Underground engineering damage prediction method and device based on adaptive fuzzy neural network, and storage medium

Abstract

The invention provides an underground engineering damage prediction method and device based on an adaptive fuzzy neural network, and a storage medium, and belongs to the technical field of damage evaluation. Comprising the following steps: carrying out scene modeling on an underground engineering and a warhead, and simulating the damage of the warhead to the underground engineering to obtain shock wave damage factors; a shock wave overpressure prediction value is obtained by using a shock wave damage prediction model constructed based on a self-adaptive fuzzy neural network model; and comparing the shock wave overpressure prediction value with the damage grade criterion of the key component in the underground engineering to obtain a damage prediction result of the underground engineering attacked by the warhead. According to the method, the adaptive fuzzy neural network is utilized to learn the mapping relation between the shock wave damage factor represented by the fuzzy rule and the shock wave overpressure in the underground engineering from the sample data set, and the shock wave overpressure in the underground engineering is further compared with the damage grade criterion of the key component; therefore, the impact wave damage effect of the warhead hitting the underground engineering is predicted.

Description

Technical Field

[0001] The present invention relates to a method, device and storage medium for predicting damage to underground engineering based on an adaptive fuzzy neural network, and belongs to the technical field of damage assessment. Background Technique

[0002] Underground engineering refers to underground protective buildings constructed separately to protect personnel and materials during wartime, civil air defense command, and medical rescue. It is an important material foundation for national security and an important part of the military threat force. It is an important facility for guarding against sudden enemy attacks and preserving war potential; it is an engineering guarantee for persevering in urban battles and supporting the war of resistance against aggression for a long time until victory.

[0003] With the continuous advancement of the urbanization process and the rapid development of the construction industry, underground engineering plays an increasingly important role in urban planning and construction, which is related to the safety and stability of the city and the lives and property of the general public.

[0004] Since the 1960s, the rapid development of weapons has forced the development of underground engineering protection systems, especially the improvement of the accuracy of new guided weapons and the emergence of weapons such as earth-penetrating bombs. Precision-guided weapons and earth-penetrating bombs are characterized by accurate hitting, deep penetration, and great explosion hazards. The main reason for their serious threat to underground engineering is that they can rely on kinetic energy and mass to penetrate the protection system of underground engineering and explode after penetrating a certain depth. Compared with ordinary missiles that explode when they touch the ground, the energy coupled into the ground after penetration and explosion will increase exponentially, and it will cause damage such as non-linear creep of the rock at the underground entrance. The shock wave overpressure caused after the explosion may also exceed the overpressure criterion threshold according to the actual situation, causing serious damage to the communication and support systems inside the underground engineering.

[0005] The traditional means of predicting damage to underground engineering is to summarize the penetration damage empirical formula from a large amount of experimental data. However, whether it is the penetration formula or the penetration damage prediction model of the adaptive fuzzy neural network, the final prediction result is only the penetration depth. This depth can only reflect whether the dynamic load section of the underground engineering is penetrated and whether the protection system is damaged, and cannot directly reflect the actual damage degree suffered by other systems inside the underground engineering. Therefore, the research on the damage caused by the shock wave after penetration and explosion has become an important way to improve the protection ability of underground engineering. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for predicting damage to underground engineering based on an adaptive fuzzy neural network, which solves the problem that in war simulation, the actual damage degree suffered by other systems inside the underground engineering cannot be directly reflected.

[0007] To achieve the above object, the present invention is implemented by the following technical solution:

[0008] In a first aspect, the present invention provides a method for predicting damage to underground engineering based on an adaptive fuzzy neural network, including: performing scenario modeling on the underground engineering and the warhead, and simulating the damage of the warhead to the underground engineering under the established scenario to obtain shock wave damage factors; the shock wave damage factors include: the warhead charge equivalent, the warhead landing speed, and the distance between the key components and the warhead explosion point;

[0009] Obtaining a predicted value of shock wave overpressure by using a pre-constructed shock wave damage prediction model; wherein, the shock wave damage prediction model is constructed by an adaptive fuzzy neural network model;

[0010] Comparing the predicted value of shock wave overpressure with the damage level criterion of key components in the underground engineering to obtain the damage prediction result of the warhead hitting the underground engineering.

