Underground space suitability evaluation method based on fuzzy hierarchical synthesis method and supervised learning

Abstract

The invention discloses an underground space suitability evaluation method based on a fuzzy hierarchical synthesis method and supervised learning, and belongs to the field of underground space suitability evaluation, and the method comprises the following steps: obtaining three-dimensional data of an underground space suitability evaluation index; preprocessing the three-dimensional data to obtain standardized evaluation index data; constructing an improved BP neural network model and an improved fuzzy comprehensive evaluation method, and performing evaluation processing on the standardized evaluation index data based on the improved fuzzy comprehensive evaluation method to obtain a label sample; training the improved BP neural network model based on the label sample to obtain an optimal model; and performing urban underground space suitability classification based on the optimal model.

Description

Technical Field

[0001] The present invention belongs to the technical field of underground space suitability assessment, and in particular relates to an underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning. Background Art

[0002] Although the evaluation methods are being continuously enriched and improved, there are still some shortcomings. For example, the weighted evaluation model has the advantage of simple calculation, but does not consider the fuzziness between classifications; the traditional fuzzy comprehensive evaluation method mostly adopts the maximum membership classification without considering the validity. In addition, the most critical thing is that most of these methods are based on traditional evaluation methods, which often have the defects of complex calculation, low efficiency and non-reusable data. The biggest feature of urban underground space suitability evaluation, especially at the macro scale, is that the data is massive (hundreds of thousands, millions, and tens of millions of grid data). The huge short board effect of traditional evaluation methods facing this feature is low efficiency and non-reusable data. It can be seen that the evaluation method for urban underground space suitability still needs further exploration, especially the combination of traditional evaluation methods and new evaluation methods.

[0003] With the continuous rise of artificial intelligence, some scholars have also applied machine learning to underground space evaluation in recent years. In the existing technology, the triangular fuzzy analytic hierarchy process, genetic algorithm and GIS platform are combined to carry out a 2D suitability evaluation of the Xiaohe Modern Industrial Park in Shanxi Province. In recent years, some scholars have also explored 3D evaluation, but the 3D evaluation carried out lacks systematic research and is still carried out under traditional evaluation methods. Therefore, it is necessary to further explore the combination of 3D suitability evaluation, especially traditional evaluation methods and new methods, in a systematic and comprehensive manner. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides an underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning, comprising:

[0006] Obtain three-dimensional data of underground space suitability evaluation indicators;

[0007] Preprocessing the three-dimensional data to obtain standardized evaluation index data;

[0008] Constructing an improved BP neural network model and an improved fuzzy comprehensive evaluation method, evaluating and processing the standardized evaluation index data based on the improved fuzzy comprehensive evaluation method to obtain label samples; training the improved BP neural network model based on the label samples to obtain an optimal model;

[0009] The suitability classification of urban underground space is carried out based on the optimal model.

[0010] Optionally, the process of obtaining three-dimensional data of underground space suitability evaluation indicators includes:

[0011] Construct 3D geological models based on geological data;

[0012] Obtain underground space suitability evaluation indicators based on quantitative and qualitative analysis;

[0013] Constructing a three-dimensional evaluation index system based on the three-dimensional geological model and the underground space suitability evaluation index;

[0014] Corresponding three-dimensional data is obtained based on the three-dimensional evaluation index system.

[0015] Optionally, the underground space suitability evaluation indicators include height difference, slope, rock mass characteristics, soil mass characteristics, surface water impact, groundwater depth, and groundwater corrosivity.

[0016] Optionally, the process of constructing the improved fuzzy comprehensive evaluation method includes weight improvement and evaluation grading improvement;

[0017] The weight improvement process includes constructing variable weights based on the idea of ​​praise and criticism, and optimizing the weights based on the variable weights;

[0018] When it is a decreasing function,

[0019] When it is an increasing function,

[0020] When the indicator weight changes slightly, ω k ′=ω k ±f(x k )(3≤k≤n-3)

[0021] In the formula, f(x i ) is a function of the indicator state value, ω1, ω2, …, ω n Represents the normalized combined weight of the objective weight and the subjective weight.

[0022] Optionally, the process of improving the evaluation grading includes: introducing effectiveness for evaluation grading, when the effectiveness is less than 0.5, adopting the maximum membership principle for evaluation grading, when the effectiveness is less than 0.5, constructing a credibility mathematical model based on the effectiveness, and performing evaluation analysis based on the credibility mathematical model.

[0023] Optionally, the process of constructing the improved BP neural network model includes optimizing the BP neural network based on an improved Red-tailed Hawk algorithm.

[0024] Optionally, the process of improving the red-tailed hawk algorithm includes: improving the high-altitude soaring stage of the red-tailed hawk, improving the low-altitude flying stage of the red-tailed hawk, and improving the stooping stage of the red-tailed hawk;

[0025] The mathematical expression of the best position for the red-tailed hawk to find food during its high-altitude soaring phase is:

[0026] X(t)=X best +(X mean -X(t-1)·Levy(dim)·TF(t)

[0027] In the formula, X(t) represents the position at iteration t, X best The best position obtained, X mean is the average position.

