[0130] In a possible implementation, the above determination coefficient formula may be:

[0131]

[0132] In the formula, represents the mean of the actual hardness values. For example, see Figure 8 , R 2 The closer the value is to 1, the better the prediction performance of the model.

[0133] In S106, determination coefficient deviation value information is generated based on the difference between the determination coefficient value information and a preset determination coefficient threshold value.

[0134] Specifically, after the terminal device generates the determination coefficient value information, the terminal device may generate the determination coefficient deviation value information according to the difference between the determination coefficient value information and a preset determination coefficient threshold value.

[0135] In S107, if the mean square error value information is smaller than the preset error threshold, and the determination coefficient deviation value information is smaller than the preset deviation threshold, prediction performance compliance information is generated.

[0136] Specifically, if the mean square error value information is less than the preset error threshold, and the determination coefficient deviation value information is less than the preset deviation threshold, the terminal device can generate prediction performance compliance information, where the prediction performance compliance information is used to describe whether the prediction performance of the trained multi-layer perceptron model has met the standard.

[0137] Without loss of generality, see Fig. 9, with the increase of the number of training iterations (Epoch), both the training error and the test error decreased significantly, and stabilized after about 40 iterations. Finally, both the training error and the test error converged to a lower level, indicating that the model performed well on both the training set and the test set, without overfitting or underfitting. In a possible implementation, the terminal device can also construct a scatter plot of the actual hardness value and the predicted hardness value. In the scatter plot, the predicted hardness value and the actual hardness value are roughly distributed on a diagonal line, indicating that the predicted hardness value of the model is highly consistent with the actual hardness value. Therefore, the prediction method provided in the present application is in line with the performance of the prediction of the hardness of high entropy alloys, and has high prediction accuracy and stability.

[0138] In S200, the hardness feature data set information is input into the trained multi-layer perceptron model to generate performance prediction result information.

[0139] Specifically, after the terminal device obtains the hardness feature data set information, the terminal device can input the hardness feature data set information into a trained multi-layer perceptron model to quickly and accurately generate performance prediction result information, wherein the performance prediction result information is used to describe the performance prediction results of the target alloy, such as the hardness, tensile strength, yield strength, elongation, electrochemical corrosion and other properties of the high entropy alloy.

[0140] The implementation principle of the high-entropy alloy performance prediction method based on the TME composite algorithm in the embodiment of the present application is: the terminal device can first obtain the hardness characteristic data set information of the target alloy, and then input the hardness characteristic data set information into the trained multi-layer perceptron model to quickly and accurately generate performance prediction result information, thereby combining the advantages of the three algorithms of the converter model, the multi-layer perceptron model and the egret swarm optimization algorithm, effectively improving the accuracy of high-entropy alloy performance prediction, accurately predicting the actual performance of high-entropy alloys, reducing prediction costs and calculations, enhancing the prediction accuracy and generalization ability of the model, and being suitable for the performance prediction of various high-entropy alloys, a multi-component material, ensuring the efficient application of the prediction model under different alloy compositions and process conditions, and providing reliable technical support and reference for the research and development and application of high-entropy alloys with high accuracy.

[0141] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0142] The embodiment of the present application also provides a high entropy alloy performance prediction system based on the TME composite algorithm. For ease of description, only the parts related to the present application are shown, such as Fig.10 As shown, the system 100 includes:

[0143] The hardness characteristic data set information acquisition module 101 is used to acquire the hardness characteristic data set information of the target alloy, wherein the target alloy includes a plurality of metal elements, and the hardness characteristic data set information includes the component specific gravity information corresponding to each metal element;

[0144] Performance prediction result information generating module 102: used to input the hardness feature data set information into the trained multi-layer perceptron model to generate performance prediction result information.

[0145] Optionally, the system 100 further includes:

[0146] Multilayer perceptron model acquisition module: used to obtain untrained multilayer perceptron model and preset alloy training data set information;

[0147] Strengthening data set information generation module: used to perform data strengthening processing on the alloy training data set information based on the preset converter model to generate strengthening data set information;

[0148] Multilayer perceptron model generation module: used to train an untrained multilayer perceptron model according to the enhanced data set information to generate a trained multilayer perceptron model.

