Laser radiation heat flow peak prediction method based on bayesian optimization neural network

Abstract

The application discloses a laser radiation heat flow peak prediction method based on a Bayesian optimization neural network. Firstly, the application utilizes pulsed laser output generated by a laser to act on a heat flow sensor to obtain sample data sets output by the heat flow sensor under different preset parameter conditions; the sample data sets are normalized and pretreated, and are randomly divided into a training data set and a verification data set; secondly, a Bayesian-BP neural network model for predicting a laser radiation heat flow peak is preliminarily established; the processed training data set is utilized to establish and train the Bayesian-BP neural network model; and then the verification data set is utilized to verify the prediction performance of the Bayesian-BP neural network model. The application adopts the Bayesian-BP neural network model to establish a black box relationship between a pulse width, laser power and a heat flow output peak, can ignore other interference terms, and realizes heat flow output peak prediction.

Description

Technical Field

[0001] This invention relates to the field of pulsed radiation heat flux research technology, and in particular to a method for predicting the peak value of laser radiation heat flux based on a Bayesian optimized BP neural network. Background Technology

[0002] Lasers are devices that can generate highly focused, high-intensity beams. To study the accuracy of heat flow sensor measurement results, they need to be calibrated. Therefore, lasers can be used to emit laser light to calibrate heat flow sensors.

[0003] However, when lasers are used improperly, the thermal effect of their instantaneous high-energy output can lead to problems such as damage to optical components and breakdown of the heat flux sensor measuring head. Therefore, it is necessary to develop an accurate and reliable method for predicting the peak heat flux of laser pulse radiation in order to solve the above-mentioned problems.

[0004] Using Ansys simulation software to build a simulation model can simulate the output of radiative heat flux from a heat flux sensor, but this method suffers from problems such as complex operation and incomplete consideration of factors affecting the prediction results. Machine learning is a feasible method for predicting the radiative heat flux of lasers. Wang Hanyu trained a GA_BP neural network model to predict the laser output temperature, solving the problem that calibrating a temperature sensor might damage it. However, the principle of a heat flux sensor differs from that of a temperature sensor. To achieve precise control over the radiative heat flux of a laser, a new machine learning model is needed to predict it. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention presents a method for predicting peak laser radiation heat flux based on a Bayesian optimized BP neural network.

[0006] In a first aspect, the present invention provides a method for predicting the peak radiation heat flux of a laser based on a Bayesian optimized BP neural network, specifically including the following steps:

[0007] S1, using the pulsed laser output generated by the laser to act on the measurement surface of the heat flow sensor, to obtain the sample dataset output by the heat flow sensor under different preset parameter conditions;

[0008] S2, normalizes the sample dataset before preprocessing and randomly divides it into training and validation datasets;

[0009] S3, A preliminary Bayesian BP neural network model for predicting the peak value of laser radiation heat flux was established.

[0010] S4. Use the training dataset processed in step S2 to build and train a Bayesian-BP neural network model.

[0011] S5 uses the validation dataset to validate the predictive performance of the Bayesian BP neural network model.

[0012] A second aspect of the present invention provides a laser radiation heat flux peak prediction device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a laser radiation heat flux peak prediction method.

[0013] A third aspect of the present invention provides a computer-readable storage medium storing a computer program for executing a method for predicting peak heat flux of a laser radiation.

[0014] The positive and beneficial effects of this invention are as follows:

[0015] 1. When studying the measurement performance of heat flux sensors using laser radiation heat flux, this invention can assist researchers in determining appropriate laser parameters before the experiment, thereby preventing irreversible damage to the experimental target material caused by laser radiation heat flux and meeting the safety requirements of physical experiments.

[0016] 2. A multi-input single-output Bayesian-BP neural network model is adopted to consider the influence of various factors on the peak value of laser radiation heat flux, thereby improving the accuracy and reliability of model prediction.