[0011] Furthermore, the shock wave damage prediction model includes: a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer;

[0012] The fuzzification layer is used to calculate the membership degree of each shock wave damage factor;

[0013] The rule layer is used to calculate the triggering intensity of each fuzzy rule according to the membership degree of each shock wave damage factor;

[0014] The normalization layer is used to normalize the triggering intensity of each fuzzy rule to obtain the triggering weight of each fuzzy rule in the entire rule base;

[0015] The defuzzification layer is used to calculate the output value of each fuzzy rule according to the triggering weight of each fuzzy rule;

[0016] The summation layer is used to perform weighted averaging on the output values of all fuzzy rules according to the triggering weight to obtain the predicted value of shock wave overpressure.

[0017] Furthermore, the calculation of the membership degree of each shock wave damage factor is specifically: ; In the formula, is the membership degree of the fuzzy subset , is the membership function of the fuzzy subset , is the warhead charge equivalent, is the membership degree of the fuzzy subset , is the membership function of the fuzzy subset , is the warhead landing speed, is the membership degree of the fuzzy subset Membership degree, is a fuzzy subset Membership function of, is the distance of the key component from the explosion point of the warhead, is the serial number of the fuzzy subset.

[0018] Furthermore, the membership function adopts a Gaussian membership function, specifically: ; In the formula, is the forward parameter.

[0019] Furthermore, the defuzzification layer calculates the output value of each fuzzy rule according to the triggering weight of each fuzzy rule, specifically: ; In the formula, , is the output value of the defuzzification layer, is the th triggering weight of the fuzzy rule, is the th output value of the fuzzy rule, is the backward parameter of the fuzzy rule.

[0020] Furthermore, when training the adaptive fuzzy neural network model, the L-BFGS method is used to optimize the forward parameter, specifically:

[0021] Forwardly transmit the data of the shock wave damage factor to the summation layer to obtain the predicted value of the shock wave overpressure;

[0022] Taking the mean square error as the loss function, calculate the error between the predicted shock wave overpressure and the real shock wave overpressure: ; In the formula, is the real value of the shock wave overpressure, is the predicted value of the shock wave overpressure, is the total number of samples, is the loss function, is the forward parameter;

[0023] Store the gradient information of the last iterations, calculate the approximation of the inverse of the low-storage Hessian matrix, and use it to update the forward parameter: ; In the formula, is the first-order partial derivative vector of the loss function with respect to , is the low-storage Hessian matrix.

[0024] Further, when training the adaptive fuzzy neural network model, the regularized least squares method is used to optimize the backward parameters. Specifically: the data of the shock wave damage factors is operated along the adaptive fuzzy neural network model to the defuzzification layer, the forward parameters are fixed, and the regularized least squares method is used to optimize the backward parameters. Specifically:

[0025] Write the shock wave overpressure prediction output in matrix form: ; In the formula, is the input matrix composed of the trigger weights and the shock wave damage factors, is the backward parameter matrix;

[0026] Establish an optimization objective function: ; In the formula, is the regularization coefficient, is the L2 regularization term;

[0027] Solve the optimization objective function to obtain the optimized backward parameter matrix: ; In the formula, is the identity matrix.

[0028] Further, it also includes: obtaining penetration damage factors: the complete mass of the warhead, the diameter of the warhead body, and the landing speed of the warhead; based on the penetration damage factors, obtaining the penetration depth of the warhead hitting the underground project through a pre-constructed penetration damage prediction model, and judging whether the underground project protection system is damaged according to the penetration depth; the penetration damage prediction model is constructed by an adaptive fuzzy neural network model.

