Loess area highway engineering construction risk evaluation method based on T-S fuzzy neural network

Abstract

The invention discloses a loess area highway engineering construction risk evaluation method based on a T-S fuzzy neural network. The loess area highway engineering construction risk evaluation method comprises the steps of obtaining a sample data set of a highway engineering construction risk source; a risk evaluation model is constructed, the risk evaluation model adopts a T-S fuzzy neural network, the risk evaluation model is optimized through a particle swarm optimization algorithm according to the sample data set, and the optimized risk evaluation model is obtained; and obtaining highway engineering construction risk data in real time, and performing risk judgment on the highway construction risk data through the optimized risk evaluation model to obtain a highway engineering construction risk level.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of highway engineering construction, and particularly relates to a highway engineering construction risk evaluation method in loess areas based on a T-S fuzzy neural network. BACKGROUND

[0002] Highway engineering construction is usually started at the same time in multiple sections, and the construction environment is complex, especially in loess areas, and the construction operation is more easily affected by factors such as topography and geological structure, natural disasters, and uneven settlement. At the same time, highway engineering construction activities are also easily affected by factors such as construction unit personnel composition, technical level, and management level. Research on the highway engineering construction risk evaluation method is an effective way to ensure personnel safety and reduce accident losses. Through effective identification, analysis, and evaluation of the risks existing in each scene of highway engineering construction in the construction phase, it is beneficial to take targeted risk control measures on this basis, so as to achieve the purpose of reducing the probability of accidents and reducing the loss of accident consequences, and improve the safety of highway engineering construction activities, thereby guiding and assisting highway engineering construction.

[0003] The existing highway engineering construction evaluation objects are mostly selected according to sub-items, and the construction risk evaluation methods are mostly based on qualitative analysis or semi-quantitative analysis. The establishment of the evaluation system lacks the support of quantitative data, and the evaluation process and results may have certain subjective one-sidedness. In addition, it still needs a lot of work to realize the real-time combination of risk evaluation technology and field management. In summary, different highway engineering construction risks cannot be effectively evaluated. SUMMARY

[0004] To solve the above technical problems, the application provides a highway engineering construction risk evaluation method in loess areas based on a T-S fuzzy neural network to solve the problems existing in the prior art.

[0005] To achieve the above purpose, the application provides a highway engineering construction risk evaluation method in loess areas based on a T-S fuzzy neural network, which comprises:

[0006] Obtaining a sample data set of highway engineering construction risk sources;

[0007] Constructing a risk evaluation model, wherein the risk evaluation model adopts a T-S fuzzy neural network, and the risk evaluation model is optimized by a particle swarm optimization algorithm according to the sample data set, to obtain an optimized risk evaluation model;

[0008] Real-time acquisition of highway engineering construction risk data, risk judgment of the highway construction risk data by the optimized risk evaluation model, and obtaining of a highway engineering construction risk grade.

[0009] Optionally, the sample data set acquisition process comprises:

[0010] The static data and real-time data of the highway engineering construction in the loess area are acquired, a risk source database of the highway engineering construction is constructed according to the static data and real-time data, a risk evaluation index system of the highway engineering construction in the loess area is constructed, the risk source database is sorted according to the risk evaluation index system, and a sample data set is obtained.

[0011] Optionally, the risk evaluation index system construction process comprises:

[0012] The data in the risk source database is decomposed step by step by using the WBS-RBS method, wherein the highway engineering work in the risk source database is decomposed step by step to obtain a highway engineering work decomposition structure, the highway engineering construction risk in the loess area in the risk source database is decomposed to obtain a highway engineering construction risk structure, the terminal risk factors of the highway engineering construction risk structure and the terminal project units of the highway engineering work decomposition structure are coupled and matched two by two, and a risk evaluation index system is constructed according to the matching result.

[0013] Optionally, the sample data set acquisition process comprises:

[0014] According to the risk evaluation index system, risk probability level standards, consequence loss level standards and risk level standards are acquired, the risk evaluation indexes and risk levels in the risk source database are quantified according to the risk probability level standards, the consequence loss level standards and the risk level standards, the quantified data are processed by using a principal component analysis method, and a sample data set is obtained.

[0015] Optionally, the risk evaluation index system comprises primary indexes and secondary factors, wherein the primary indexes comprise engineering geology, natural environment, design construction and construction management, the secondary factors of the engineering geology comprise surrounding rock grade, water enrichment condition, geological structure and non-uniform settlement, the secondary factors of the natural environment comprise annual average precipitation, topography, vegetation coverage, vegetation matching type and surface water condition, the secondary factors of the design construction comprise parameter design, geological exploration condition, excavation method, support timeliness and construction defect, and the secondary factors of the construction management comprise site management, safety training, emergency plan and engineering water conservation measures.

[0016] Optionally, in the risk assessment model, risk indexes in the risk assessment index system are taken as input data, membership degrees of the risk indexes are calculated according to fuzzy rules, the membership degrees are calculated by using a multiplication algorithm to obtain weighting coefficients, the weighting coefficients are de-fuzzied to obtain weight values of different attributes, the input data are processed through a full connection layer to obtain output results of a consequent network, and the weight values of the different attributes and the output results of the consequent network are weighted to obtain a highway engineering construction risk grade.