[0017] 3. By adopting a Bayesian-BP neural network model, a black-box relationship between pulse width and laser power and peak heat flux output was established. Other interference terms can be ignored, enabling the prediction of peak heat flux of laser radiation. This also provides a new technical solution for verifying the accuracy of heat flux sensor measurement results. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method implementation embodiment;

[0019] Figure 2 This is a laser thermal flux output testing platform.

[0020] Figure 3 The diagram shows the convergence process of training the mean squared error of the network model.

[0021] Figure 4 Figure 1 shows an embodiment of a laser output heat flux peak prediction device. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention is described below with reference to specific embodiments shown in the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0023] This application discloses a method for predicting the peak radiation heat flux of a laser based on a Bayesian-BP neural network algorithm, specifically including the following steps, such as... Figure 1 As shown:

[0024] S1, Build a laser heat flow output experimental platform to obtain sample datasets output by the heat flow sensor under different preset parameter conditions;

[0025] S1.1, such as Figure 2 As shown, the laser heat flow output experimental platform mainly consists of a pulse trigger 3 controlling the laser 2 to emit pulsed laser light, which radiates onto the measurement surface of the heat flow sensor 1. Simultaneously, a signal acquisition card 6 collects the voltage signal of the heat sensor, and a constant-temperature water cooler 4 provides constant-temperature water cooling for the heat flow sensor. In this embodiment, the high-power laser is the excitation heat source, and the laser radiation heat flow is mainly determined by two parameters: the laser pulse width t. width The laser power P can be adjusted by setting the repetition frequency f and duty cycle k, while the laser power P can be controlled by adjusting the power output percentage.

[0026] S1.2, Set the output parameters of the high-power laser and output the radiated laser to act on the measurement surface of the heat flow sensor. Record the laser repetition frequency f, duty cycle k, power P, radiated laser absorptivity λ of the heat flow sensor, and peak heat flow Q of the heat flow sensor output. Obtain the input vector X = (x1, x2, x3, x4) = (P, f, k, λ) and output layer vector Y = (Q) of the Bayesian-BP neural network, and obtain a set of sample data. The radiated laser is a one-dimensional planar shock wave. In this embodiment, several typical Gordon heat flow sensors have different laser absorption effects on the black coating of their measurement surfaces.

[0027] S1.3 Repeat the above steps multiple times, applying laser pulses with different repetition frequencies, duty cycles, and powers to the measurement surface of the heat flow sensor with different laser absorption rates to obtain several sets of sample data, thereby obtaining a sample dataset.

[0028] S2, normalizes the sample dataset before preprocessing and randomly divides it into a 75% training set and a 30% validation set:

[0029] S2.1, For each input layer vector and each sample element in each output layer vector of the sample dataset, apply the formula (based on the minimax method) y k =(yy) min ) / (y max -y min ) are mapped sequentially to the interval [0,1]. Where y k y represents the normalized value of the experimental data. min y is the minimum value in the sample dataset.max This represents the maximum value in the sample dataset; in one example, the max-min method is implemented by calling the mapminmax function in the MATLAB toolbox.

[0030] S2.2, randomly divide it into a 75% training set and a 30% validation set.

[0031] S3, Preliminary establishment of a Bayesian-BP neural network architecture for predicting the peak value of laser radiative heat flux:

[0032] S3.1, based on empirical formula Determine the number of hidden layer nodes m to improve the prediction accuracy of the BP neural network, where a is the number of input layer vectors, b is the number of output layer vectors, and c is a constant from 1 to 10.

[0033] S3.2, the specific implementation formula of the initially established Bayesian-BP neural network model is as follows:

[0034]

[0035] Where w k m represents the connection weights between the hidden layer and the output layer. k denoted as , where is the computation process between the input layer and the hidden layer, 'a' is the bias between the hidden layer and the output layer, 'j' is the optimal number of hidden layers, 'δ1' is the activation function to be determined, and 'Y' is the predicted peak value of the laser radiation heat flux.

