A multi-element alloy property prediction method based on particle swarm optimization BP neural network

Abstract

The present invention discloses a multi-component alloy performance prediction method based on a particle swarm optimized (PSO) BP neural network, comprising the following steps: Step 1: establishing a database based on historical material composition-performance data; Step 2: establishing and training a PSO BP neural network composition-performance prediction model, using the coefficient of determination as an evaluation criterion for prediction effectiveness; Step 3: subjecting the computational set obtained from the prediction model trained in Step 2 to a genetically optimized (GMO) BP neural network, using the coefficient of determination as an evaluation criterion for prediction effectiveness; Step 4: if the coefficient of determination calculated in Step 3 meets the judgment criteria, the multi-component alloy composition and microalloying element array to be predicted are substituted into the PSO BP neural network obtained in Step 2 to complete performance prediction; otherwise, Steps 2 and 3 are repeated. This prediction method is suitable for predicting the performance of multi-component alloys and microalloying alloys, and is applicable to the development of new material compositions. Industrially, it can predict performance changes due to smelting burnout.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-component alloy performance prediction, and in particular to a multi-component alloy performance prediction method based on particle swarm optimization BP neural network. Background Art

[0002] With the development of society and the advancement of science and technology, higher and more stringent requirements are being placed on the complex environments in which various materials are used and their service performance. In particular, the performance of metal materials such as magnesium alloys, aluminum alloys, zirconium alloys, titanium alloys, copper alloys, high-temperature alloys, and high-entropy alloys used in high-end fields such as aerospace, deep-sea exploration, electronic information, and energy transmission and storage will directly affect the service performance of equipment. Design methods for customized and proprietary materials have emerged, which propose basic alloy systems based on basic operating conditions and alloying requirements based on more specific operating environments. How to quickly and accurately determine the performance corresponding to the design composition has become a technical challenge that needs to be solved in the field of performance prediction of multi-component microalloying alloys.

[0003] Some relevant research has been carried out at home and abroad on the prediction of alloy composition and performance. For example, the patent "CN107609647A" discloses a neural network (BPNN) model for predicting the influence of different alloy compositions and heat treatment process parameters on mechanical properties; "CN111063401 A" discloses a method for predicting the microstructure and mechanical properties of heat-treated Mg-Zn-Zr alloys; "CN110010210 A" discloses a multi-element alloy prediction method based on the BP neural network model. These prediction methods are all based on the BP model, which uses the steepest descent method to continuously adjust the network weights and thresholds through back propagation. For the composition-performance prediction needs of complex materials, even if a multi-hidden layer BP model is used, its prediction effect is difficult to meet strict evaluation standards, and it is prone to overfitting, making it difficult to apply to actual predictions. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a multi-component alloy performance prediction method based on particle swarm optimization BP neural network, which can accurately predict the performance of multi-component alloys and micro-alloyed alloys while ensuring low computing time and computing resource consumption, and can also predict performance changes caused by smelting burnout.

[0005] The present invention provides a multi-element alloy property prediction method based on particle swarm optimization BP neural network, comprising:

[0006] Step 1: Establish a database based on the composition-property history data of a material in a specific state, including but not limited to: rolled, annealed, forged, cast, or extruded state;

[0007] Step 2: Establish and train a particle swarm optimization BP neural network component-performance prediction model, using the coefficient of determination as the evaluation criterion for the prediction effect. If the calculated coefficient of determination meets the judgment criterion, execute step 3; otherwise, execute step 2 again.

[0008] Step 3: The calculation set obtained from the prediction model trained in step 2 is introduced into the genetic optimization BP neural network, and the determination coefficient is also used as the evaluation standard for the prediction effect to verify the accuracy of the model;

[0009] Step 4: If the coefficient of determination calculated in step 3 meets the judgment criteria, the multi-element alloy composition and microalloying element array to be predicted are substituted into the particle swarm optimization BP neural network obtained in step 2 to complete the performance prediction; otherwise, repeat steps 2 and 3.