[0149] Optionally, the system 100 further includes:

[0150] Alloy training data set information partitioning module: used to partition the alloy training data set information based on a preset partitioning ratio to generate test set information and training set information.

[0151] Optionally, the enhanced data set information generation module includes:

[0152] Enhanced data set information generation submodule: used to perform data enhancement processing on the training set information based on the preset converter model to generate enhanced data set information.

[0153] Optionally, the above-mentioned enhanced data set information generation submodule includes:

[0154] Transformation data set information generation unit: used to linearly transform and map the training set information to a high-dimensional space to generate transformation data set information;

[0155] A sequence information set information generating unit: used to generate sequence information set information based on a preset position coding formula set and transformation data set information;

[0156] Data similarity information generation unit: used to generate data similarity information based on the core formula of the attention mechanism corresponding to the preset converter model and the transformed data set information;

[0157] Extended data information generating unit: used for performing multi-head expansion processing on data similarity information based on a preset multi-head self-attention mechanism to generate extended data information;

[0158] A multi-head output data information determining unit: used to determine the multi-head output data information according to the extended data information;

[0159] A first optimized data information generating unit: used for performing residual connection processing and layer normalization processing on multi-head output data information to generate first optimized data information;

[0160] A second optimized data information generating unit: configured to perform feature enhancement processing on the optimized data information based on a preset feedforward neural network to generate second optimized data information;

[0161] Enhanced data set information generating unit: used for performing secondary residual connection processing and secondary layer normalization processing on the second optimized data information to generate enhanced data set information.

[0162] Optionally, the multilayer perceptron model generation module includes:

[0163] Initial position information generation submodule: used to initialize the preset egret group optimization algorithm and generate initial position information and initial speed information of multiple egrets;

[0164] Initial fitness information generation submodule: used to generate initial fitness information of multiple egrets according to a preset objective function;

[0165] The optimal egret determination submodule of the current population is used to determine the egret with the best initial fitness information as the optimal egret of the current population;

[0166] Update position information determination submodule: used to determine the updated position information of the optimal egret in the current population based on a preset position update formula;

[0167] Update speed information determination submodule: used to determine the update speed information of the optimal egret of the current population based on a preset speed update formula;

[0168] Current iteration number information acquisition submodule: used to obtain the current iteration number information;

[0169] Current iteration number information comparison submodule: used to compare the current iteration number information with the preset iteration number upper limit information;

[0170] Loop execution submodule: used for generating the current fitness information of the optimal egret of the current population according to the objective function if the current iteration number information is less than the iteration number upper limit information, and executing again based on the preset position update formula to determine the updated position information of the optimal egret of the current population to compare the current iteration number information with the preset iteration number upper limit information, until the current iteration number information reaches the preset iteration number upper limit information, or satisfies the preset convergence condition;

[0171] Optimal hyperparameter combination information determination submodule: used to determine that the final update position information is the optimal hyperparameter combination information;

[0172] Multilayer perceptron model generation submodule: used to train an untrained multilayer perceptron model based on the optimal hyperparameter combination information to generate a trained multilayer perceptron model.

[0173] Optionally, the multilayer perceptron model generation submodule includes:

[0174] Neuron output information generation unit: used to map the optimal hyperparameter combination information to the hidden layer of the multilayer perceptron model based on the input layer of the multilayer perceptron model to generate neuron output information;

[0175] Predicted hardness value information generating unit: used to generate predicted hardness value information based on a preset output layer calculation formula and neuron output information;

[0176] Predicted hardness value information judgment unit: used to judge whether the loss meets the requirements based on the preset minimization loss function and the predicted hardness value information;

[0177] Loop execution unit: used to determine that the multilayer perceptron model is a trained multilayer perceptron model if the loss meets the requirements; otherwise, based on the back propagation algorithm, adjust the preset weight values ​​and bias values, and execute again based on the preset output layer calculation formula and neuron output information, generate predicted hardness value information based on the preset minimization loss function and predicted hardness value information, and judge whether the loss meets the requirements until the loss meets the requirements.