[0036]

[0037] Where w ik x represents the connection weights between the input layer and the hidden layer. i denoted as the value after preprocessing the i-th type of input parameter, b is the bias between the input layer and the hidden layer, k is the optimal number of hidden layers, and δ2 is the activation function to be determined.

[0038] S4. Use the data processed in step S2 to build and train a Bayesian-BP neural network model.

[0039] S4.1, Define the hyperparameters mentioned in step S4 (number of hidden layer nodes k, weights w of the BP neural network). k ,w ik Search the activation functions δ1, δ2 and the bias terms a, b) in the search space and initialize them;

[0040] S4.2, select the mean square error (MSE) between the actual value and the predicted value as the objective function. The smaller the MSE, the higher the accuracy of the predicted data.

[0041] n is the number of samples.

[0042] S4.3, Define the maximum number of iterations for model training;

[0043] S4,4. Using the normalized training set data obtained in step S2, train the Bayesian-BP neural network model until the maximum number of iterations is reached, and the mean square error prediction convergence curve of the prediction model can be obtained.

[0044] S5. Use the validation set data from step S2 to predict and validate the Bayesian-BP neural network model.

[0045] The following examples illustrate in detail the specific implementation of the laser radiation heat flux peak prediction method based on Bayesian-BP neural network of the present invention.

[0046] S1. Multiple sets of experiments were conducted on the laser heat flow output test platform to generate sample datasets. During the experiments, pulsed laser outputs with different laser repetition frequencies f, duty cycles k, and power P were applied to a heat flow sensor with a radiation laser absorptivity of λ, generating a total of 200 sets of sample data.

[0047] The laser repetition frequency f, duty cycle k, power P, and the laser absorptivity λ of the heat flux sensor are selected as the input layer vector X = (x1, x2, x3, x4) = (P, f, k, λ). A signal acquisition card with a sampling rate of 1000 Sa / s is used to acquire the heat flux output signal of the heat flux sensor from the rising edge of the pulse to the falling edge of the pulse, and the peak heat flux Q is selected as the output layer vector Y = (Q).

[0048] The laser pulse is a one-dimensional planar pulse and exhibits an ideal Gaussian distribution.

[0049] In this way, we obtain 200 sets of sample data, distributed as X1=(f1,k1,P1,λ1),Y1=(Q1);

[0050] X2=(f2,k2,P2,λ2),Y2=(Q2);

[0051] …

[0052] X 200 =(f 200 k 200 P 200 , λ 200 ),Y 200 =(Q 200 ).

[0053] S2, using the formula y k =(yy) min ) / (y max -y minThe sample dataset was preprocessed and standardized for scaling, and then randomly divided into a 75% training set and a 30% validation set.

[0054] S2.1, Suppose we let Y be f1, f2, ... f n Given a dataset, the maximum value is selected as y. max The minimum value is taken as y min Then let y be f1, f2, ... f n Calculate the corresponding sample feature values ​​y respectively. k Similarly, the sample feature values ​​of each sample element in each input layer vector and each output layer vector in the sample dataset can be obtained.

[0055] S2.2, divide the 200 data points into 4 groups of 50 data points each, with three groups serving as the training set and one group serving as the validation set.

[0056] S3. The architecture of a Bayesian-BP neural network model for predicting the peak value of laser radiation heat flux is initially established, and the hyperparameters that need to be optimized are obtained.

[0057] S4, using the training dataset processed in step S2, build and train a Bayesian BP neural network model:

[0058] S4.1, Define the search space for the Bayesian-BP neural network hyperparameters obtained in S3 as follows:

[0059]

[0060] S4.2, Select the mean square error (MSE) between the actual value and the predicted value as the objective function;

[0061] S4.3, Define the maximum number of iterations for training the model as 1500;

[0062] S4.4, using the normalized training set data obtained in step S2, train the Bayesian-BP neural network model until the maximum number of iterations is reached, and obtain the mean squared error prediction convergence curve of the prediction model as shown in the figure. Figure 3 As shown. The optimal hyperparameter combination after optimization is: ReLU activation function, 12 hidden layer nodes, BP neural network weights of -1.981 and -0.464, and BP neural network biases of -0.452 and -0.282. At this time, the mean squared error (MSE) of the objective function is minimized, with a minimum MSE value of 0.00964.