[0010] In the multi-element alloy property prediction method based on particle swarm optimization BP neural network of the present invention, the step 1 is specifically as follows:

[0011] Step 1-1: Collecting multi-component alloy composition data and performance data associated with the composition and used to determine the alloy usage standards to form a basic database;

[0012] Step 1-2: Divide the basic database into two parts: the particle swarm optimization BP neural network training set and the particle swarm optimization BP neural network validation set;

[0013] Step 1-3: Perform normalization processing between 0 and 1 on the particle swarm optimization BP neural network training set and the particle swarm optimization BP neural network verification set as the particle swarm optimization BP neural network operation set.

[0014] In the multi-component alloy property prediction method based on particle swarm optimization BP neural network of the present invention, the multi-component alloy includes: magnesium alloy, aluminum alloy, zirconium alloy, titanium alloy, copper alloy, high-temperature alloy and high-entropy alloy basic component system and added microalloying elements.

[0015] In the multi-component alloy performance prediction method based on particle swarm optimization BP neural network of the present invention, the performance data associated with the multi-component alloy system components and used to determine the alloy usage standard include: yield strength, ultimate tensile strength, hardness, elongation, compressibility, conductivity and corrosion rate.

[0016] In the multi-element alloy property prediction method based on particle swarm optimization BP neural network of the present invention, the steps 1-2 are specifically as follows:

[0017] When the number of basic databases is between 200 and 400, the particle swarm optimization BP neural network training set is 95% of the basic database, and the particle swarm optimization BP neural network validation set is 5% of the basic database; when the number of basic databases is greater than 400, the particle swarm optimization BP neural network training set is 90% of the basic database, and the particle swarm optimization BP neural network validation set is 10% of the basic database.

[0018] In the multi-element alloy property prediction method based on particle swarm optimization BP neural network of the present invention, the step 2 is specifically as follows:

[0019] Step 2-1: using the component data in the particle swarm optimization BP neural network operation set as input and the performance data as output, and setting the particle swarm optimization algorithm parameters, and using the particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network. The particle swarm parameters include population size, maximum update generations, learning factor, and inertia weight method;

[0020] Step 2-2: Establish a BP neural network, substitute the particle swarm optimized BP neural network operation set and the initial weights and thresholds after particle swarm optimization into the BP neural network, set the number of hidden layer nodes of the BP neural network, and establish a component-performance prediction model;

[0021] Step 2-3: Select the transfer function and training method to train the particle swarm optimization BP neural network, and select different determination coefficients as evaluation criteria according to the number of basic databases. When the number of basic databases is between 200 and 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤15%. When the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤10%. If it is within the evaluation standard range, go to step 3, otherwise recalculate.

[0022] In the multi-element alloy property prediction method based on particle swarm optimization BP neural network of the present invention, the step 3 is specifically as follows:

[0023] Step 3-1: Using the component data in the operation set screened by the particle swarm optimization BP neural network training as the input of the genetic algorithm to optimize the BP neural network, and the performance data as the output of the genetic algorithm to optimize the BP neural network, and setting the genetic algorithm parameters, and using the genetic algorithm to optimize the initial weights and thresholds of the BP neural network. The genetic algorithm parameters include population size, iteration termination generation number, crossover and mutation probability;

[0024] Step 3-2: Set the BP neural network parameters, select the transfer function and training method, and substitute the initial weights and thresholds optimized by the genetic algorithm into the BP neural network;

[0025] Step 3-3: When the number of basic databases is less than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤20%; when the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤15%.

[0026] The multi-element alloy property prediction method based on particle swarm optimization BP neural network of the present invention has at least the following beneficial effects:

[0027] 1. The present invention first optimizes the BP neural network through the particle swarm algorithm, and substitutes the calculation data obtained by the particle swarm algorithm to the genetic optimization BP neural network to verify the accuracy of the composition-performance prediction model. While ensuring low computing time and computing resource consumption, it can achieve accurate prediction of the properties of multi-element alloys and micro-alloyed alloys.

