Photovoltaic frame bearing performance prediction method

Abstract

The invention discloses a photovoltaic frame bearing performance prediction method, and the method specifically comprises the following steps: collecting the characteristic parameters of a photovoltaic frame to be detected, inputting the characteristic parameters into a photovoltaic frame bearing performance prediction model, and outputting the photovoltaic frame bearing performance corresponding to the characteristic parameters through the photovoltaic frame bearing performance prediction model. On the basis of a small amount of experimental data, a rapid prediction method of'photovoltaic frame structure design parameter-bearing performance 'is obtained by combining finite element simulation and small sample machine learning technologies, and is used for predicting the strength performance of the photovoltaic frame under different structure parameter combinations, so that technical reference is provided for practical engineering application.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic assembly frames, and in particular to a method for predicting the bearing performance of a photovoltaic frame. Background Art

[0002] Photovoltaic modules work in complex environments such as high temperature, strong winds, and even hail for a long time. In order to ensure the safety of the modules, guarantee power generation efficiency, and extend service life, the frame must have good load-bearing performance. Generally, the factors that affect the load-bearing capacity of photovoltaic frames mainly include: material selection (high-strength aluminum alloy, reinforced composite materials, etc.), structural design (cross-sectional shape, setting of reinforcement ribs, etc.) and manufacturing process (heat treatment and surface treatment). Among them, the selection of frame material and structural design are the most critical.

[0003] In actual production, the single factor rotation method (or control variable method) is usually used to conduct a large number of experiments to study the influence of the above-mentioned different structural parameters on the overall frame bearing performance. However, due to the interaction of these structural parameters, they show a highly nonlinear law, so it is a multidimensional nonlinear mapping problem. The use of traditional experimental analysis methods will be very time-consuming, labor-intensive and costly. Therefore, if we can build a mathematical model between "different structural design parameters-frame bearing performance" with the help of new technical means under the condition of as few experiments as possible, and use the model to realize the rapid prediction and analysis of the frame bearing performance under different structural designs, it will be of great significance to improve the service life of photovoltaic modules and ensure power generation efficiency.

[0004] For structural optimization design problems, researchers often use finite element simulation calculation methods. Compared with physical experiments, finite element methods have obvious advantages in research cost and cycle, and the accuracy and repeatability of analysis results are good. However, for actual enterprise applications, it seems somewhat impractical because completing a complete process of structural design simulation requires professional computing resources and a long computing time, and the technical difficulty is relatively high. Summary of the invention

[0005] The present invention provides a method for predicting the load-bearing performance of a photovoltaic frame, aiming to improve at least one of the above-mentioned problems.

[0006] The present invention is implemented as follows: a method for predicting the load-bearing performance of a photovoltaic frame, the method being specifically as follows:

[0007] The characteristic parameters of the photovoltaic frame to be tested are collected, and the characteristic parameters are input into the photovoltaic frame load-bearing performance prediction model, and the photovoltaic frame load-bearing performance prediction model outputs the photovoltaic frame load-bearing performance corresponding to the characteristic parameters.

[0008] Furthermore, the BP neural network model is trained based on the samples, and after the test accuracy of the BP neural network model reaches a set accuracy threshold, the BP neural network model is used as a photovoltaic frame load-bearing performance prediction model.

[0009] Furthermore, the fireworks algorithm is used to optimize the initial weights and thresholds of the neural network in the BP neural network model.

[0010] Furthermore, the samples are divided into training samples and test samples, the BP neural network model is trained based on the training samples, and the BP neural network model in the training process is tested based on the test samples.

[0011] Furthermore, the ratio of test samples to training samples is 8:2.

[0012] Furthermore, the sample construction process is as follows:

[0013] A finite element simulation model of the photovoltaic frame is constructed, and the load-bearing performance of the photovoltaic frame under different characteristic parameters is simulated through the finite element simulation model of the photovoltaic frame. The current characteristic input parameters of the photovoltaic frame are used as features, and the load-bearing performance of the photovoltaic frame obtained by simulation is used as a label.