[0063] S5. Finally, the optimal Bayesian BP neural network model for predicting the peak value of laser radiation heat flux is validated, with the input layer vector set to X = (f, k, P, λ). iThe vector X is given by the expression (10, 4%, 15, 95%), where the first value is in 10 MHz and the third value is in W. When this vector X is input into a Bayesian backpropagation (BP) neural network model, the predicted output vector is Y = (0.0356), in kW / m². 2 This means obtaining the peak value of the laser radiation heat flux under the specified parameters.

[0064] like Figure 4 As shown in the illustration, this invention also discloses a laser radiation heat flux peak prediction device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a laser radiation heat flux peak prediction method. The memory may include main memory, such as high-speed random access memory, or it may also include non-volatile memory, such as at least one disk storage device. The processor, network interface, and memory are interconnected via an internal bus, which may be an industry-standard architecture bus, a peripheral component interconnection standard bus, an extended industry-standard architecture bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory is used to store the program; specifically, the program may include program code, which includes computer operation instructions. The memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0065] This invention also discloses a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of a laser radiation heat flux peak prediction method. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for predicting peak laser radiative heat flux based on Bayesian optimized neural networks, characterized in that: Includes the following steps: S1, using the pulsed laser output generated by the laser to act on the side surface of the heat flow sensor, to obtain the sample dataset output by the heat flow sensor under different preset parameter conditions; S2, normalizes the sample dataset before preprocessing and randomly divides it into training and validation datasets; S3, A preliminary Bayesian-BP neural network model for predicting the peak value of laser radiation heat flux was established; S4. Use the training dataset processed in step S2 to build and train a Bayesian-BP neural network model. S5. Validate the predictive performance of the Bayesian BP neural network model using the validation dataset. The preset parameters mentioned in step S1 include the repetition frequency, duty cycle, and power of the laser; wherein the repetition frequency and duty cycle determine the laser pulse width; Step S1 is as follows: a. Apply a pulsed laser with a specific power density and pulse width to the measurement surface of the heat flux sensor, and record the peak value of the radiative heat flux of the heat flux sensor. The input layer vector of the Bayesian-BP neural network model is obtained. Output layer vector ,in For repetition frequency, For duty cycle, For power, The absorption rate of the radiated laser for the heat flow sensor; b. Repeat the above steps multiple times to apply heat flow radiation with different laser powers and pulse widths to the measurement surfaces of different heat flow sensors, thereby obtaining a sample dataset.

2. The method for predicting the peak heat flux of laser radiation according to claim 1, characterized in that: The heat flow sensor mentioned in step S1 is a Gordon heat flow sensor.

3. The method for predicting the peak heat flux of laser radiation according to claim 1, characterized in that: The heat flow radiation mentioned is a one-dimensional planar heat flow.

4. The method for predicting the peak heat flux of laser radiation according to claim 1, characterized in that: The mathematical expression of the Bayesian-BP neural network model in step 3 is as follows: in The connection weights between the hidden layer and the output layer. This refers to the computation process between the input layer and the hidden layer. The bias between the hidden layer and the output layer. This represents the optimal number of hidden layers. As the first activation function, Predict the peak value of the laser's radiative heat flux.

5. The method for predicting the peak heat flux of laser radiation according to claim 4, characterized in that: The calculation process is as follows: The connection weights between the input layer and the hidden layer. For the first The values ​​after preprocessing the input parameters. The bias between the input layer and the hidden layer. This represents the optimal number of hidden layers. This is the second activation function.

6. A device for predicting the peak radiation heat flux of a laser, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the laser radiation heat flux peak prediction method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing a laser radiation heat flux peak prediction method according to any one of claims 1-5.

Images