[0178] Optionally, the system 100 further includes:

[0179] Mean square error value information generation module: used to generate mean square error value information based on a preset mean square error formula and a multi-layer perceptron model;

[0180] A determination coefficient value information generation module: used to generate determination coefficient value information based on a preset determination coefficient formula and a multi-layer perceptron model;

[0181] A determination coefficient deviation value information generation module: used to generate determination coefficient deviation value information according to the difference between the determination coefficient value information and a preset determination coefficient threshold value;

[0182] Prediction performance compliance information generation module: used to generate prediction performance compliance information if the mean square error value information is less than the preset error threshold and the determination coefficient deviation value information is less than the preset deviation threshold.

[0183] It should be noted that the information interaction, execution process and other contents between the above-mentioned modules are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0184] The present application also provides a terminal device, such as Fig.11 As shown, the terminal device 110 of this embodiment includes: a processor 111, a memory 112, and a computer program 113 stored in the memory 112 and executable on the processor 111. When the processor 111 executes the computer program 113, the steps in the above-mentioned high entropy alloy performance prediction method embodiment are implemented, such as Figure 1 Steps S100 to S200 shown; or, when the processor 111 executes the computer program 113, the functions of each module in the above device are realized, for example Fig.10 The functions of modules 101 to 102 are shown.

[0185] The terminal device 110 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server, and the terminal device 110 includes but is not limited to a processor 111 and a memory 112. Those skilled in the art will appreciate that Fig.11 It is only an example of the terminal device 110 and does not constitute a limitation of the terminal device 110. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device 110 may also include input and output devices, network access devices, buses, etc.

[0186] Among them, the processor 111 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.; the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0187] The memory 112 may be an internal storage unit of the terminal device 110, such as a hard disk or memory of the terminal device 110, or an external storage device of the terminal device 110, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the terminal device 110; further, the memory 112 may also include both an internal storage unit and an external storage device of the terminal device 110, and the memory 112 may also store a computer program 113 and other programs and data required by the terminal device 110, and the memory 112 may also be used to temporarily store data that has been output or is to be output.

[0188] One embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the steps of each of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc.; the computer-readable medium may include: any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0189] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, all equivalent changes made according to the methods, principles, and structures of the present application should be included in the protection scope of the present application.

Claims

1. A high entropy alloy performance prediction method based on TME composite algorithm, characterized in that: The method comprises: Acquire hardness characteristic data set information of a target alloy, wherein the target alloy includes a plurality of metal elements, and the hardness characteristic data set information includes component specific gravity information corresponding to each of the metal elements; The hardness feature data set information is input into a trained multi-layer perceptron model to generate performance prediction result information.

2. The method according to claim 1, characterized in that After obtaining the hardness characteristic data set information of the target alloy, the method further includes: Get the untrained multilayer perceptron model and the preset alloy training data set information; Based on a preset converter model, performing data enhancement processing on the alloy training data set information to generate enhanced data set information; Training an untrained multilayer perceptron model according to the enhanced data set information to generate a trained multilayer perceptron model; Wherein, after obtaining the untrained multilayer perceptron model and the preset alloy training data set information, the method further includes: Based on a preset division ratio, the alloy training data set information is divided and processed to generate test set information and training set information; Accordingly, the data enhancement processing is performed on the alloy training data set information based on the preset converter model to generate enhanced data set information, including: Based on a preset converter model, data enhancement processing is performed on the training set information to generate enhanced data set information; The step of performing data enhancement processing on the training set information based on a preset converter model to generate enhanced data set information includes: Linearly transform and map the training set information to a high-dimensional space to generate transformed data set information; Generate sequence information set information based on a preset position coding formula set and the transformed data set information; Generate data similarity information based on the core formula of the attention mechanism corresponding to the preset converter model and the transformed data set information; Based on a preset multi-head self-attention mechanism, multi-head expansion processing is performed on the data similarity information to generate extended data information; Determining multi-head output data information according to the extended data information; Performing residual connection processing and layer normalization processing on the multi-head output data information to generate first optimized data information; Based on a preset feedforward neural network, feature enhancement processing is performed on the optimized data information to generate second optimized data information; The second optimized data information is subjected to secondary residual connection processing and secondary layer normalization processing to generate enhanced data set information.

