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

Abstract

The invention discloses an underground space suitability evaluation method based on a fuzzy hierarchical synthesis method and semi-supervised learning. The method comprises the following steps: acquiring three-dimensional data of underground space suitability evaluation indexes; performing normalization processing on the three-dimensional data to obtain normalized data, and labeling the normalized data through an improved fuzzy comprehensive evaluation method; constructing an initial suitability evaluation model based on an XGBoost model, and training the initial suitability evaluation model through the labeled normalized data to obtain a suitability evaluation model; and urban underground space suitability classification is carried out through the suitability evaluation model. According to the method, the improved fuzzy hierarchical comprehensive evaluation method and the improved XGBoost model are combined to evaluate the suitability of the underground space. The XGBoost model is improved through the improved mantis algorithm, and the defect that optimal combination parameters are difficult to obtain due to the fact that a traditional XGBoost model parameter selection base has randomness can be overcome.

Description

Technical Field

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

[0002] The development and utilization of urban underground space has important advantages in saving resources and reducing environmental pollution, so it is currently being vigorously developed. However, blind development may affect the stability of building foundations, surface cracking, rock and soil instability, and may even damage the ecological environment. Therefore, it is of great significance to conduct a suitability evaluation for the development and utilization of urban underground space before planning. In terms of the perspective and dimension of suitability evaluation, most of the existing results are only two-dimensional evaluations carried out from the perspective of engineering geological environment, and the traditional two-dimensional evaluation has the shortcomings of weak visualization and low evaluation accuracy. Most of the existing evaluation methods are supervised learning evaluation methods. The most prominent shortcoming is that the traditional evaluation method has a large amount of calculation and the data cannot be reused, resulting in low efficiency. Summary of the invention

[0003] 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 semi-supervised learning to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides an underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning, comprising the following steps:

[0005] Collect three-dimensional data of underground space suitability evaluation indicators;

[0006] Normalizing the three-dimensional data to obtain normalized data, and labeling the normalized data using an improved fuzzy hierarchical comprehensive evaluation method;

[0007] Building an initial suitability assessment model based on the XGBoost model, and training the initial suitability assessment model through labeled normalized data to obtain a suitability assessment model;

[0008] The suitability classification of urban underground space is carried out through the suitability assessment model.

[0009] Preferably, the three-dimensional data includes height difference, slope, rock mass characteristics, soil mass characteristics, surface water impact, groundwater depth and groundwater corrosivity.

[0010] Preferably, the method for improving the fuzzy comprehensive evaluation method comprises:

[0011] Based on the principle of praise and criticism, a variable combination mathematical model is established to improve the traditional hierarchical analysis method;

[0012] The effectiveness of the traditional fuzzy comprehensive method in evaluation and grading is analyzed. When the effectiveness fails, credibility is established based on the effectiveness principle, and evaluation and grading are performed based on the credibility.

[0013] Preferably, the method for constructing the initial suitability assessment model includes optimizing the XGBoost model by improving the Mantis algorithm to obtain the initial suitability assessment model.

[0014] Preferably, the method for optimizing the XGBoost model by improving the Mantis algorithm comprises:

[0015] Design the parameter range of the XGBoost model and initialize it to obtain the initialized population;

[0016] By training the initialized population using the XGBoost model, the fitness value of each individual can be obtained;

[0017] The improved mantis algorithm is used to optimize the fitness value and obtain the optimal fitness value;

[0018] A model weight is obtained based on the optimal fitness value, and the XGBoost model is updated by the model weight to obtain an initial suitability evaluation model.

[0019] Preferably, the method for improving the mantis algorithm comprises:

[0020] Use Tent mapping to initialize the Mantis algorithm population;

[0021] The solutions of finding and attacking prey in the mantis algorithm are disturbed through reverse learning;

[0022] The solution of the neutral phagocytosis phase of the Mantis algorithm is perturbed by Gauss-Cauchy mutation.

[0023] Preferably, the expression of the reverse solution in reverse learning is:

[0024]

[0025] In the formula, C ubj and C lbj are the upper and lower bounds of the current t-th iteration mantis individual in the j-th dimension, is the solution of the i-th mantis in the j-th dimension.