[0028] 2. It can realize the rapid and accurate prediction of the performance of multi-component alloys and micro-alloyed alloys, significantly improve the efficiency of new material composition research and development, meet the preliminary prediction of material properties after component burnout in actual industrial production processes, and promote the innovation ability and level of multi-component micro-alloyed alloys in the field of science and technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention provides a flowchart of a multi-element alloy performance prediction method based on particle swarm optimization BP neural network. DETAILED DESCRIPTION

[0030] like Figure 1 As shown in the figure, the present invention provides a multi-component alloy performance prediction method based on a particle swarm optimization BP neural network. This method can achieve rapid and accurate prediction of the performance of multi-component alloys and microalloyed alloys, significantly improving the efficiency of new material composition research and development, meeting the preliminary prediction of material performance after component burnout in actual industrial production processes, and promoting the innovation capability and level of multi-component microalloyed alloys in the field of science and technology. Specifically, it includes the following steps:

[0031] Step 1: Establish a database based on the composition-property history data of a material in a specific state. The specific state of the material includes but is not limited to: rolled state, annealed state, forged state, cast state or extruded state, specifically:

[0032] Step 1-1: Collecting multi-component alloy composition data and performance data associated with the composition and used to determine the alloy usage standards to form a basic database;

[0033] In specific implementation, a basic composition-performance database can be obtained by collecting and organizing the composition and performance data of multi-element alloys in publicly published literature.

[0034] In a specific implementation, the multi-component alloy includes: a basic component system of magnesium alloy, aluminum alloy, zirconium alloy, titanium alloy, copper alloy, high-temperature alloy and high-entropy alloy and added micro-alloying elements.

[0035] In a specific implementation, the performance data associated with the multi-component alloy system components and used to determine the alloy usage standard include: yield strength, ultimate tensile strength, hardness, elongation, compression rate, conductivity and corrosion rate.

[0036] Step 1-2: Divide the basic database into two parts: the particle swarm optimization BP neural network training set and the particle swarm optimization BP neural network validation set;

[0037] Duplicate data with the same composition but different performance parameters due to different heat treatment processes are cleaned and classified according to whether a second phase is generated.

[0038] When the number of basic databases is between 200 and 400, the particle swarm optimization BP neural network training set is 95% of the basic database, and the particle swarm optimization BP neural network validation set is 5% of the basic database; when the number of basic databases is greater than 400, the particle swarm optimization BP neural network training set is 90% of the basic database, and the particle swarm optimization BP neural network validation set is 10% of the basic database.

[0039] Step 1-3: In order to make the calculation more accurate, both the particle swarm optimization BP neural network training set and the particle swarm optimization BP neural network verification set are normalized between 0 and 1 as the particle swarm optimization BP neural network operation set.

[0040] Step 2: Establish and train a particle swarm optimization BP neural network component-performance prediction model, using the determination coefficient as the evaluation criterion for the prediction effect. If the calculated determination coefficient meets the judgment criterion, execute step 3, otherwise execute step 2 again. Step 2 is specifically as follows:

[0041] Step 2-1: using the component data in the particle swarm optimization BP neural network operation set as input and the performance data as output, and setting the particle swarm optimization algorithm parameters, and using the particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network; the particle swarm parameters include population size, maximum update generations, learning factor, and inertia weight method;

[0042] In practice, methods for adjusting the inertia weight of the particle swarm algorithm include linearly decreasing weights, adaptively adjusting weights, and random weights. The following examples all use linearly decreasing weights. Methods for adjusting the learning factor of the particle swarm algorithm include shrinkage factors, synchronous learning factors, and asynchronous learning factors. The following examples all use synchronous learning factors.

[0043] Step 2-2: Establish a BP neural network, substitute the particle swarm optimized BP neural network operation set and the initial weights and thresholds after particle swarm optimization into the BP neural network, set the number of hidden layer nodes of the BP neural network, and establish a component-performance prediction model;

[0044] The BP neural network model automatically divides the normalized training set in the particle swarm optimization BP neural network operation set into a BP model internal training set, a BP model internal validation set, and a BP model internal test set, with the sum of the proportions of each set being 1. The BP model internal training set is used to train the neural network, and the BP model internal validation set is used to prevent overfitting of the neural network. The training set and validation set are used during BP model training; the BP model internal test set is used to test the model's prediction accuracy and is used after training.

[0045] The internal validation of the BP model here differs from the 5% PSO BP neural network validation set in steps 1-2, which verifies model accuracy from a different perspective and with more stringent standards. The internal training set of the BP model here differs from the PSO BP neural network training set in steps 1-2, which verifies the entire PSO BP model, while this validation only verifies the internal model of the BP model.