[0029] In a second aspect, the present invention provides an electronic device, including: a processor and a memory;

[0030] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the underground project damage prediction method based on the adaptive fuzzy neural network as described in the first aspect is realized.

[0031] In a third aspect, the present invention provides a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the underground project damage prediction method based on the adaptive fuzzy neural network as described in the first aspect is realized.

[0032] Compared with the prior art, the beneficial effects achieved by the present invention:

[0033] (1) The method provided by the present invention uses an adaptive fuzzy neural network to learn from the sample dataset the mapping relationship between the shock wave damage factors represented by fuzzy rules and the shock wave overpressure inside the underground project, and further compares the shock wave overpressure inside the underground project with the damage level criteria of key components, so as to realize the prediction of the shock wave damage effect of the warhead hitting the underground project; this analysis method through target vulnerability analysis and damage effectiveness evaluation uses the degree of damage suffered by important systems such as the protection system, communication system, and support system of the underground project after being hit by the warhead to reflect the damage degree caused by precision-guided weapons and penetrator bombs and other weapons.

[0034] (2) The method provided by the present invention adopts an adaptive fuzzy neural network system with a fast convergence speed, which can accurately predict the penetration damage and shock wave damage effects that may occur when the underground project encounters an attack. It provides a direction and verification for the upgrading and transformation of the underground project and increasing its protection ability. In addition, by studying the damage degree of the underground entrance, it can provide data and technical support for the rapid emergency repair and reconstruction of the project, which has important theoretical significance and military value. Brief Description of the Drawings

[0035] Figure 1 It is a schematic flow chart of the underground project damage prediction method in Embodiment 1 of the present invention;

[0036] Figure 2 It is a network structure diagram of the shock wave damage prediction model in Embodiment 1 of the present invention;

[0037] Figure 3 It is a network structure diagram of the penetration damage prediction model in Embodiment 1 of the present invention;

[0038] Figure 4 It is an error analysis diagram taking the penetration depth as the error analysis index in Embodiment 3 of the present invention. Detailed Description of the Embodiment

[0039] The terms "including" and "having" in the specification, claims and above-mentioned drawings of this application, as well as any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments.

[0040] Embodiment 1

[0041] This embodiment provides a method for predicting damage to underground engineering based on an adaptive fuzzy neural network, as Figure 1 shown. This method includes the following steps:

[0042] Step 1: Perform scene modeling on the underground engineering and the warhead, and simulate the damage of the warhead to the underground engineering under the established scene to obtain shock wave damage factors.

[0043] Specifically, the scene modeling includes: modeling of building structures and facilities and equipment. First, import the CAD drawings of the building, draw facilities and equipment such as doors, windows, stairs, electrical and communication equipment on the drawings, then use modeling software to model the building structures and facilities and equipment of the underground engineering, and finally export the building model.

[0044] According to experience, the factors affecting shock wave damage include: the complete mass of the warhead, the actual penetration depth of the warhead, the landing speed of the warhead, the thickness of the protection section of the underground engineering, and the distance from the key components to the explosion point of the warhead. Among them, the thickness of the protection section of the underground engineering is set in advance in actual engineering projects, and it is not independent of the complete mass of the warhead and the actual penetration depth of the warhead. In order to simplify the input variables of the ANFIS network and reduce the error of shock wave damage prediction, the charge equivalent of the warhead after penetration is used to replace the above three factors.

[0045] Therefore, in the technical solution of the present invention, the shock wave damage factors include: the charge equivalent of the warhead, the landing speed of the warhead, and the distance from the key components to the explosion point of the warhead.

[0046] Step 2: Use the pre-constructed shock wave damage prediction model to obtain the predicted value of shock wave overpressure, where the shock wave damage prediction model is constructed by an adaptive fuzzy neural network model.