3. The method according to claim 2, characterized in that The step of training an untrained multilayer perceptron model according to the enhanced data set information to generate a trained multilayer perceptron model includes: Initialize the preset egret group optimization algorithm to generate initial position information and initial speed information of multiple egrets; According to the preset objective function, the initial fitness information of multiple egrets is generated; Determine the egret with the best initial fitness information as the optimal egret in the current population; Based on a preset position update formula, determining the updated position information of the optimal egret in the current population; Based on a preset speed update formula, determining the update speed information of the optimal egret of the current population; Get the current iteration number information; Comparing the current number of iterations information with preset upper limit information of the number of iterations; If the current iteration number information is less than the iteration number upper limit information, then according to the objective function, generate the current fitness information of the optimal egret in the current population, and execute the preset position update formula again to determine the updated position information of the optimal egret in the current population to compare the current iteration number information with the preset iteration number upper limit information, until the current iteration number information reaches the preset iteration number upper limit information, or satisfies the preset convergence condition; Determining the final updated position information as the optimal hyperparameter combination information; Based on the optimal hyperparameter combination information, an untrained multilayer perceptron model is trained to generate a trained multilayer perceptron model.

4. The method according to claim 3, characterized in that The step of training an untrained multilayer perceptron model based on the optimal hyperparameter combination information to generate a trained multilayer perceptron model includes: Based on the input layer of the multilayer perceptron model, mapping the optimal hyperparameter combination information to the hidden layer of the multilayer perceptron model to generate neuron output information; Generate predicted hardness value information based on a preset output layer calculation formula and the neuron output information; Based on the preset minimization loss function and the predicted hardness value information, judging whether the loss meets the requirements; If the loss meets the requirements, the multilayer perceptron model is determined to be a trained multilayer perceptron model; otherwise, based on the back propagation algorithm, the preset weight values ​​and bias values ​​are adjusted, and the preset output layer calculation formula and the neuron output information are executed again to generate predicted hardness value information based on the preset minimization loss function and the predicted hardness value information, and determine whether the loss meets the requirements until the loss meets the requirements.

5. The method according to claim 4, characterized in that After determining that the multilayer perceptron model is a trained multilayer perceptron model if the loss meets the requirement, the method further includes: Based on a preset mean square error formula and the multi-layer perceptron model, generating mean square error value information; Based on a preset coefficient of determination formula and the multilayer perceptron model, generating coefficient of determination value information; Generate determination coefficient deviation value information according to the difference between the determination coefficient value information and a preset determination coefficient threshold value; If the mean square error value information is smaller than a preset error threshold, and the determination coefficient deviation value information is smaller than a preset deviation threshold, prediction performance compliance information is generated.

6. A high entropy alloy performance prediction system based on TME composite algorithm, characterized in that: The system comprises: A hardness characteristic data set information acquisition module is used to acquire hardness characteristic data set information of a target alloy, wherein the target alloy includes a plurality of metal elements, and the hardness characteristic data set information includes component specific gravity information corresponding to each of the metal elements; Performance prediction result information generation module: used to input the hardness feature data set information into the trained multi-layer perceptron model to generate performance prediction result information.

7. The system according to claim 6, characterized in that The system comprises: Multilayer perceptron model acquisition module: used to obtain untrained multilayer perceptron model and preset alloy training data set information; Strengthening data set information generation module: used for performing data strengthening processing on the alloy training data set information based on a preset converter model to generate strengthening data set information; Multilayer perceptron model generation module: used to train an untrained multilayer perceptron model according to the enhanced data set information to generate a trained multilayer perceptron model.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images