[0046] The division ratio of the BP model internal training set, BP model internal validation set and BP model internal test set is determined by the number of particle swarm optimization BP neural network training sets. When the number of particle swarm optimization BP neural network training sets is between 200 and 400, the proportion of the BP model internal training set is 85% or 90%, the proportion of the BP model internal validation set is 10%, and the proportion of the BP model internal test set is 5% or 0; when the number of particle swarm optimization BP neural network training sets is more than 400, the proportion of the BP model internal training set is 70% to 90%, the proportion of the BP model internal validation set is 10% to 20%, and the proportion of the BP model internal test set is 10% to 20%.

[0047] An early termination strategy is adopted. After the validation set error decreases continuously for a set number of times, training is stopped and the results are output to prevent overfitting.

[0048] Step 2-3: Select the transfer function and training method to train the particle swarm optimization BP neural network, and select different determination coefficients as evaluation criteria according to the number of basic databases;

[0049] When the number of basic databases is between 200 and 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤15%; when the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤10%. If it is within the evaluation standard range, go to step 3, otherwise recalculate.

[0050] The training methods mainly include the steepest descent method, the additional momentum method, the adaptive learning rate method, the momentum-adaptive learning rate adjustment method and the quantized conjugate gradient algorithm. The following embodiments all adopt the steepest descent method.

[0051] In specific implementation, the determination coefficient is calculated according to the following formula:

[0052]

[0053] Among them, R 2 is the coefficient of determination; y i To verify the actual performance of the concentrated alloy; i Predicting performance for neural networks; is the average value of the actual performance of the alloy in the validation set, and n is the number of validation sets.

[0054] Step 3: The calculation set obtained from the prediction model trained in step 2 is brought into the genetic optimization BP neural network, and the determination coefficient is also used as the evaluation standard for the prediction effect to verify the accuracy of the model. The specific steps of step 3 are:

[0055] Step 3-1: Using the component data in the operation set screened by the particle swarm optimization BP neural network training as the input of the genetic algorithm to optimize the BP neural network, and the performance data as the output of the genetic algorithm to optimize the BP neural network, and setting the genetic algorithm parameters, and using the genetic algorithm to optimize the initial weights and thresholds of the BP neural network. The genetic algorithm parameters include population size, iteration termination generation number, crossover and mutation probability;

[0056] A small population size leads to fast convergence but reduces population diversity; a high crossover probability easily destroys the excellent structure already formed in the population, making the search too random; a smaller crossover probability makes it too slow to discover new individuals; a too small mutation probability leads to poor ability of mutation operations to generate new individuals and suppress premature maturation; a too high mutation probability leads to excessive randomness.

[0057] Step 3-2: Set the BP neural network parameters, select the transfer function and training method, and substitute the initial weights and thresholds optimized by the genetic algorithm into the BP neural network;

[0058] The BP parameter selection, transfer function, and training method settings are the same as in steps 2-2 and 2-3.

[0059] Step 3-3: When the number of basic databases is less than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤20%; when the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤15%.

[0060] Step 4: If the coefficient of determination calculated in step 3 meets the judgment criteria, the multi-element alloy composition and microalloying element array to be predicted are substituted into the particle swarm optimization BP neural network obtained in step 2 to complete the performance prediction; if the coefficient of determination calculated in step 3 does not meet the judgment criteria, repeat steps 2 and 3.

[0061] Since the ratio of the training set and the validation set of the particle swarm optimization BP neural network is fixed in the BP model, the data is randomly assigned, so the calculation can be repeated many times to obtain the optimal prediction effect. If a satisfactory result is not obtained after repeated calculations, the number of hidden layer nodes, learning factor parameters, inertia weight parameters, crossover and mutation probabilities can be modified and the above steps can be repeated.