[0047] As is well known, the adaptive fuzzy neural network model includes: an input layer, a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer; among them, the fuzzification layer fuzzifies the input variables and converts them into membership degrees of different fuzzy sets, that is, calculates the degree to which the input data belongs to a certain fuzzy set. Each node is a membership function, representing a fuzzy rule. The rule layer is used to calculate the triggering strength of each rule, which is obtained by multiplying the membership degrees of different fuzzy sets; the normalization layer normalizes the triggering strength obtained from the previous layer; the defuzzification layer is used to calculate the rule output. Finally, through the summation layer, defuzzification is performed to obtain an exact output.

[0048] In some embodiments, as Figure 2 shown, for each input variable, the fuzzification layer uses 3 fuzzy subsets to calculate the membership degrees of each shock wave damage factor; specifically: ; In the formula, is the membership degree of the fuzzy subset , is the membership function of the fuzzy subset , is the warhead charge equivalent is the membership degree of the fuzzy subset , is the membership function of the fuzzy subset , is the warhead landing speed is the membership degree of the fuzzy subset , is the membership function of the fuzzy subset , is the distance between the key component and the warhead explosion point is the fuzzy subset serial number

[0049] There are various membership functions. In some specific embodiments, a Gaussian membership function is adopted, specifically: ; In the formula, is the forward parameter

[0050] Figure 2 In, the rule layer and the normalization layer are combined through the logical operation AND. The rule layer calculates the triggering strength of each fuzzy rule according to the membership degrees of each shock wave damage factor; the normalization layer normalizes the triggering strength of each fuzzy rule to obtain the triggering weight of each fuzzy rule in the entire rule base, which represents the degree of using this rule in the entire reasoning process

[0051] In some embodiments, the defuzzification layer calculates the output value of each fuzzy rule according to the triggering weight of each fuzzy rule. Here, the input variables need to be substituted again, and the output value of each fuzzy rule is represented by a linear combination of the input variables, specifically:

[0052] ;

[0053] In the formula, , is the output value of the defuzzification layer is the th triggering weight of the fuzzy rule is the th output value of the fuzzy rule is the backward parameter of the fuzzy rule

[0054] Finally, the summation layer weights and averages the output values of each fuzzy rule to obtain the shock wave overpressure prediction value

[0055] Step 3: Compare the predicted value of the shock wave overpressure with the damage level criterion of the key components in the underground project to obtain the damage prediction result of the warhead hitting the underground project.

[0056] Specifically, the damage level criterion of the key components in the underground project is shown in Table 1.

[0057] Table 1 Overpressure damage criterion for key components of the system Component Minor damage Moderate damage Severe damage Protective door 5 5~10 ≥10 Supply and exhaust fan 0.067 0.067~0.1 ≥0.1 Ventilation duct 0.08 0.08~0.15 ≥0.15 Dust removal filter 0.02 0.02~0.05 ≥0.05 Generator 0.6 0.6~0.9 ≥0.9 Power transformation equipment 0.6 0.6~0.9 ≥0.9 Power transmission line 0.03 0.03~0.042 ≥0.042 Communication equipment 0.03 0.03~0.05 ≥0.05 Transmission line 0.03 0.03~0.042 ≥0.042 Command electronic equipment 0.03 0.03~0.05 ≥0.05

[0058] In some specific embodiments, it further includes: obtaining penetration damage factors: the complete mass of the warhead, the diameter of the warhead body, and the landing speed of the warhead; based on the penetration damage factors, obtaining the penetration depth of the warhead hitting the underground project through a pre-constructed penetration damage prediction model, and judging whether the underground project protection system is damaged according to the penetration depth; the penetration damage prediction model is constructed by an adaptive fuzzy neural network model.