[0028] Optionally, the process of improving the low-altitude flight stage of the red-tailed hawk includes constructing a reverse solution based on elite reverse learning; wherein, the expression for obtaining a new generation of individuals based on the reverse solution and the current solution is:

[0029]

[0030] k is a random number between 0 and 1, a j (t), b j (t) is the interval constructed by the elite population.

[0031] Optionally, the process of improving the stooping phase of the red-tailed hawk includes adding disturbances to the optimization process using an improved sine-cosine algorithm;

[0032] Among them, the expression of the improved sine-cosine algorithm is:

[0033]

[0034] t is the number of iterations, T max is the maximum number of iterations, ω t is the inertia weight, x i (t) is the i-th position of individual x in the t-th iteration, p i(t) is the component of the i-th position of the best individual position in the t-th iteration, r1 is a decreasing function usually between [0,a], r2 is a random number on the interval [0,2π]], a is a constant, r3 is a random number on the interval [-2,2], and r4 is a random number on the interval [0,1].

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] Aiming at the outstanding defects of traditional suitability evaluation methods, such as large amount of calculation and non-reusable data, resulting in low efficiency, a set of supervised learning urban underground space evaluation methods based on the combination of improved fuzzy hierarchical comprehensive evaluation method and machine learning was explored. The research of this invention solves the key problems of low accuracy, low efficiency and weak visualization of traditional suitability evaluation methods. In addition, the research of this invention also provides new ideas for the suitability evaluation of urban underground space. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0038] Figure 1 It is a flow chart of the improved fuzzy hierarchical synthesis method evaluation system of an embodiment of the present invention;

[0039] Figure 2 This is a flow chart of the improved BP neural network structure according to an embodiment of the present invention;

[0040] Figure 3 This is an evaluation statistical chart under multiple constraints and single constraints in the study area of ​​an embodiment of the present invention. DETAILED DESCRIPTION

[0041] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0042] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] Embodiment 1

[0044] In this embodiment, a method for evaluating the suitability of underground space based on fuzzy hierarchical synthesis method and supervised learning is provided, comprising the following steps: obtaining three-dimensional data of underground space suitability evaluation indicators; preprocessing the three-dimensional data to obtain standardized evaluation indicator data; constructing an improved BP neural network model and an improved fuzzy comprehensive evaluation method, and evaluating and processing the standardized evaluation indicator data based on the improved fuzzy comprehensive evaluation method to obtain labeled samples; training the improved BP neural network model based on the labeled samples to obtain an optimal model; and classifying the suitability of urban underground space based on the optimal model.

[0045] like Figure 1 As shown in the figure, the improved fuzzy comprehensive evaluation method mainly improves the weight construction and evaluation classification. The weights constructed by the traditional hierarchical analysis and combined weighting method do not take into account the actual environment of the indicators. Based on the principle of praise and criticism, a variable combination mathematical model is established to improve the traditional hierarchical analysis method. Secondly, the traditional fuzzy comprehensive method does not consider the validity when evaluating and grading. Through analysis and demonstration, it is found that the validity is invalid when it is lower than 0.5. The credibility is established according to the validity principle, and the credibility is used for improvement when the validity is invalid. The improved fuzzy hierarchical comprehensive method provides theoretical support for suitability evaluation.

[0046] The present invention adopts the idea of ​​praise and criticism to construct variable weights. Assume that the weights obtained by subjective method are g1, g2, ..., g n , the weights obtained by the objective method are s1, s2, ..., s n , the normalized combination weights are ω1, ω2, …, ω n At the same time, let the variable weight coefficients be f(x1), f(x2), ..., f(x n ), when the weight is too high and the indicator is not so important, it needs to be reduced, which is called a devaluation function. Then we have:

[0047]

[0048] On the contrary, when the weight of some indicators is relatively low and the impact of the indicators is very important, they need to be increased, which is called a reward (increasing function). Then, we have:

[0049]

[0050] In addition, some indicators may have small changes in weight, in which case they can be set as:

[0051] ω k ′=ω k ±f(x k ) (3≤k≤n-3) (3)

[0052] Among them, f(x i) is a function about the indicator state value. Assume that the function is a parabola. When the weight decreases, it is a decreasing function about x. When the weight increases, it is an increasing function about x. When the weight does not change or changes very little, it is infinitely close to 0.

[0053] The process of improving the evaluation grading includes: introducing validity for evaluation grading; when the validity is less than 0.5, adopting the maximum membership principle for evaluation grading; when the validity is less than 0.5, constructing a credibility mathematical model based on the validity; and conducting evaluation analysis based on the credibility mathematical model.