[0062] Example 1

[0063] Using the method provided by the present invention, taking copper alloy as an example, specifically a Cu-Ni-Si alloy of lead frame material, a microalloying element Zn is added to the alloy system to improve its electrical conductivity and ultimate tensile strength, wherein Ni (2.5wt% to 3.5wt%, selected in steps of 0.01%), Si (0.4wt% to 0.8wt%, selected in steps of 0.01%), and Cr (0.1wt% to 0.6wt%, selected in steps of 0.01%) are the components for which performance is to be predicted. The prediction method is specifically described as follows:

[0064] (1) Database establishment: Collect and organize the cold-rolled and aged Cu-Ni-Si system data from publicly available literature to obtain a basic composition-property database with 510 data items. The properties mainly focus on ultimate tensile strength and electrical conductivity (international annealed copper standards);

[0065] (2) The repeated data with the same composition but different performance parameters due to different heat treatment processes were cleaned, and the Cr element was classified as a second-phase element. The database had 200 data items, 190 of which were used as training sets and 10 as validation sets;

[0066] (3) Data normalization: To make the calculation more accurate and convenient, the 190 training sets and 10 validation sets were normalized between 0 and 1 and used as the calculation set;

[0067] (4) using the integrated computing data as input and the performance data as output, setting the parameters of the particle swarm optimization algorithm, and optimizing the initial weights and thresholds of the BP neural network;

[0068] (5) Establishing a particle swarm optimization BP neural network model: Substitute the normalized operation set in (3) and the initial weights and thresholds calculated in (4) into the model, set the number of nodes in the hidden layer of the BP neural network to 10, and establish a Cu-Ni-Si-Zn alloy performance prediction model;

[0069] (6) 170 of the 190 particle swarm optimization BP neural network operation sets were used as the internal training set of the BP model, 10 were used as the internal validation set of the BP model, and 10 were used as the internal test set of the BP model. The number of early terminations was set to 10, that is, the training was stopped after the validation set error decreased 10 times in a row.

[0070] (7) The hidden layer transfer function selects the tan-sigmoid function, the output layer transfer function selects the linear transfer function, and the training method adopts the steepest descent method;

[0071] (8) The number of databases in this example is 200. The prediction effect is evaluated by the absolute value of the deviation of the determination coefficient from 1 ≤ 15%. The determination coefficient of the ultimate tensile strength is 0.8873 and the determination coefficient of the conductivity is 0.8988, which achieves the ideal prediction effect.

[0072] (9) Substitute the 190 operation data selected by the particle swarm optimization BP neural network training into the genetic optimization BP neural network, and set the genetic algorithm parameters including population size, termination evolution generation, crossover and mutation probability;

[0073] (10) The ultimate tensile strength and conductivity determination coefficients of the genetically optimized BP neural network were calculated to be 0.8895 and 0.8349, respectively. The determination coefficients met the criteria. The various parameters and data set positions in the particle swarm optimization BP neural network can be fixed to predict the properties of the Cu-Ni-Si-Cr alloy.

[0074] (11) Ni (2.5 wt% to 3.5 wt%, with a step size of 0.01%), Si (0.4 wt% to 0.8 wt%, with a step size of 0.01%), and Cr (0.1 wt% to 0.6 wt%, with a step size of 0.01%) were input into the composition permutation function, and each element was fully permuted in order and interval. The composition with an ultimate tensile strength of not less than 750 MPa and an electrical conductivity of not less than 40% IACS was selected, and the data points with a Ni / Si ratio between 4 and 4.5 were selected. The data shown in Table 1 were obtained:

[0075] Table 1

[0076]