[0059] According to experience, the factors affecting penetration damage include: the complete mass of the warhead, the landing speed of the warhead, the diameter of the warhead body, the warhead coefficient, and the protection coefficient of the material of the underground project protection section. Since the warhead coefficient and the protection coefficient of the material of the underground project protection section are fixed in actual engineering projects, the present invention only constructs a penetration damage prediction model for the complete mass of the warhead, the diameter of the warhead body, and the landing speed of the warhead.

[0060] In some more specific embodiments, as Figure 3 shown, in the penetration damage prediction model, for each penetration damage factor, the fuzzy layer uses 2 fuzzy subsets to calculate the membership degree of each penetration damage factor.

[0061] The processing processes of other layers are similar to those of the shock wave damage prediction model and will not be elaborated here.

[0062] Embodiment 2

[0063] On the basis of Embodiment 1, this embodiment provides a process for optimizing the forward parameters and backward parameters in the shock wave damage prediction model.

[0064] For the Gaussian membership function, different means and standard deviations will affect the coverage range of the rules. For a large number of data sets, appropriate means and standard deviations can be quickly obtained through the gradient descent algorithm. However, for the present invention with a small data set, the conventional gradient descent algorithm falls into the phenomena of overfitting and unstable convergence. To solve this problem, in this embodiment, when training the shock wave damage prediction model, the Limited-memory BFGS method is used to optimize the forward parameters specifically as follows:

[0065] Forwardly transmit the data of the shock wave damage factor to the summation layer to obtain the predicted value of the shock wave overpressure;

[0066] Taking the mean square error as the loss function, calculate the error between the predicted shock wave overpressure and the true shock wave overpressure:

[0067] ;

[0068] In the formula, is the true value of the shock wave overpressure, is the predicted value of the shock wave overpressure, is the total number of samples, is the loss function, is the forward parameter;

[0069] Store the gradient information of the last iterations, calculate the approximation of the inverse of the low-storage Hessian matrix, and use it to update the forward parameter:

[0070] ;

[0071] In the formula, is the first-order partial derivative vector of the loss function with respect to , is the low-storage Hessian matrix.

[0072] In the optimization of the backward parameter, similarly, the least squares method is usually used to optimize the backward parameter in an appropriate adaptive fuzzy neural network model for a normal data set. For the present invention, a small data set will lead to overfitting and unstable convergence. To solve this problem, in the training of the shock wave damage prediction model in this embodiment, the regularized least squares method (Ridge Regression, L2 regularization) is used to optimize the backward parameter. It includes:

[0073] Forwardly transmit the data of the shock wave damage factor along the adaptive fuzzy neural network model to the defuzzification layer, fix the forward parameter, and use the regularized least squares method to optimize the backward parameter. Specifically:

[0074] Write the predicted output of the shock wave overpressure in matrix form:

[0075] ;

[0076] In the formula, is the input matrix composed of the triggering weight and the shock wave damage factor, is the backward parameter matrix;

[0077] Establish an optimization objective function:

[0078] ;

[0079] In the formula, is the regularization coefficient, is the L2 regularization term;

[0080] Solve the optimization objective function to obtain the optimized backward parameter matrix:

[0081] ;

[0082] In the formula, is the identity matrix, which is used to ensure is invertible. In the conventional least squares method, When the data is scarce, will be close to a singular matrix, the inverse operation is unstable and may have large numerical errors. The regularized least squares method is used to avoid the instability problem of matrix inversion, improve the calculation stability, reduce overfitting, and improve the generalization ability.

[0083] Example 3

[0084] Based on Example 2, this example provides the performance of the penetration damage prediction model and the shock wave damage prediction model in the simulation experiment.

[0085] Experiment 1: Use the penetration damage prediction model to simulate and predict the penetration damage of underground engineering. By presetting the fixed warhead shape and the protection coefficient of the target underground engineering protection section material, the penetration depth is deduced using the warhead mass, projectile diameter, warhead landing speed, etc. The obtained fuzzy rules will be applicable to the penetration damage calculation under different warhead masses, projectile diameters, and warhead landing speeds under the same warhead shape and the same protection material state.