[0054] In the evaluation of suitability of urban underground space, data is often massive and requires a large amount of calculation. In addition, traditional evaluation methods are not repeatable, that is, when new data or new plots are evaluated, complex calculations need to be performed again, which is inefficient. Machine learning is a cross-disciplinary subject involving probability theory, statistics, approximation theory, linear algebra, advanced mathematics, etc. It uses computers as tools and truly simulates human learning methods to effectively improve efficiency by dividing existing content into knowledge structures. Therefore, this chapter explores the evaluation method of combining improved fuzzy hierarchical synthesis method with machine learning. Machine learning can simulate human learning and perform automatic classification by training labeled samples.

[0055] Since the evaluation of urban underground space suitability is a complex system involving multiple disciplines, its evaluation involves hydrological environment, geotechnical engineering, social economics, etc. The evaluation indicators of different disciplines are quite different and the units are not uniform. In order to eliminate the influence of different dimensions, the indicator data needs to be processed. Common data processing methods in machine learning include normalization and standardization.

[0056] The purpose of normalization is to convert the data to between 0 and 1. The principle is to subtract the minimum value from the evaluation index value and divide it by the difference between the maximum and minimum values. The formula is:

[0057]

[0058] Standardization is to scale the value of a column of numerical features in the training set (assuming it is the i-th column) into a normal distribution with a mean of 0 and a variance of 1. The formula is as follows:

[0059]

[0060] in

[0061]

[0062] Building an Improved Red-Tailed Hawk Algorithm Model

[0063] The Red-tailed Hawk (RTH) is a new algorithm that simulates hunting behavior. The hunting process of the Red-tailed Hawk includes three stages: high flying, low altitude, and stooping. The high-altitude soaring stage is the area where the Red-tailed Hawk explores the space to determine the exploration position of the prey. The low-altitude soaring stage moves around the prey to choose the best hunting position, and then it swings and hits the target in the sharp turn and dive stage. The algorithm simulates the predation of the Red-tailed Hawk to solve the optimization problem. Although the Red-tailed Hawk algorithm has the advantages of fast convergence and high accuracy compared with traditional algorithms, it lacks global exploration ability and is easy to fall into the local optimal solution. Therefore, an improved Red-tailed Hawk algorithm is proposed, which is mainly improved from three stages.

[0064] Bernoulli mapping improvement: Since the initial population of the Red-tailed Hawk algorithm is random, the search for solutions by this randomly distributed population is also random, which will affect the search space. The use of chaotic mapping replaces the randomness in the intelligent algorithm search process during the initialization position update process. The use of chaotic mapping method can expand the search range of the space. In addition, the use of chaotic mapping method can not only expand the search range but also increase the diversity of the population, so chaotic mapping is used to initialize the population. Common chaotic mappings include Tent mapping, Singer mapping, Bernoulli mapping, Cubic mapping, and Logistic mapping.

[0065] By comparing and analyzing the four chaotic mapping methods of Cubic mapping, Logistic mapping, Bernoulli mapping, and Singer mapping, it is found that the three mapping methods of Cubic, Logistic, and Singer are all densely distributed in the upper right corner. Compared with these three methods, Bernoulli mapping is more uniform, so Bernoulli mapping is used to initialize the red-tailed hawk population. The formula is as follows:

[0066]

[0067] Among them, 0<λ<1,λ=0.5

[0068] (1) High-altitude soaring stage: During the high-altitude soaring stage, the red-tailed hawk flies far into the sky to find the best position for food. The mathematical model is as follows:

[0069] X(t)=X best +(X mean -X(t-1)·Levy(dim)·TF(t) (8)

[0070] X(t) represents the position at iteration t, X best The best position obtained, X meanis the average position, Levy is calculated according to formula (9), and TF(t) is the conversion factor function according to formula (10);

[0071]

[0072] Where S is a constant (taken as 0.01), dim is the dimension, β is a constant 1.5, and μ and v are random numbers between [0,1].

[0073] TF(t)=1+sin(2.5+t / T max ) (10)

[0074] Where T max The maximum number of iterations, t is the number of iterations

[0075] (2) Low-altitude flight: The eagle flies in a spiral pattern much lower to the ground, circling its prey. Its model can be expressed as follows:

[0076]

[0077] Where x and y represent the direction coordinates, and the formula is as follows

[0078]

[0079] R0 represents the initial value of the radius between [0.5,3], A represents the gain between [5,15], rand represents a random number between [0,1], and r represents the control gain between [1,2]. These parameters help the eagle fly around its prey in a spiral motion.

[0080] 2.0 uses elite opposition-based learning for improvement: The solution obtained in the second stage of the Red-tailed Hawk algorithm may be a local optimal solution rather than a global optimal solution, so elite opposition-based learning is used for improvement. Elite opposition-based learning (EOBL) constructs the reverse solution of the feasible solution of the current problem to increase the diversity of the population, and selects the optimal solution from the current solution and the reverse solution as the new generation of individuals. The formula is as follows:

[0081]

[0082] k is a random number between 0 and 1, a j (t), b j (t) is the interval constructed by the elite population.