[0077] The experimental results show that the relative errors between the actual ultimate tensile strength (UTS) and electrical conductivity (EC) of the alloy after aging and the predicted ultimate tensile strength (UTS) and electrical conductivity (EC) are less than 2%. Particle swarm optimization BP neural network can be used to predict the properties of multi-component alloys.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the concept of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-element alloy performance prediction method based on particle swarm optimization BP neural network, characterized in that: include: Step 1: Establish a database based on the composition-property history data of a material in a specific state, including but not limited to: rolled, annealed, forged, cast, or extruded state; Step 2: Establish and train a particle swarm optimization BP neural network component-performance prediction model, using the coefficient of determination as the evaluation criterion for the prediction effect. If the calculated coefficient of determination meets the judgment criterion, execute step 3; otherwise, execute step 2 again. Step 3: The calculation set obtained from the prediction model trained in step 2 is introduced into the genetic optimization BP neural network, and the determination coefficient is also used as the evaluation standard for the prediction effect to verify the accuracy of the model; Step 4: If the coefficient of determination calculated in step 3 meets the judgment criteria, the multi-element alloy composition and microalloying element array to be predicted are substituted into the particle swarm optimization BP neural network obtained in step 2 to complete the performance prediction; otherwise, repeat steps 2 and 3; The step 1 is specifically as follows: Step 1-1: Collecting multi-component alloy composition data and performance data associated with the composition and used to determine the alloy usage standards to form a basic database; Step 1-2: Divide the basic database into two parts: the particle swarm optimization BP neural network training set and the particle swarm optimization BP neural network validation set. Specifically: When the number of basic databases is between 200 and 400, the particle swarm optimization BP neural network training set is 95% of the basic database, and the particle swarm optimization BP neural network validation set is 5% of the basic database; when the number of basic databases is greater than 400, the particle swarm optimization BP neural network training set is 90% of the basic database, and the particle swarm optimization BP neural network validation set is 10% of the basic database; Step 1-3: Normalize the PSO BP neural network training set and the PSO BP neural network validation set between 0 and 1 to serve as the PSO BP neural network operation set; The step 2 is specifically as follows: Step 2-1: using the component data in the particle swarm optimization BP neural network operation set as input and the performance data as output, and setting the particle swarm optimization algorithm parameters, and using the particle swarm algorithm to optimize the initial weights and thresholds of the BP neural network. The particle swarm parameters include population size, maximum update generations, learning factor, and inertia weight method; Step 2-2: Establish a BP neural network, substitute the particle swarm optimized BP neural network operation set and the initial weights and thresholds after particle swarm optimization into the BP neural network, set the number of hidden layer nodes of the BP neural network, and establish a component-performance prediction model; The BP neural network model automatically divides the training set in the normalized particle swarm optimization BP neural network operation set into a BP model internal training set, a BP model internal validation set, and a BP model internal test set, and the sum of the proportions of each set is 1; The division ratio of the BP model internal training set, BP model internal validation set and BP model internal test set is determined by the number of particle swarm optimization BP neural network training sets. When the number of particle swarm optimization BP neural network training sets is between 200 and 400, the BP model internal training set accounts for 85% or 90%, the BP model internal validation set accounts for 10%, and the BP model internal test set accounts for 5% or 0; when the number of particle swarm optimization BP neural network training sets is more than 400, the BP model internal training set accounts for 70% to 90%, the BP model internal validation set accounts for 10% to 20%, and the BP model internal test set accounts for 10% to 20%; Step 2-3: Select the transfer function and training method to train the particle swarm optimization BP neural network, and select different determination coefficients as evaluation criteria according to the number of basic databases. When the number of basic databases is between 200 and 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤15%. When the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the BP neural network performance prediction value and the determination coefficient of the actual value in the particle swarm optimization BP neural network validation set and 1 is ≤10%. If it is within the evaluation standard range, proceed to step 3, otherwise recalculate. The step 3 is specifically as follows: Step 3-1: Using the component data in the operation set screened by the particle swarm optimization BP neural network training as the input of the genetic algorithm to optimize the BP neural network, and the performance data as the output of the genetic algorithm to optimize the BP neural network, and setting the genetic algorithm parameters, and using the genetic algorithm to optimize the initial weights and thresholds of the BP neural network. The genetic algorithm parameters include population size, iteration termination generation number, crossover and mutation probability; Step 3-2: Set the BP neural network parameters, select the transfer function and training method, and substitute the initial weights and thresholds optimized by the genetic algorithm into the BP neural network; Step 3-3: When the number of basic databases is less than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤20%; when the number of basic databases is greater than 400, the prediction effect is evaluated by the absolute value of the deviation between the coefficient of determination of the BP neural network performance prediction value and the actual performance value in the genetic algorithm optimized BP neural network verification set and 1 is ≤15%.

2. The multi-element alloy property prediction method based on particle swarm optimization BP neural network according to claim 1, characterized in that: The multi-component alloy includes: magnesium alloy, aluminum alloy, zirconium alloy, titanium alloy, copper alloy, high-temperature alloy and high-entropy alloy basic component system and added micro-alloying elements.

3. The multi-element alloy property prediction method based on particle swarm optimization BP neural network according to claim 1, characterized in that: The performance data associated with the multi-component alloy system components and used to determine the alloy usage standard include: yield strength, ultimate tensile strength, hardness, elongation, compressibility, electrical conductivity and corrosion rate.

Images