[0086] Table 2 shows the comparison between the output results obtained by simulating the test samples using the penetration damage prediction model and the actual results.

[0087] Table 2 Comparison of experimental results and simulation results of penetration depth for fixed warhead shape and underground protection section material Warhead mass Projectile body diameter Warhead landing speed Experimental penetration depth Simulated penetration depth 1360 0.594 532.44 3.419 3.527 670 0.340 158.35 0.755 0.744 361 0.311 764.88 5.670 5.773 361 0.188 606.62 4.012 4.250 1360 0.533 442.61 2.754 2.721 1000 0.501 1414.65 11.967 12.188 110 0.192 345.48 0.970 0.871 2270 0.370 330.36 5.323 5.326

[0088] Figure 4 Shows the results of error analysis with the penetration depth as the error analysis index. As can be seen from the figure, the results trained by this model have a small error compared with the actual results.

[0089] In fact, in the test, MATLAB was used to compare the penetration depth experiment with the actual results, and the error fluctuated between 5% and 7%. This error is completely acceptable compared with the error brought by the empirical formula. To ensure that the experimental results are not accidental, in this embodiment, an experiment was also conducted with all parameters of the warhead fixed and the protection coefficient of the material of the protected section of the target underground project. Based on the experimental data and error comparison of different warhead landing speeds, the comparison results are shown in Table 3.

[0090] Table 3 Comparison of Experimental Results and Simulation Results of Different Warhead Landing Speeds Warhead mass Warhead diameter Warhead coefficient Warhead landing speed Experimental penetration depth Simulated penetration depth Error % 227 0.273 1.00 343.11 4.373 4.233 3.2 227 0.273 1.00 390.45 4.938 4.701 4.8 227 0.273 1.00 470.24 5.842 6.175 5.7 227 0.273 1.00 510.50 6.288 6.577 4.6 227 0.273 1.00 630.76 7.604 7.163 5.8

[0091] The average error of this experiment remained at about 5.2%. The simulation experiment results proved the feasibility of using ANFIS to study the penetration damage effect. However, since the final results of the penetration formula and the ANFIS penetration damage prediction model are only the penetration depth, and this depth can only reflect whether the dynamic load section of the underground project is penetrated and whether the protection system is damaged, and cannot directly reflect the actual damage degree suffered by other systems inside the underground project, it is necessary to study the shock wave damage generated after penetration.

[0092] In Experiment 2, the shock wave damage prediction model was trained using the shock wave damage data set. During the training process, ANFIS converged quickly. Table 4 shows the output results of some training data sets, and the average error of the training results was only about 5%.

[0093] Table 4 Error Analysis of Training Data Set of Shock Wave Damage Prediction Model Warhead mass Warhead landing speed Penetration depth Distance from explosion point Experimental shock wave overpressure Simulated shock wave overpressure 2270 343.62 6.843 140 1.460 1.461 227 340.13 1.265 140 0.147 0.159 110 344.15 0.858 100 0.051 0.063 2270 443.01 7.076 100 1.911 2.005 110 399.85 1.426 100 0.070 0.076 227 347.89 1.303 120 0.171 0.163

[0094] After training, the heavy and medium mass precision-guided weapons of the US military's AGM series were used again to vertically strike the dynamic load section protection layer at 90°. The striking speed was at least the speed of sound and not more than 3 Mach at most. A damage experiment was conducted on an underground project model with one end open. Using the trained shock wave damage prediction model, the penetration depth was simulated by linear interpolation of the warhead landing speed and compared with the 5m-thick underground protection concrete layer to judge whether the mouth protection system of the underground project was penetrated under this condition. The relevant data related to the shock wave damage factors were extracted and combined as Test Data Set 1 and Test Data Set 2.

[0095] The test data was tested in the already trained shock wave damage prediction model. Table 5 and Table 6 respectively correspond to the training results of Test Data Set 1 and Test Data Set 2.