[0083]

[0084] (3) Stooping stage: In this stage, the eagle suddenly bends down and attacks the prey from the best position obtained during the low-altitude soaring stage. The formula is as follows:

[0085] X(t)=α(t)·X best +x(t)·StepSize1(t)+y(t)·StepSize2(t) (15)

[0086] Each step size can be calculated as follows:

[0087]

[0088] Where α and G are acceleration and gravity factors respectively, which can be defined as follows:

[0089]

[0090] Among them, α represents the acceleration of the eagle which increases with the increase of t to improve the convergence speed; G is the gravity effect, which is reduced when the eagle is very close to the prey in order to reduce the diversity of exploitation.

[0091] 3.0 uses sine and cosine algorithms for improvement: At the end of the red-tailed hawk algorithm, it may fall into a local search state. If the previous calculation deviates from the current value, the optimization in the local area is invalid. Therefore, in order to ensure that the local area can be jumped out again in the final stage, the search group is randomly expanded to an area far away from the target to check whether the optimal target solution is reached, so as to achieve the effect of global development. Therefore, the disturbance method is often used for improvement. Common disturbance mutation methods include western mutation, differential mutation, Gaussian mutation, sine and cosine, etc. The present invention uses an improved sine and cosine algorithm to improve the red-tailed hawk.

[0092] Sine Cosine Algorithm (SCA) is a relatively new swarm intelligence optimization algorithm with the advantages of few parameters, easy implementation, and fast convergence. However, it also has the disadvantages of premature convergence and low solution accuracy, so some scholars have made improvements. The present invention constructs nonlinear adaptive weights based on existing research results, and improves the Sine Cosine Algorithm with nonlinear adaptive weights. The improved Sine Cosine Algorithm optimizes the last stage of the Red Tailed Hawk Algorithm.

[0093] Assuming the population size is N and the search dimension is d, the position of individual i in the d-dimensional space can be expressed as The improved sine-cosine algorithm formula is as follows:

[0094]

[0095] in:

[0096] k is the adjustment coefficient, which is a number greater than 0. Through experimental analysis, k=3.

[0097] t is the number of iterations, T max is the maximum number of iterations, ω t is the inertia weight, x i (t) is the i-th position of individual x in the t-th iteration, p i (t) is the component of the i-th position of the best individual position in the t-th iteration, r1 is a decreasing function usually between [0,a], r2 is a random number in the interval [0,2π]]. a is a constant usually 2, r3 is a random number in the interval [-2,2], and r4 is a random number in the interval [0,1].

[0098] Improved Red Tailed Hawk Algorithm Test:

[0099] In order to verify the superiority of the improved red-tailed hawk algorithm, 8 internationally used benchmark functions were selected and compared using particle swarm optimization (PSO), northern goshawk algorithm (NGO), gray wolf algorithm (GWO), genetic algorithm (GA), red-tailed hawk algorithm (RTH), and improved red-tailed hawk algorithm (CRTH). The population size of the six algorithms was set to 30, and the maximum number of iterations was 100. The improved red-tailed hawk algorithm converged faster, especially on functions F7 and F8. When the number of iterations reached the maximum number of 100, the fitness value of the red-tailed hawk algorithm was better than that of the other algorithms, and the improved red-tailed hawk algorithm was better than that of the unimproved one. The above results verified that the solution obtained by the improved red-tailed hawk algorithm was optimal.

[0100] In order to quantitatively analyze each algorithm, the average value and the optimal value are used as evaluation indicators. From the analysis of the following table, it can be seen that the average values ​​obtained by the improved Red-tailed Hawk algorithm on the 8 test functions are the smallest, and the optimal values ​​are obtained on the 8 test functions, especially the optimal values ​​obtained on the test functions F1, F3, F4 and F9 are closer to the standard value than the other algorithms. The above results once again prove that the use of the multi-effect improved Red-tailed Hawk algorithm is reliable and correct. The algorithm analysis of this embodiment verifies the improvement of the evaluation accuracy of urban underground space in subsequent supervised learning.

[0101] Table 1

[0102]

[0103] Analysis of supervised learning evaluation methods: Combination of IMFAHP-SL evaluation methods

[0104] Since supervised learning has the advantages of high accuracy and good stability, this chapter mainly combines the improved fuzzy analytic hierarchy process and supervised learning (IMFAHP-SL) to carry out the suitability evaluation research of urban underground space development. The key technical process is as follows:

[0105] Key steps to adopt IMFAHP-SL combination:

[0106] (1) Extract three-dimensional data of evaluation indicators such as height difference, slope, rock mass characteristics, soil mass characteristics, surface water impact, groundwater depth, and groundwater corrosivity.

[0107] (2) Standardize the evaluation index data.

[0108] (3) Use the improved fuzzy comprehensive evaluation method (IMFAHP) to label the data.

[0109] (4) The evaluation data is divided into training set, validation set and test set. Five models including BP, SVC, improved BP1, BP2 and BP3 are used for training and the appropriate classification model is selected.