[0014] Furthermore, the characteristics of the sample include: lining material modulus E, lining width D, lining height H, lining thickness d, lining support ribs, and the sample is labeled: maximum load bearing F of the photovoltaic frame.

[0015] Furthermore, the ABAQUS-2021 version software was used to construct a finite element simulation model of the photovoltaic frame.

[0016] Furthermore, the BP network topology consists of three layers, namely, input layer, hidden layer and output layer. The number of input layer nodes is 4, the number of output layer nodes is 1, and the number of hidden layer nodes ranges from [5-12].

[0017] Based on a small amount of experimental data, the present invention designs a method combining finite element simulation and small sample machine learning technology to obtain a rapid prediction method of "photovoltaic frame structure design parameters-bearing performance", which is used to predict the strength performance of photovoltaic frames under different structural parameter combinations, thereby providing technical reference for practical engineering applications. At the same time, in order to overcome the shortcomings of the existing ordinary BP neural network technology, which has poor prediction accuracy for small sample data and is prone to local minimization, the improved fireworks algorithm IFWA is introduced to improve and optimize the BP network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 A flow chart of the overall technical route of the method for predicting the load-bearing performance of a volt frame provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a photovoltaic frame lining structure provided by an embodiment of the present invention;

[0021] Figure 3 A finite element simulation model diagram of a photovoltaic frame lining-compression test provided in an embodiment of the present invention;

[0022] Figure 4 A diagram showing the prediction results of the IFWA-BP model for 22 groups of training samples in a stress resistance experiment provided by an embodiment of the present invention;

[0023] Figure 5 A comparison chart of test sample predictions of the BP model and the IFWA-BP model provided in an embodiment of the present invention;

[0024] Figure 6 A comparison chart of the prediction errors of the compression test-BP model and the IFWA-BP model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The specific implementation modes of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0026] Based on a small amount of experimental data, the present invention designs a method combining finite element simulation and small sample machine learning technology to obtain a rapid prediction method of "photovoltaic frame structure design parameters-bearing performance", which is used to predict the strength performance of photovoltaic frames under different structural parameter combinations, thereby providing technical reference for practical engineering applications. At the same time, in order to overcome the shortcomings of the existing ordinary BP neural network technology, which has poor prediction accuracy for small sample data and is prone to local minimization, the improved fireworks algorithm IFWA is introduced to improve and optimize the BP network model.

[0027] The photovoltaic frame load-bearing performance prediction method provided by the present invention is specifically as follows: characteristic parameters of the photovoltaic frame to be tested are collected, the characteristic parameters are input into a photovoltaic frame load-bearing performance prediction model, and the photovoltaic frame load-bearing performance prediction model outputs the photovoltaic frame load-bearing performance corresponding to the characteristic parameters.

[0028] The BP neural network model is trained based on samples. After the test accuracy of the BP neural network model reaches the set accuracy threshold, the BP neural network model is used as the photovoltaic frame load-bearing performance prediction model. The BP network topology structure consists of three layers, namely the input layer, hidden layer and output layer. The number of input layer nodes is 4, the number of output layer nodes is 1, and the number of hidden layer nodes ranges from [5-12].

[0029] In the embodiment of the present invention, the sample construction process is specifically as follows:

[0030] Construct a finite element simulation model of the photovoltaic frame, and simulate the load-bearing performance of the photovoltaic frame under different characteristic parameters through the finite element simulation model of the photovoltaic frame. Use the current characteristic input parameters of the photovoltaic frame as features, and use the load-bearing performance of the photovoltaic frame obtained by simulation as labels. Divide the samples into training samples and test samples, train the BP neural network model based on the training samples, and test the BP neural network model based on the training process of the test samples. The ratio of test samples to training samples is 8:2.

[0031] In the embodiment of the present invention, the sample characteristics include: lining material modulus E (unit: MPa), lining width D (unit: mm), lining height H (unit: mm), lining thickness d (unit: mm), lining support ribs (horizontal, vertical), and the sample label is the maximum load F (unit: kN).