[0096] Table 5 Error Analysis of Test Data Set 1 Warhead mass Warhead landing speed Penetration depth Distance from explosion point Experimental shock wave overpressure Simulated shock wave overpressure 1360 449.90 3.381 100 1.161 1.104 1360 488.92 3.796 100 1.294 1.332 1360 478.51 3.683 120 1.072 1.125 1360 524.62 4.189 140 1.040 0.998 1360 515.83 4.091 200 0.757 0.843

[0097] Table 6 Error Analysis of Test Dataset 2 Warhead mass Warhead landing speed Penetration depth Distance from explosion point Experimental shock wave overpressure Simulated shock wave overpressure 361 606.72 4.989 100 0.402 0.443 361 719.19 6.402 100 0.490 0.484 361 877.62 8.586 120 0.540 0.542 361 764.78 7.008 140 0.414 0.403 361 709.52 6.276 200 0.287 0.305

[0098] The final average errors of the experiments are 4.674% and 3.851% respectively. By analyzing the errors, it is found that the ANFIS trained is more accurate in simulating the damage of small-mass warheads. The main reason is that the damage effect of small-mass earth-penetrating missiles on underground projects changes more significantly with the change of speed. Of course, the average error between the prediction results and the actual results is 4.262%, which is completely acceptable. This experiment demonstrates the feasibility of ANFIS in predicting shock wave damage.

[0099] Experiment 2 is an experiment on an underground project model with one end open. To avoid the contingency of the experiment, Experiment 3 is a comparison of the experimental data and simulation data of the destruction of an underground project model with both ends open to ensure the feasibility of ANFIS in predicting shock wave damage. The experimental results are shown in Table 7.

[0100] Table 7 Error Analysis of Test Dataset 3 Warhead mass Warhead landing speed Penetration depth Distance from explosion point Experimental shock wave overpressure Simulated shock wave overpressure 2270 353.54 6.887 140 1.152 1.153 227 340.45 1.268 140 0.116 0.125 110 382.11 1.338 100 0.058 0.050 2270 343.01 6.826 100 1.552 1.580 1000 667.75 6.576 200 2.112 2.004 1360 532.63 4.279 120 0.972 1.040

[0101] Through MATLAB, the error analysis of the experimental results and simulation results is carried out. The final average error is 7.42%. Although the error is slightly larger than that of the above model, it is still smaller than the error between the empirical formula results and the experimental results and is within the acceptable range. At the same time, it also proves the feasibility of ANFIS in predicting shock wave damage under different conditions.

[0102] Example 4

[0103] This embodiment provides an electronic device, including: a processor, a communication interface, a communication bus, and a memory; wherein, the processor, the communication interface, and the memory complete mutual communication through the communication bus, and a computer-readable instruction is stored on the memory. The processor can call the computer-readable instruction in the memory to execute the underground project damage prediction method based on the adaptive fuzzy neural network described in Embodiment 1 or Embodiment 2.

[0104] In addition, when the computer-readable instructions in the above-mentioned memory can be implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0105] Example 5

[0106] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for predicting damage to underground engineering based on an adaptive fuzzy neural network described in Embodiment 1 or Embodiment 2. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0107] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. An underground engineering damage prediction method based on an adaptive fuzzy neural network, characterized in that Including: Model the scenarios of underground engineering and warheads, and simulate the damage of warheads to underground engineering under the established scenarios to obtain shock wave damage factors; The shock wave damage factors include: warhead charge equivalent, warhead landing speed, and the distance between key components and the warhead explosion point; Use a pre-constructed shock wave damage prediction model to obtain the predicted value of shock wave overpressure; Compare the predicted value of shock wave overpressure with the damage level criterion of key components in the underground engineering to obtain the damage prediction result of the warhead hitting the underground engineering; Among them, the shock wave damage prediction model is constructed by an adaptive fuzzy neural network model.

2. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 1, characterized in that The shock wave damage prediction model includes: a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer; The fuzzification layer is used to calculate the membership degree of each shock wave damage factor; The rule layer is used to calculate the triggering intensity of each fuzzy rule according to the membership degree of each shock wave damage factor; The normalization layer is used to normalize the triggering intensity of each fuzzy rule to obtain the triggering weight of each fuzzy rule in the entire rule base; The defuzzification layer is used to calculate the output value of each fuzzy rule according to the triggering weight of each fuzzy rule; The summation layer is used to perform weighted averaging on the output values of all fuzzy rules according to the triggering weight to obtain the predicted value of shock wave overpressure.

3. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 2, wherein The calculation of the membership degree of each shock wave damage factor is specifically: ； Wherein, is the membership degree of the fuzzy subset ; is the membership function of the fuzzy subset ; is the charge equivalent of the warhead, is the membership degree of the fuzzy subset ; is the membership function of the fuzzy subset ; is the landing speed of the warhead, is the membership degree of the fuzzy subset ; is the membership function of the fuzzy subset ; is the distance between the key component and the explosion point of the warhead, is the serial number of the fuzzy subset.

4. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 3, characterized in that, The membership function adopts a Gaussian membership function, specifically: ； In the formula, is the forward parameter.

5. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 4, characterized in that, The defuzzification layer calculates the output value of each fuzzy rule according to the triggering weight of each fuzzy rule, specifically: ； wherein, , is the output value of the clarification layer, is the trigger weight of the th fuzzy rule, is the output value of the th fuzzy rule, and is the backward parameter of the fuzzy rule.

6. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 5, characterized in that, When training the adaptive fuzzy neural network model, use the L-BFGS method to optimize the forward parameters, specifically: Forwardly transfer the data of shock wave damage factors to the summation layer to obtain the predicted value of shock wave overpressure; Use the mean square error as the loss function to calculate the error between the predicted shock wave overpressure and the real shock wave overpressure: ； In the formula, is the true value of the shock wave overpressure, is the predicted value of the shock wave overpressure, is the total number of samples, is the loss function, is the forward parameter; Store the most recent gradient information of the iteration, calculate an approximation of the inverse of the low-storage Hessian matrix, and use it to update the forward parameters: ； In the formula, is the first-order partial derivative vector of the loss function with respect to , and is the low-storage Hessian matrix.

7. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 6, characterized in that, When training the adaptive fuzzy neural network model, use the regularized least squares method to optimize the backward parameters, specifically: Operate the data of shock wave damage factors along the adaptive fuzzy neural network model to the defuzzification layer, fix the forward parameters, and use the regularized least squares method to optimize the backward parameters, specifically: Write the shock wave overpressure prediction output in matrix form: ； In the formula, is the input matrix composed of the trigger weight and the shock wave damage factor, is the backward parameter matrix; Establish an optimization objective function: ； In the formula, is the regularization coefficient, is the L2 regularization term; Solve the optimization objective function to obtain the optimized backward parameter matrix: ； In the formula, is the identity matrix.

8. The method for predicting damage of underground engineering based on an adaptive fuzzy neural network according to claim 1, characterized in that Also including: Obtain penetration damage factors: the complete mass of the warhead, the diameter of the warhead body, and the warhead landing speed; Based on the penetration damage factors, obtain the penetration depth of the warhead hitting the underground engineering through a pre-constructed penetration damage prediction model, and judge whether the underground engineering protection system is damaged according to the penetration depth; The penetration damage prediction model is constructed by an adaptive fuzzy neural network model.

9. An electronic device, characterized in that, Including: A processor and a memory; The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method for predicting damage to underground engineering based on an adaptive fuzzy neural network according to any one of claims 1-8 is implemented.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the method for predicting damage to underground engineering based on an adaptive fuzzy neural network according to any one of claims 1-8.

Images