[0110] (5) The data is then fed into a suitable model for training and testing, and the improved Mantis algorithm is used for parameter optimization.

[0111] (6) Model comparison and analysis were performed and verified using indicators such as accuracy, precision, and support. The optimized optimal model was used to classify the suitability of urban underground space.

[0112] Model verification and evaluation indicators: In order to verify the superiority of improved BP1 (RTH-BP) and improved BP2 (CRTH-BP), the superiority of the improved BP neural network is also compared with the traditional classical particle swarm algorithm (POS) to improve BP. In addition, in order to verify the superiority of the improved BP neural network model, the three improved BP neural network models are verified with the traditional BP neural network model and support vector machine (SVC).

[0113] The results of the evaluation of the suitability of the development and utilization of the urban underground space in Chengdu Airport New Town using the traditional BP neural network model are shown in the following table. From Table 1, it can be concluded that in order to comprehensively evaluate the performance of the model, the model is evaluated using precision, recall, F1 score, support, and accuracy. The evaluation formula is as follows:

[0114]

[0115]

[0116] Improved BP neural network model construction:

[0117] BP neural network initial weight and bias optimization

[0118] Since the weights and biases of traditional BP neural networks are random, and weights and biases are two key parameters that have an important impact on model performance and learning process. Appropriate initial weights and biases can accelerate model convergence and find the optimal solution. If the initial value is too small, the gradient may disappear and the weight update may be very slow. On the contrary, if the initial value is set too large, it may cause the gradient to explode and the weight update to be too large, making it difficult for the model to converge. The distribution of initial weights and biases has random differences, and the distribution range is approximately -0.5 to -0.3 is 66, -0.3 to -0.1 is 546, -0.1 to 0.1 is 552, 0.1 to 0.3 is 510, 0.3 to 0.5 is 54, and the initial bias is basically around 0.

[0119] Based on the shortcomings of the traditional BP neural network model, an improved BP neural network model is proposed. The BP neural network is improved by using the red-tailed hawk algorithm (RTH) and the improved red-tailed hawk algorithm (CRTH) proposed in the previous article. Figure 2 shown.

[0120] BP neural network optimal model construction:

[0121] Different BP neural network structures, such as different network layers, training times, and number of neurons, have a great impact on the results. Therefore, quantitative analysis was performed on different network layers, training times, and number of neurons. When the number of training times was 10, the loss values ​​on the training set and validation set were 0.024274 and 0.01470, respectively. When the number of training times was 50, the loss values ​​on the training set and validation set were 0.00921 and 0.00858, respectively. The results show that when the number of training times was 10, the loss value did not converge and the loss value was relatively large, while when the number of training times was 50, the loss value did not decrease and the loss value was relatively small. Therefore, 50 training times was the optimal training time for the model. At the same time, different accuracy rates were obtained when the number of neurons was 4, 8, 16, 32, and 64, respectively, but the accuracy was the highest when the number of neurons was 32, so 32 was the optimal number of neurons. Similarly, different accuracy rates are obtained when the number of network structure layers is 3, 4, 5, and 6 respectively, but the highest accuracy rate is obtained when the network structure is 3 layers. Therefore, the network structure is designed to be 3 layers as the optimal number of network structure layers.

[0122] Supervised learning evaluation method system development:

[0123] The system mainly uses Python to develop backend logic, and combines the Web system to develop supervised learning evaluation methods, and develops a supervised learning urban underground space suitability evaluation system. The system uses two modules, front-end and back-end. The front-end mainly uses HTML, CSS, and JavaScript languages. The framework is built using the Vue3 combined API. Pinia is used to implement browser persistent storage, Tailwind CSS is used to beautify the interface, and Vue Router is used to implement route navigation and interface switching. The back-end of the system uses Python's Flask framework and RESTful API architecture to access and process data, thereby realizing the separation of front-end and back-end. Through API calls, the front-end can interact with the back-end, including calling the back-end neural network model and data interaction.

[0124] The developed urban underground space evaluation system (DUUSS-1) uses a BP neural network to train a model 50 times. The system can set parameters such as the number of training times, learning rate, number of input features, number of hidden layers, and output features. After setting the parameters, the evaluation results and graphs can be automatically generated by importing the data.

[0125] Supervised learning evaluation method case application:

[0126] Supervised learning research on regional evaluation and grading: The constructed three-dimensional geological model is used to divide the model into 150×150×30 grids, and the centroid of the grid is used as the center point of data storage. The coordinate data and ID number of the center point of the grid are extracted to divide a total of 104,501 square grid data, and these grids are attached with attribute values ​​and labels. First, the sample data is standardized, and then the data is divided into a training set and a test set with a ratio of 0.2, and then the training set is divided into a training set and a validation set with a ratio of 0.2. Finally, the developed underground space evaluation system is used for training, validation and testing.