[0026] Preferably, the expression of the Gauss-Cauchy variation is:

[0027] X newi (t) = X b (t)×(1+μ1×Cauchy(0,δ)+μ2×Causs(0,δ))

[0028]

[0029] Where, X b (t) is the position of individual X at the tth iteration, X newi (t) is the optimal position of the tth iteration (the position after the Gaussian-Cauchy mixture perturbation), Causs(δ) is the Gaussian mutation operator, Cauchy(δ) is the Cauchy mutation operator, δ is the standard deviation of the Cauchy-Gaussian mutation strategy, μ i is the weight coefficient, k is the adjustment coefficient greater than 0. Through experimental comparison and analysis, k = 3.2, T max is the maximum number of iterations.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] From the perspective of semi-supervised learning, this paper proposes a three-dimensional evaluation method for the suitability of urban underground space by improving XGBoost. The latest mantis algorithm is adopted and improved, and the modified mantis algorithm is optimized to XGBoost, and then an improved XGBoost model is proposed. The key problem of high labeling cost and low efficiency of supervised learning is solved. At the same time, it also provides new ideas for the evaluation of urban underground space suitability and has significant engineering practical research value. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] 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:

[0033] Figure 1 This is a flow chart of an underground space suitability assessment method according to an embodiment of the present invention;

[0034] Figure 2 It is a flow chart of the improved fuzzy hierarchical synthesis method evaluation system of an embodiment of the present invention;

[0035] Figure 3 This is a flowchart of the improved XGBoost structure according to an embodiment of the present invention;

[0036] Figure 4 This is a diagram showing the effects of different parameters on the model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] 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.

[0038] 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.

[0039] Embodiment 1

[0040] like Figure 1 As shown, in this embodiment, a method for evaluating underground space suitability based on fuzzy hierarchical synthesis method and semi-supervised learning is provided, comprising the following steps:

[0041] Collect three-dimensional data of underground space suitability evaluation indicators;

[0042] Normalize the three-dimensional data to obtain normalized data, and label the normalized data using an improved fuzzy comprehensive evaluation method;

[0043] An initial suitability assessment model is constructed based on the XGBoost model, and the initial suitability assessment model is trained using labeled normalized data to obtain a suitability assessment model;

[0044] The suitability classification of urban underground space is carried out through the suitability assessment model.

[0045] Furthermore, the three-dimensional data includes elevation difference, slope, rock mass characteristics, soil mass characteristics, surface water impact, groundwater depth and groundwater corrosivity.

[0046] Further, the method of improving the fuzzy comprehensive evaluation method includes:

[0047] Based on the principle of praise and criticism, a variable combination mathematical model is established to improve the traditional hierarchical analysis method;

[0048] The effectiveness of the traditional fuzzy comprehensive method in evaluation and grading is analyzed. When the effectiveness fails, credibility is established based on the effectiveness principle, and evaluation and grading are performed based on the credibility.

[0049] Furthermore, the present invention adopts the idea of ​​praise and criticism to construct variable weights. Assume that the weights obtained by the 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:

[0050]

[0051] 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:

[0052]

[0053] In addition, some indicators may have small changes in weight, in which case they can be set as:

[0054] ω k ′=ω k ±f(x k ) (3≤k≤n-3) (3)

[0055] 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.

[0056] 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.

[0057] The improved fuzzy hierarchical comprehensive method evaluation system is as follows: Figure 2 shown.

[0058] Furthermore, the method for constructing an initial suitability assessment model includes optimizing the XGBoost model by improving the Mantis algorithm to obtain the initial suitability assessment model.

[0059] The specific method is as follows:

[0060] The mantis algorithm simulates the hunting and sexual cannibalism of mantises. The mantis algorithm consists of four stages: the first stage is the population initialization stage, in which the population is randomly distributed in the search space; the second stage is the exploration stage (finding prey), which imitates the mantis's behavior of finding prey; the third stage is the development and utilization stage, which imitates the mantis's behavior of attacking and catching prey; the fourth stage is the sexual cannibalism stage. Although the mantis algorithm has the advantages of high accuracy and fast convergence speed, it lacks global exploration capabilities and is prone to falling into local optimal solutions. Therefore, an improved mantis algorithm is proposed, which is mainly improved from three stages.