[0032] Specifically, the present invention constructs a sample based on steps (1) to (3), and constructs a photovoltaic frame load-bearing performance prediction model based on steps (4) and (5), such as Figure 1 shown.

[0033] (1) Obtain experimental data of photovoltaic frame samples. Take the load-bearing performance (compressive strength) of photovoltaic frame samples as the research object, and process experimental samples (such as Figure 2 As shown), carry out mechanical experiments and obtain a small amount of experimental data;

[0034] (2) Construct a finite element simulation model of the photovoltaic frame. Based on the experimental data, with the help of finite element simulation software, set the material model, geometric model and related boundary conditions to construct a finite element simulation model of the photovoltaic frame mechanical experiment, such as Figure 3 As shown, the finite element simulation software is the commercial ABAQUS-2021 version software.

[0035] (3) Design a simulation experiment plan. In combination with actual engineering requirements, select the key structural parameters of the photovoltaic frame as characteristic parameters, design different characteristic parameter combinations, and use the photovoltaic frame mechanical experiment finite element simulation model to simulate the maximum load under different characteristic parameters. The current characteristic input parameters of the photovoltaic frame are used as features, and the load-bearing performance of the photovoltaic frame obtained by simulation is used as a label to obtain data samples for machine learning;

[0036] (4) Training the BP network model. The fireworks algorithm is introduced into the BP neural network model to optimize the initial weights and thresholds of the neural network to achieve the most effective learning and prediction of the neural network; a BP network model of the training structure design parameters - photovoltaic frame load-bearing performance is constructed by randomly selecting training samples from the constructed data samples;

[0037] (5) Testing the BP network model. The constructed BP network model is tested by randomly selecting test samples from the data sample. After the BP network model reaches the set accuracy threshold, the trained BP network model is used as the photovoltaic frame load-bearing performance prediction model.

[0038] The fireworks algorithm is used to optimize the initial weights and thresholds of the BP neural network, so that the optimized BP neural network can better predict the function output. And the prediction error is compared with that of the ordinary BP model to verify the superior performance of the BP network model based on the fireworks algorithm; the parameters of the fireworks algorithm are as follows: fireworks population size N = 10, fireworks explosion radius D = 20, fireworks explosion spark number M = 5, fireworks explosion spark number upper limit Lm = 0.6, fireworks explosion spark number lower limit Bm = 0.3, Gaussian variation spark number G = 5, maximum number of iterations T = 100.

[0039] When verifying the BP neural network model, samples that have not participated in the BP neural network model training are selected from the samples, and a prediction test is performed on the constructed network model. The generalization ability of the improved model is evaluated by comparing the relative errors between the measured and predicted values.

[0040] The present invention aims at predicting the influence of different structural design parameters on the load-bearing performance of photovoltaic frames. Based on a very small amount of experimental results, a finite element simulation model is constructed to obtain data samples for machine learning through simulation. Then, with the BP neural network model as the center, the improved fireworks algorithm IFWA is combined to optimize the initial weights and thresholds, and finally a new IFWA-BP network model with high prediction accuracy and the ability to quickly predict the load-bearing performance of products is obtained. It not only solves the problem of high R&D cost of traditional experimental methods and large computational complexity of finite element simulation technology, but also overcomes the problems of overfitting, slow convergence speed and poor generalization ability commonly found in ordinary BP network methods, providing a new technical means for the design and development of photovoltaic frame products.

[0041] Based on the finite element simulation model (such as Figure 3 The data samples are obtained by using the compressive strength data of the photovoltaic frame under different structural design parameters as training samples, as shown in Table 1. The input variables are the lining material modulus E (X1), lining width D (X2), lining height H (X3), lining thickness d (X4), lining support ribs (X5), and the output variable is the maximum load F (unit: kN) of the sample.