[0127] The Red-tailed Hawk Algorithm (RTH) and the Improved Red-tailed Hawk Algorithm (CRTH) were used to improve the BP neural network, which were abbreviated as Improved BP2 (RTH-BP) and Improved BP3 (CRTH-BP). At the same time, in order to verify the superiority of the Red-tailed Hawk Algorithm (RTH) and the Improved Red-tailed Hawk Algorithm (CRTH), they were also compared with the traditional classical particle swarm algorithm (PSO) and abbreviated as Improved BP2 (PSO-BP). In addition, in order to verify the superiority of the improved BP neural network model, it was also compared with the traditional BP neural network model and support vector machine (SVC). These five evaluation methods were applied to the suitability evaluation of the urban underground space development in the Airport New Town of the Eastern New District of Chengdu.

[0128] Table 2

[0129]

[0130] In order to verify the reliability of the evaluation of the BP neural network, the more classic support vector machine (SVC) was also used for verification. The results are shown in the following table. From the table, it can be seen that the accuracy of using the BP neural network and SVC is not much different.

[0131] Table 3

[0132]

[0133] Due to the advantages of particle swarm algorithm (POS), such as simplicity, fewer parameters and fast speed, the particle swarm algorithm is used to improve the BP neural network, namely the improved BP1. Then the improved BP1 is used to carry out the suitability evaluation of the development and utilization of the urban underground space in Chengdu Airport New City. The results are shown in Table 4.

[0134] Table 4

[0135]

[0136] Although the POS-improved BP has improved precision and accuracy over the traditional BP, the POS algorithm is prone to fall into the local optimal solution. Therefore, the Red Tailed Hawk (RTH) algorithm is used to improve the BP neural network, that is, the improved BP2. Then the improved BP2 is used to conduct a suitability evaluation on the development and utilization of the urban underground space in Chengdu Airport New Town. The results are shown in Tables 5 and 6.

[0137] Table 5

[0138]

[0139] Table 6

[0140]

[0141] Supervised learning to study the three-dimensional suitability of the region:

[0142] The idea of ​​block assignment and block evaluation is adopted to integrate the previous evaluation results with the three-dimensional geological model to obtain a three-dimensional evaluation zoning map. Five models, BP, SVC, improved BP1, improved BP2, and improved BP3, are used to classify the airport new city area in the eastern new district of Chengdu. The evaluation grades of BP are 11.52%, 28.92%, 4.76%, and 54.81% respectively; the evaluation grades of SVC are 13.53%, 27.91%, 6.41%, and 52.15% respectively; the evaluation grades of improved BP1 are 11.49%, 29.08%, 9.75%, and 49.68% respectively; the evaluation grades of improved BP2 are 13.58%, 29.64%, 6.97%, and 49.81% respectively; the evaluation grades of improved BP3 are 13.41%, 29.70%, 6.71%, and 50.17% respectively.

[0143] Comparative analysis under multivariate and univariate constraints:

[0144] Different methods were used to carry out a three-dimensional suitability evaluation study under the same constraints of geological environment, disaster environment, ecological environment, and existing resource constraints. In order to study the impact of evaluation under different constraints on the results, the same method was used to carry out an evaluation study on the study area. The same classification standards as those under multiple constraints were used to classify only under the constraints of engineering geological environment (see Table 7).

[0145] Table 7

[0146]

[0147]

[0148] It shows that the improved BP3 is superior to other methods. Therefore, this method is used to conduct three-dimensional evaluation of the study area.

[0149] The total volume of the evaluation area analyzed and evaluated using the improved BP3 model only under the engineering geological environment is 71561318753.387m 3 The volumes of level 1, 2, 3 and 4 are: 9725085523.87m 3 ,21265590181.87m 3 ,7300521642.11m 3 ,33270121405.53m 3 ; The proportions of level 1, level 2, level 3 and level 4 are 13.59%, 29.72%, 10.20% and 46.49% respectively.

[0150] The total volume of the evaluation area is 71561318753.3873m based on the evaluation zoning analysis using the improved BP3 model under multivariate constraints. 3 , where the volumes of level 1, 2, 3 and 4 are: 9695136288.506m 3 ,21350905605.796m 3 ,4611591017.926m 3 ,35903685841.159m 3 ; The proportions of level 1, level 2, level 3 and level 4 are 13.55%, 29.84%, 6.44% and 50.17% respectively.

[0151] like Figure 3 As shown in the figure, the differences between the evaluation under multiple constraints and the single constraint are mainly distributed in Level 3 and Level 4, among which the difference in Level 3 is 2495.67 million m 3 , the difference between the 4 levels is 2635.03 million m 3 , the difference was 3.49% and 3.68% respectively. This is mainly due to the increase in the proportion of unsuitable areas through the introduction of ecological forest protection in the ecological environment, permanent basic farmland protection, and restrictions on underground tracks and cultural relics in the existing resource protection.

[0152] On-site field verification:

[0153] The above analysis shows that the improved BP3 method is more reasonable and efficient than other traditional methods. Therefore, this method is used to further divide the study area into suitable zoning areas and verify it with the actual situation on the spot.