[0061] The initial population of the mantis algorithm is random, and the search for solutions by this randomly distributed population is also random, which will affect the search space, so chaotic mapping is used to initialize the population. This embodiment uses Tent mapping to initialize the mantis algorithm population, and its formula is as follows:

[0062]

[0063] (1) Population initialization stage

[0064]

[0065] and represents the upper and lower bounds of the j dimension; is a vector of random numbers between 0 and 1 according to a uniform distribution.

[0066] (2) Searching for prey (exploration phase): The exploration phase includes two behaviors: hiding, ambush, and pursuit.

[0067] (a) Pursuit and predation

[0068]

[0069] Formula (6) combines the Levy flight t1 and the normally distributed random number t2 to update the mantis. represents the position of the ith mantis (solution) at the function t; · represents the product of two numbers; * represents the product of two vectors, represents a numerical vector generated based on the Levy flight strategy, |τ2| is a random number with a standard deviation of 1 and a mean of 0, and r1 and r2 are numbers randomly generated between 0 and 1 according to a uniform distribution. is a vector of randomly generated numbers in the interval (0,1) according to a uniform distribution. is a solution randomly selected from the current population such that is a binary vector generated based on the following formula:

[0070]

[0071] and is a vector randomly generated from a uniform distribution in the interval 0-1.

[0072] (b) Hidden ambush hunting

[0073]

[0074] α is a factor that controls the position of the mantis's head to cover the ambush distance. The mathematical formula of this coefficient is as follows:

[0075] α=cos(πr6)·μ (9)

[0076] r6 is a random number between 0 and 1, μ is the distance factor, and the formula is as follows:

[0077]

[0078] T is the maximum number of function evaluations, simulating the behavior of a prey moving in an environment in search of prey. The behavior of this prey to ambush distance is determined by the following formula:

[0079]

[0080] and They represent the upper and lower bounds of the dimension in the problem being processed, respectively, and r7 is a randomly generated number between 0 and 1. A randomly generated number vector between 0 and 1 according to a uniform distribution. represents a solution randomly selected from the archive, which is used to represent the position of the i-th mantis. Formula (9) At the beginning of the optimization iteration, the distance between the camouflaged place and the prey is large. As the number of current iterations increases, this distance will gradually decrease because it will move towards the mantis. The ambush behavior of the mantis and its prey is mathematically formulated as follows:

[0081]

[0082] R9 and R 10 is a randomly selected value uniformly distributed between 0 and 1 to achieve a trade-off between the ambush behavior and the prey behavior of the mantis; the behaviors of the pursuer and the spearer are integrated into the proposed optimizer by using the loop control factor, which divides the optimization process into multiple parts, each of which helps to explore the possible search space of the optimization problem. Its formula is as follows:

[0083]

[0084] % represents the remainder operator, also known as the modulo operator, and P is an integer (estimated in the experimental section) representing the number of loops used to achieve the trade-off between equations (5) and (11).

[0085] (3) Attacking prey: Prey search phase

[0086] Mantis captures prey in two stages: approach and capture. The relevant parameters are as follows

[0087] d s Estimated strike distance, v s Calculate the strike speed,

[0088]

[0089] v s represents the attack speed of the mantis; ρ represents the acceleration of the mantis's impact, a constant value determined in subsequent experiments; l is a number generated between -1 and -2 to control the acceleration of gravity. s When it is close to 0, the mantis finds that it is not the right time to attack the prey. s When it is close to 1, the mantis will move quickly, attack the prey to be captured and eat it before the prey escapes; each time the mantis attacks the prey, it will be updated by the following formula:

[0090]

[0091] represents the new position of the mantis in the j dimension at time t+1, v s Indicates the attack speed of the mantis; is the position of the prey, so as to minimize the distance between the prey and the mantis; The attack distance varies according to the size of the mantis. The larger the mantis, the greater the attack distance. Calculated according to the following formula:

[0092]

[0093] represents the jth dimension of the prey or the best solution obtained so far; represents the current position of the i-th mantis in the j-th dimension; when When far away from prey, the attack distance is large, and vice versa, the attack distance is small.