[0042] Table 1 Characteristic parameters and value distribution table

[0043]

[0044]

[0045] Table 2 lists the 27 groups of data samples actually measured, 22 of which are randomly selected as training samples, and the remaining 5 groups of data are taken as test samples for the model.

[0046] Table 2 Data sample example table

[0047]

[0048]

[0049] The BP neural network model was constructed by programming in MATLAB R2021a software. In order to further improve the prediction accuracy of the ordinary BP network model, the improved fireworks algorithm was used to optimize the initial weights and thresholds of the BP neural network (abbreviated as IFWA-BP). By setting: fireworks population size N = 10, fireworks explosion radius D = 20, number of fireworks explosion sparks M = 5, upper limit of the number of fireworks explosion sparks Lm = 0.6, lower limit of the number of fireworks explosion sparks Bm = 0.3, number of Gaussian variation sparks G = 5, maximum number of iterations T = 100, a new IFWA-BP prediction model for photovoltaic frame bearing performance was obtained. From Figure 4 Judging from the prediction results of the model for 22 groups of training samples, the predicted values ​​are consistent with the true values, indicating that the model has achieved a high prediction accuracy.

[0050] Then, 5 groups of test samples were input into the established IFWA-BP network model for simulation prediction and comparison. Figure 5 and Figure 6 As shown in the figure, after optimization by the IFWA algorithm, the maximum error of the new IFWA-BP network model has been further reduced to 3.5%, which is significantly higher than the 13.1% before optimization. This shows that the new IFWA-BP model has achieved a high prediction accuracy and can be used for further performance prediction research.

[0051] The present invention has been described exemplarily. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A method for predicting the load-bearing performance of a photovoltaic frame, characterized in that: The method is specifically as follows: The characteristic parameters of the photovoltaic frame to be tested are collected, and the characteristic parameters are input into the photovoltaic frame load-bearing performance prediction model, and the photovoltaic frame load-bearing performance prediction model outputs the photovoltaic frame load-bearing performance corresponding to the characteristic parameters.

2. The photovoltaic frame load-bearing performance prediction method according to claim 1, characterized in that: The BP neural network model is trained based on samples. After the test accuracy of the BP neural network model reaches the set accuracy threshold, the BP neural network model is used as a prediction model for the load-bearing performance of the photovoltaic frame.

3. The photovoltaic frame load-bearing performance prediction method according to claim 2, characterized in that: The fireworks algorithm is used to optimize the initial weights and thresholds of the neural network in the BP neural network model.

4. The photovoltaic frame load-bearing performance prediction method according to claim 2, characterized in that: The samples are divided into training samples and test samples. The BP neural network model is trained based on the training samples, and the BP neural network model in the training process is tested based on the test samples.

5. The photovoltaic frame load-bearing performance prediction method according to claim 4, characterized in that: The ratio of test samples to training samples is 8:

2.

6. The photovoltaic frame load-bearing performance prediction method according to claim 1, characterized in that: The sample construction process is as follows: A finite element simulation model of the photovoltaic frame is constructed, and the load-bearing performance of the photovoltaic frame under different characteristic parameters is simulated through the finite element simulation model of the photovoltaic frame. The current characteristic input parameters of the photovoltaic frame are used as features, and the load-bearing performance of the photovoltaic frame obtained by simulation is used as a label.

7. The photovoltaic frame load-bearing performance prediction method according to claim 6, characterized in that: The characteristics of the sample include: lining material modulus E, lining width D, lining height H, lining thickness d, lining support ribs, and the sample is labeled: maximum load bearing F of the photovoltaic frame.

8. The photovoltaic frame load-bearing performance prediction method according to claim 6, characterized in that: ABAQUS-2021 software is used to construct a finite element simulation model of the photovoltaic frame.

9. The photovoltaic frame load-bearing performance prediction method according to claim 1, characterized in that: The BP network topology consists of three layers, namely the input layer, hidden layer and output layer. The number of input layer nodes is 4, the number of output layer nodes is 1, and the number of hidden layer nodes ranges from [5-12].

Images