[0154] Poor suitability: Based on the improved BP3 method, the 4th level is mainly concentrated near the Longquan Mountain belt, such as Anquan Village, Baoan Village, Guojiayan Village, Qingsong Village, Chenba Village, Ping'an Village, Hongyanzi Village, Jianguo Village, Xinjiazui Village, Zhuanfang Village, Zhangjiagou Village, Baishu Village, Shiheyan Village, and Qingshuiyan Village. These areas are mainly affected by the terrain, such as the large height difference and slope, and the influence of the Ma'anshan Fault, Sancha Lake Fault and Jiajiachang Syncline nearby, and there are also many landslides and collapses. At the same time, they are also restricted by the protection of Sancha Lake and ecological forest land. In addition, due to the influence of weathering, the physical and mechanical properties of some bedrock outcrops are relatively low, which has a great impact on the development of underground space, and the development cost is also relatively high. These areas are not suitable for large-scale and block development, and can be developed in a point-like or linear manner. Si'er Village, Jinjiaba Village, Dingshui Village and other areas are affected by Metro Line 18, so the protection of underground resources should be strengthened during development.

[0155] General suitability area: mainly concentrated in the shallow surface parts of Mast Dam Village, Wangjia Village, Dagu Village and other places in the northeast. Since most of these places are covered with thin layers of soil, such as (silty) clay, plain fill, cultivated soil, silt and silty soil, the soil compression coefficient is 0.14-0.45, and the average value is 0.21-0.27. According to the specification, the compression characteristics are medium. The compressive strength and integrity coefficient of some strongly weathered bedrocks such as silty mudstone and mudstone are relatively low. Moreover, mudstone is easily softened and disintegrated by surface water, resulting in reduced mechanical properties, which greatly affects the stability of the project. However, it is less affected by landslides, collapses, lakes, reservoirs, and faults. Therefore, surface drainage measures need to be taken during the excavation process, and the slopes need to be supported in time.

[0156] Suitable and better area: This area is mainly distributed below the shallow surface, in the sub-shallow layer, the middle layer and some deep layers, and is less affected by landslides, collapses and faults. These areas are mainly slightly weathered mudstone, shale and some sand and mudstone interlayers. The bedrock stability is relatively high and can be used as the bearing capacity of the foundation. However, under the influence of groundwater, mudstone softens and expands when it encounters water, and its physical and mechanical properties are reduced, affecting the stability of the slope and cavern. When developing underground space, precipitation treatment should be carried out and waterproofing measures should be taken in time. At the same time, the infiltration of the main surface water is also required to cause the groundwater level to rise.

[0157] Good suitability area: This area is mainly distributed in the sub-shallow, middle and deep layers and is basically not affected by landslides, collapses and faults. These areas are mainly composed of micro-differentiated sandstone, sandstone (sand and mudstone interlayers), siltstone (mainly sandstone) and other rock masses with relatively high saturated compressive strength and integrity coefficient, which have good stability for the development of underground space. However, since the interlayers of mudstone and sandstone are affected by groundwater, it is easy to cause certain impacts on underground projects, so precipitation treatment is also required. These areas can be developed in a comprehensive manner such as points, lines, and surfaces.

[0158] In view of the characteristics of massive urban underground space suitability evaluation data and low evaluation efficiency, a supervised learning urban underground space development suitability evaluation method is proposed by combining the improved fuzzy hierarchical synthesis method with machine learning. The main conclusions are as follows:

[0159] (1) In view of the fact that the Red-tailed Hawk Algorithm has insufficient global exploration ability and is prone to falling into local optimal solutions, a multi-effect improved Red-tailed Hawk Algorithm is proposed. The improved Red-tailed Hawk Algorithm is tested on the benchmark function and the results show that the average values ​​of the improved Red-tailed Hawk Algorithm on the eight test functions are the smallest, and the optimal values ​​on the eight test functions are obtained. In particular, the optimal values ​​obtained on the test functions F1, F3, F4 and F9 are closer to the standard values ​​than those of the other algorithms.

[0160] (2) Since the weights and biases of the traditional BP neural network are random, and the weights and biases generated by this randomness affect the convergence speed and accuracy of the model, the improved red-tail algorithm is used to improve the traditional BP neural network. The improved fuzzy hierarchical synthesis method is combined with supervised learning to construct five supervised learning evaluation models: BP, SVC, improved BP1, improved BP2, and improved BP3.

[0161] (3) The different BP neural network layers, training times, and number of neurons used also have a great impact on the results. Therefore, a quantitative analysis was conducted on different network layers, training times, and number of neurons to construct the optimal BP neural network model.

[0162] (4) A supervised learning urban underground space suitability evaluation system (DUUSS-1) was developed using Python, Web systems, JavaScript languages, etc.