[0094] When the mantis fails in its attack, it needs to change direction and attack again. The formula is as follows: The mantis changes direction to attack based on the improved direction selected by randomly selecting two mantises from the previous group.

[0095]

[0096] and are two mantises randomly selected from the current population; r 12 It is based on a random number uniformly distributed between 0 and 1. The failure of the mantis to attack means that it has fallen into a local optimal solution. Individuals trapped in a local optimal solution need significant exploration and utilization capabilities to escape the optimal solution. The following mathematical model is proposed to update the mantis to take a better position to attack the prey again to prevent falling into the local optimal solution:

[0097]

[0098] r 13It is a randomly generated value between 0 and 1 according to a uniform distribution; Formula 9-18 is used together with the failure rate to avoid falling into the local minimum and to improve the convergence speed of the optimal solution of the optimization problem; the probability of Formula 9-19 gradually decreases with the increase of the current function evaluation to reduce the maximization search process at the beginning of the optimization process, and gradually increases the utilization operator to accelerate the convergence to the optimal solution.

[0099]

[0100] a is a predefined fixed value between 0 and 1, which is used to control the exploration and mining operations. The larger a is, the more exploration can be increased and the more mining can be reduced, and vice versa. Formula (15) and Formula (17) are randomly exchanged within MSA, r2 is a random number in the interval 0-1, and for all dimensions in each solution, r2 is a constant, and r4 is a randomly generated number for each dimension in the updated solution.

[0101] 2.0 uses reverse learning to improve:

[0102] In the second and third stages of the Mantis algorithm, it is easy to fall into the local optimal solution, which reduces the accuracy of the solution. Therefore, reverse learning is used to perturb the solution to jump out of the local optimal solution. Suppose the reverse solution of the Mantis algorithm solution generated by direction learning is

[0103] is a feasible solution to the problem of generation t, is the solution of the i-th mantis in the j-th dimension, yes The corresponding inverse solution is The expression is:

[0104]

[0105] Among them C ubj and C lbj are the upper and lower bounds of the mantis individual in the jth dimension at the current tth iteration, i∈[1,N], N is the number of mantis individuals, j∈[1,D], D is the spatial dimension of feasible solutions.

[0106] (4) Sexual cannibalism: Female mantises eat males during or after mating. This behavior is more sexual cannibalism. First, females attract their mates to attract males to their locations. The formula is:

[0107]

[0108] Represents a female praying mantis; One solution, chosen at random, represents a male being attracted to a female for mating and then being eaten; is a vector representing an attraction factor based on a random number uniformly distributed between [0,1]; mate attraction is performed at the beginning of the optimization process, with the goal of attracting males to the position of females. Therefore, it is very likely to be performed at the beginning of the optimization process, and the probability gradually decreases as the population increases. The mathematical formula for this probability, namely Pt, is the probability function:

[0109] P t =r 17 ·μ (22)

[0110] r 17 According to the random numbers uniformly distributed between [0,1], the male mates with the female through the uniform crossover operator in the genetic operator to produce new offspring. The formula is as follows:

[0111]

[0112] is a vector of random numbers uniformly distributed between [0,1]; represents the value of the first dimension of the first mantis, which represents the male mantis mating with the female. This formula is established to cover dimensions where the loading converges within different dimensions.

[0113]

[0114] represents males; μ is the portion of males eaten; cos(2lπ) is used to give females the flexibility to turn males during feeding; male behavior focuses on reducing the risk of females provoking aggression, while several males reduce the risk of being eaten by their own kind.

[0115] 3.0 uses Gaussian Cauchy improvement:

[0116] At the end of the Mantis algorithm, it often falls 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 and achieve the effect of global development. It is necessary to perturb the last stage to jump out of the local area. The perturbation methods often used are Cauchy mutation, differential mutation, Gaussian mutation, etc. Because the search range of Cauchy mutation is larger than that of Gaussian, it has better global search ability, but too large a step size is easy to jump out of the optimal solution and produce a poor solution. Compared with Cauchy mutation, Gaussian mutation has better search ability in a small range. Therefore, the two are organically combined to propose a perturbation method of Gauss-Cauchy mutation. Therefore, the formula for using Gauss-Cauchy for perturbation is as follows:

[0117] X newi (t) = Xb (t)×(1+μ1×Cauchy(0,δ)+μ2×Causs(0,δ)) (25)

[0118]

[0119] The Cauchy density function formula is:

[0120]

[0121] Where, X b (t) is the position of individual X at the tth iteration, X newi (t) is the optimal position of the tth iteration (the position after the Gaussian-Cauchy mixture perturbation), Causs(δ) is the Gaussian mutation operator, Cauchy(δ) is the Cauchy mutation operator, δ is the standard deviation of the Cauchy-Gaussian mutation strategy, μ i is the weight coefficient, k is the adjustment coefficient greater than 0. Through experimental comparison and analysis, k = 3.2. From the formula, it can be concluded that the early stage is mainly based on Cauchy mutation search, and the later stage is mainly based on Gaussian mutation.

[0122] Eight internationally used benchmark functions were selected for comparative analysis using particle swarm optimization (PSO), northern goshawk algorithm (NGO), gray wolf optimization algorithm (GWO), mantis algorithm (MSA), genetic algorithm (GA), and improved mantis algorithm (CMSA). The population size of the six algorithms was set to 30, and the maximum number of iterations was 100.

[0123] From the analysis of Table 1, it can be concluded that the mean and optimal values ​​of the improved mantis algorithm on the benchmark test functions F1 to F8 are basically better than those of PSO, NGO, WGO, GA, and MSA algorithms. Only the mean on the test function F1 is higher than that of PSO, but the optimal values ​​are better than those of the other five algorithms. The results further prove that the CMSA algorithm is superior. The above results show that the use of the multi-effect improved mantis algorithm is reliable and correct. The verification of the algorithm of the present invention verifies the improvement of the accuracy of urban underground space evaluation by subsequent semi-supervised learning.

[0124] Table 1

[0125]

[0126] Analysis of semi-supervised learning evaluation methods:

[0127] Semi-supervised learning is a key issue in the field of pattern recognition and machine learning. It is a learning method that combines supervised learning with unsupervised learning. The most prominent advantage of semi-supervised learning is to use a small number of labeled samples to predict a large number of unlabeled samples. At present, the classification methods of semi-supervised learning mainly include Random Forest (RF), XGboost, and Support Vector Machine. Random Forest (RF): Based on the bootstrap related theory, Breiman combined decision trees to establish the RF model. RF extracts multiple samples from the original sample based on the bootstrap resampling method, and bootstraps each sample to model the decision tree. Then, the constructed decision trees are combined and the final classification or prediction value is obtained by voting. XGBoost (ExtremeGradient Boosting) is an efficient gradient boosting framework, which is improved based on the GradientBoosting Decision Tree (GBDT) algorithm. XGboost, support vector machine, and clustering method are compared and analyzed in terms of evaluation effect and efficiency, and XGboost is selected as the semi-supervised evaluation method.

[0128] Improve XGboost evaluation model construction:

[0129] XGBoost is a machine learning algorithm based on the gradient boosting framework, which has the advantages of fast speed, ability to process big data, and high accuracy. Since XGBoost has many parameters and different parameters have different effects on the performance of the model, its main parameters are as follows:

[0130] The learning rate (learning_rate) controls the adjustment of the weights at each iteration. Different weights have an important impact on the model. A smaller learning rate can improve the stability of the model, but it will also make the model converge more slowly. On the contrary, a larger learning rate may speed up the convergence of the model, but it may also cause overfitting.

[0131] The number of trees (n_estimators) is also called the number of iterations. The number of different trees has an important impact on the model. Increasing the number of this value can improve the performance of the model but also reduce the efficiency.

[0132] Maximum depth of the tree (max_depth): The maximum depth of the decision tree, which controls the complexity of the tree and avoids overfitting. Generally speaking, the larger the value, the more complex the model is and the better it fits the training data, but too high a value may also lead to overfitting.

[0133] Subsample ratio: The ratio of randomly selected samples used to train each tree. If the subsample ratio is set too small, overfitting can be prevented, but it may also reduce the performance of the model.

[0134] The feature subsampling ratio of each tree (colsample_bytree) is used to randomly extract the feature ratio when training each tree. A smaller feature subsampling ratio can increase the performance of the model.