[0163] (5) Based on the idea of ​​block assignment and block evaluation of the three-dimensional geological model, five supervised learning evaluation models and the developed evaluation system were used to carry out a three-dimensional evaluation of the suitability of the Urban Airport New Town in the Eastern New District of Chengdu. The accuracies of the five models, BP, SVC, improved BP1, improved BP2, and improved BP3, were 89.21%, 88.64%, 91.48%, 94.83%, and 99.10%, respectively. The results showed that improved BP3 had the highest accuracy. The three-dimensional suitability divisions of the five model evaluation grades were: BP were 11.52%, 28.92%, 4.76%, and 54.81%, respectively; SVC were 13.53%, 27.91%, 6.41%, and 52.15%, respectively; improved BP1 were 11.49%, 29.08%, 9.75%, and 49.68%, respectively; improved BP2 were 13.58%, 29.64%, 6.97%, and 49.81%, respectively; improved BP3 were 13.41%, 29.70%, 6.71%, and 50.17%, respectively.

[0164] (6) The three-dimensional suitability evaluation of the study area was carried out using multivariate constraints and univariate constraints respectively. The results showed that there were differences between the evaluation under multivariate constraints and the evaluation under the single constraint, mainly distributed in level 3 and level 4, with a difference of 3.49% in level 3 and 3.68% in level 4. This is mainly due to the constraints of ecological forest protection, permanent basic farmland protection, and underground rail and cultural relics in the evaluation indicators of ecological environment.

[0165] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for evaluating underground space suitability based on fuzzy hierarchical synthesis method and supervised learning, characterized in that: include: Obtain three-dimensional data of underground space suitability evaluation indicators; Preprocessing the three-dimensional data to obtain standardized evaluation index data; Constructing an improved BP neural network model and an improved fuzzy comprehensive evaluation method, and evaluating and processing the standardized evaluation index data based on the improved fuzzy comprehensive evaluation method to obtain label samples; Based on the label samples, the improved BP neural network model is trained to obtain an optimal model; The suitability classification of urban underground space is carried out based on the optimal model.

2. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 1 is characterized in that: The process of obtaining three-dimensional data of underground space suitability evaluation indicators includes: Construct 3D geological models based on geological data; Obtain underground space suitability evaluation indicators based on quantitative and qualitative analysis; Constructing a three-dimensional evaluation index system based on the three-dimensional geological model and the underground space suitability evaluation index; Corresponding three-dimensional data is obtained based on the three-dimensional evaluation index system.

3. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 2 is characterized in that: The underground space suitability evaluation indicators include height difference, slope, rock characteristics, soil characteristics, surface water impact, groundwater depth, and groundwater corrosiveness.

4. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 1 is characterized in that: The process of constructing the improved fuzzy comprehensive evaluation method includes weight improvement and evaluation grading improvement; The weight improvement process includes constructing variable weights based on the idea of ​​praise and criticism, and optimizing the weights based on the variable weights; When it is a decreasing function, When it is an increasing function, When the indicator weight changes slightly, ω′ k =ω k ±f(x k ) (3≤k≤n-3) In the formula, f(x i ) is a function of the indicator state value, ω1, ω2, ..., ω n Represents the normalized combined weight of the objective weight and the subjective weight.

5. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 4 is characterized in that: The process of improving the evaluation grading includes: introducing validity for evaluation grading, when the validity is less than 0.5, adopting the maximum membership principle for evaluation grading, when the validity is less than 0.5, constructing a credibility mathematical model based on the validity, and performing evaluation analysis based on the credibility mathematical model.

6. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 1 is characterized in that: The process of constructing the improved BP neural network model includes optimizing the BP neural network based on the improved Red-tailed Hawk algorithm.

7. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 6 is characterized in that: The process of improving the red-tailed hawk algorithm includes: improving the high-altitude soaring stage of the red-tailed hawk, improving the low-altitude flying stage of the red-tailed hawk, and improving the stooping stage of the red-tailed hawk; The mathematical expression of the best position for the red-tailed hawk to find food during its high-altitude soaring phase is: X(t)=X best +(X mean -X(t-1)·Levy(dim)·TF(t) In the formula, X(t) represents the position at iteration t, X best The best position obtained, X mean is the average position.

8. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 7 is characterized in that: The process of improving the low-altitude flight stage of the red-tailed hawk includes constructing a reverse solution based on elite reverse learning; wherein, the expression for obtaining a new generation of individuals based on the reverse solution and the current solution is: k is a random number between 0 and 1, a j (t), b j (t) is the interval constructed by the elite population.

9. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and supervised learning according to claim 8 is characterized in that: The process of improving the stoop phase of the Red-tailed Hawk includes adding disturbances to the optimization process using a modified sine-cosine algorithm; Among them, the expression of the improved sine-cosine algorithm is: t is the number of iterations, T max is the maximum number of iterations, ω t is the inertia weight, x i (t) is the i-th position of individual x in the t-th iteration, p i (t) is the component of the i-th position of the best individual position in the t-th iteration, r1 is a decreasing function usually between [0,a], r2 is a random number on the interval [0,2π]], a is a constant, r3 is a random number on the interval [-2,2], and r4 is a random number on the interval [0,1].

Images