[0135] Minimum loss reduction required for node splitting (gamma): Controls the condition for node splitting. Node splitting occurs only when the loss reduction brought by node splitting is greater than gamma.

[0136] from Figure 4 (a) The analysis shows that different tree depths have a great influence on the model. When the tree depth is 2, the minimum accuracy is 0.6579, and when the value is 12, the maximum accuracy is 0.7991. When the value is 12, it does not decrease but increases very little, so the range of this value is between 2-12. Figure 4 (b) It is found that different learning rate values ​​also have a great impact on the model. When the learning rate is 0.01, the accuracy is the lowest at 0.6382. When the learning rate increases to 0.5, the maximum value is 0.9705. When the value is 0.4 and 0.5, the increase is very small. Therefore, 0.01-0.5 is taken as the range of this value. Figure 4 (c) It is concluded that the number of iterations also has an impact on the model. When the number of iterations is 20, the accuracy is the lowest at 0.8327. As the number of iterations increases, the accuracy stops increasing at 120, so its range is between 20 and 120. Figure 4 (d) The minimum value is 0.7196 when the sample proportion is 0.15, and the maximum value is 0.7233 when the sample proportion is 0.45 and 0.9, so the range of this value is 0.15-0.9. Figure 4 (e) It is concluded that different feature proportions have a certain impact on the model. When the sub-sample proportion is 0.1, the lowest accuracy is 0.7192, when the value is 0.7, the maximum value is 0.9617, and when the value is 1, the accuracy is lower than the maximum value, so the value range is 0.1-1. Figure 4 (f) It is found that the accuracy remains unchanged when different loss thresholds are taken, indicating that the loss threshold has little impact on the model, so this parameter is not used as an indicator for model optimization.

[0137] In semi-supervised learning, different label ratios have a certain impact on the results. Therefore, the label ratio is set to 0.2, 0.3, 0.4, 0.5, and 0.7 respectively. As the label ratio increases, the accuracy also increases, but when it increases to 0.4, the growth is relatively slow. Therefore, considering the time efficiency, labeling cost and accuracy, the label ratio is set to 0.4 in this experiment.

[0138] From the above content, we can conclude that setting different parameters has a great impact on the performance of the XGBoost model, and it is difficult to find the optimal combination of these parameters based on human experience. Therefore, the Mantis algorithm and the improved Mantis algorithm proposed in the previous article are used to optimize these parameters to obtain the optimal combination of parameters. The main technical flow chart is shown in Figure 3 shown.

[0139] The model of this embodiment is verified as follows:

[0140] First, the sample data is standardized, and then the data is designed with a ratio of 0.4 between labeled and unlabeled data according to the analysis in the previous article. Secondly, the labeled 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 evaluation.

[0141] The Grey Wolf Algorithm, Mantis Algorithm and Improved Mantis Algorithm were used to optimize the hyperparameters of XGBoost, and then an improved XGBoost model was proposed. The improved models are called Improved XGBoost-1 (WGO-XGBoost), Improved XGBoost-2 (MSA-XGBoost), and Improved XGBoost-3 (CMSA-XGBoost). At the same time, in order to verify the reliability of the model, the traditional XGBoost model and the random forest (RF) model were used for comparative analysis. These models were used to classify the suitability of the development and utilization of the urban underground space in the Airport New Town of the Eastern New District of Chengdu, and the results are shown in Tables 2, 3, 4, 5, and 6 respectively.

[0142] Table 2

[0143]

[0144] The results of the suitability evaluation of the development and utilization of urban underground space in Airport New Town, Eastern New District, Chengdu, using improved XGBoost-1 are shown in the following table:

[0145] Table 3

[0146]

[0147] The results of the suitability evaluation of the development and utilization of urban underground space in the Airport New Town of Chengdu Eastern New District using the improved XGBoost-2 are shown in the following table:

[0148] Table 4

[0149]

[0150] The results of the suitability evaluation of the development and utilization of urban underground space in Airport New Town, Eastern New District, Chengdu, using improved XGBoost-3 are shown in the following table:

[0151] Table 5

[0152]

[0153] The results of the suitability evaluation of the development and utilization of urban underground space in the Airport New Town of Chengdu Eastern New District using random forest (RF) are shown in the following table:

[0154] Table 6

[0155]

[0156]

[0157] The above results show that the evaluation results of different methods are quite different. In order to quantitatively analyze the differences among these methods, the results of XGBoost, Random Forest (RF), Improved XGBoost-1, Improved XGBoost-2, and Improved XGBoost-3 were statistically analyzed. The classification results of the five models were 85.40%, 82.13%, 87.52%, 91.35%, and 96.26%, respectively. The classification results of XGBoost were close to those of RF, which verified that the constructed XGBoost model was reliable, and the classification accuracy of the improved XGBoost model was higher than that of the traditional XGBoost model and also better than the RF model. The differences among the five evaluation models were 2.12%, 3.83%, 4.91%, 10.86%, and 14.13%, respectively. The results showed that the improved XGBoost-3 had the highest accuracy, which was 10.86% higher than the traditional XGBoost, 8.74% higher than the improved XGBoost-1, 4.91% higher than the improved XGBoost-2, and 14.13% higher than the random forest. Although the accuracy of improved XGBoost-2 is lower than that of improved XGBoost-3, it is higher than that of XGBoost improved by Grey Wolf Algorithm, traditional XGBoost and random forest model. It can be seen from the above that the model of the present invention can achieve the best effect.

[0158] 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 semi-supervised learning, characterized in that: The following steps are involved: Collect three-dimensional data of underground space suitability evaluation indicators; Normalizing the three-dimensional data to obtain normalized data, and labeling the data using an improved fuzzy hierarchical comprehensive evaluation method; Building an initial suitability assessment model based on the XGBoost model, and training the initial suitability assessment model through labeled normalized data to obtain a suitability assessment model; The suitability classification of urban underground space is carried out through the suitability assessment model.

2. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 1 is characterized in that: The three-dimensional data includes elevation difference, slope, rock mass characteristics, soil mass characteristics, surface water impact, groundwater depth and groundwater corrosivity.

3. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 1 is characterized in that: The methods to improve the fuzzy comprehensive evaluation method include: Based on the principle of praise and criticism, a variable combination mathematical model is established to improve the traditional hierarchical analysis method; The effectiveness of the traditional fuzzy comprehensive method in evaluation and grading is analyzed. When the effectiveness fails, credibility is established based on the effectiveness principle, and evaluation and grading are performed based on the credibility.

4. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 1 is characterized in that: The method for constructing an initial suitability assessment model includes optimizing the XGBoost model by improving the Mantis algorithm to obtain the initial suitability assessment model.

5. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 4 is characterized in that: The method for optimizing the XGBoost model by improving the Mantis algorithm includes: Design the parameter range of the XGBoost model and initialize it to obtain the initialized population; By training the initialized population using the XGBoost model, the fitness value of each individual can be obtained; The improved mantis algorithm is used to optimize the fitness value and obtain the optimal fitness value; A model weight is obtained based on the optimal fitness value, and the XGBoost model is updated by the model weight to obtain an initial suitability evaluation model.

6. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 4 is characterized in that: The method for improving the mantis algorithm comprises: Use Tent mapping to initialize the Mantis algorithm population; The solutions of finding and attacking prey in the mantis algorithm are disturbed through reverse learning; The solution of the neutral phagocytosis phase of the Mantis algorithm is perturbed by Gauss-Cauchy mutation.

7. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 6 is characterized in that: The expression of the reverse solution in reverse learning is: In the formula, C ubj and C lbj are the upper and lower bounds of the current t-th iteration mantis individual in the j-th dimension, is the solution of the i-th mantis in the j-th dimension.

8. The underground space suitability assessment method based on fuzzy hierarchical synthesis method and semi-supervised learning according to claim 6 is characterized in that: The expression of the Gauss-Cauchy variation is: X newi (t)=X b (t)×(1+μ1×Cauchy(0,δ)+μ2×Causs(0,δ)) Where, X b (t) is the position of individual X at the tth iteration, X newi (t) is the optimal position of the tth iteration, Causs(δ) is the Gaussian mutation operator, Cauchy(δ) is the Cauchy mutation operator, δ is the standard deviation of the Cauchy-Gaussian mutation strategy, μ i is the weight coefficient, k is the adjustment coefficient, T max is the maximum number of iterations.

Images