[0017] Optionally, the process of optimizing the risk assessment model by using the particle swarm optimization algorithm comprises the following steps.

[0018] Algorithm parameters of the particle swarm optimization algorithm are initialized, and network parameters of the risk assessment model are initialized, the initialized network parameters of the risk assessment model are taken as initial particles of the particle swarm optimization algorithm, and the network parameters comprise membership function centers, widths and network weights;

[0019] The mean square error of the risk assessment model is taken as a fitness function, and minimization of a training error of the risk assessment model trained according to a sample data set is taken as a target;

[0020] The initial particles are iteratively optimized by using the particle swarm optimization algorithm according to the target and the fitness function, and when a maximum iteration number or the fitness function converges, final particle values of the optimized risk assessment model are obtained.

[0021] Optionally, in the iterative optimization process, an adaptive inertia weight is introduced in a particle swarm updating process, a mutation strategy and a particle hierarchical updating strategy are used, and a particle with a local optimal solution less than a certain threshold is disturbed by introducing Gaussian noise.

[0022] Compared with the prior art, the present application has the following advantages and technical effects:

[0023] The present application realizes accurate quantification and real-time evaluation of highway engineering construction risks in loess areas by constructing a risk assessment model fusing a T-S fuzzy neural network and a particle swarm optimization algorithm. The method effectively overcomes the problems of strong subjectivity and insufficient data support in traditional qualitative or semi-quantitative evaluation, significantly improves the objectivity and accuracy of risk assessment, and greatly improves the risk response speed and site management efficiency with the help of real-time data input and model dynamic optimization mechanism, thereby providing scientific and reliable safety protection for highway engineering construction under complex geological conditions in loess areas. BRIEF DESCRIPTION OF DRAWINGS

[0024] The accompanying drawings, which form a part of the present application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the present application and together with the description serve to explain the present application. In the drawings:

[0025] Figure 1 A method flowchart of an embodiment of the present application;

[0026] Figure 2 A risk source database schematic diagram of an embodiment of the present application;

[0027] Figure 3 An index system schematic diagram of an embodiment of the present application;

[0028] Figure 4 A neural network structure diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0029] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0030] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0031] The present application discloses a loess area highway engineering construction risk evaluation method based on a T-S fuzzy neural network, and belongs to the technical field of highway engineering construction. The steps are as follows: S1, based on historical data and real-time data, a loess area highway engineering construction risk source database is constructed, a risk evaluation index system is established, and sample data sets are obtained; S2, the sample data sets are divided into a training set and a test set, an improved T-S fuzzy neural network model is designed and trained, the input layer is a risk index, the output layer is a risk level, and the particle swarm optimization algorithm (PSO) is used to optimize the neural network parameters, and finally a risk evaluation model is constructed; S3, real-time risk index data is input into the evaluation model, the risk level is determined, and real-time risk assessment is realized. The method effectively reduces the subjectivity in the evaluation process by combining real-time data and historical data, improves the accuracy and response speed of risk assessment, and has important significance for the safety protection of loess area highway construction.

[0032] To overcome the deficiencies in the related art, an embodiment of the present application discloses a loess area highway engineering construction risk evaluation method based on a T-S fuzzy neural network, including the following steps:

[0033] S1, Preparation of sample data set. For highway engineering construction projects in the loess region, based on static data and real-time data, a highway engineering construction risk source database is constructed, a highway engineering construction risk evaluation index system is established, the risk source database is sorted, and a sample data set is obtained;

[0034] Further, as preferred, the preparation of the sample data set in step S1 includes the following steps:

[0035] S1.1, combined with project construction real-time risk source identification collection and historical typical highway engineering construction case risk source data research, a highway engineering construction risk source database in the loess region is constructed;

[0036] S1.2, based on the WBS-RBS method, the highway engineering work is decomposed into unit engineering, sub-item engineering and construction activities, and the highway engineering construction risks in the loess region are decomposed into personnel, mechanical equipment, materials, technology and environment 5 aspects;

[0037] Specifically, the present embodiment is based on the highway engineering work model, and specific highway engineering work needs to be determined during specific decomposition, such as bridge engineering, tunnel engineering, etc. Taking tunnel engineering in highway engineering work as an example, the main construction activities of highway tunnel engineering can be decomposed as follows according to the “Highway Tunnel Engineering Construction Safety Risk Assessment Guide” provided by the Ministry of Transport:

[0038] 1) WBS decomposition of highway tunnel engineering construction. Tunnel engineering is mainly divided into four parts: portal engineering, excavation and support engineering, waterproof and drainage engineering, and lining engineering. It can be seen that the specific construction activities include portal excavation and slope support, tunnel excavation (blasting or machinery), primary support, inverted arch and filling, waterproof and drainage system, secondary lining, and installation of auxiliary facilities such as ventilation, lighting and monitoring.

[0039] 2) RBS risk decomposition dimension. The dimensions include personnel (operation skill, illegal operation, lack of safety awareness), mechanical equipment (failure, failure, inadequate maintenance), materials (specification inconsistency, quality failure, timely supply), technology (design deviation, parameter selection error, insufficient monitoring) and environment (surrounding rock mutation, underground water, gas, climate, collapse).

[0040] S1.3, the end risk factors of the highway engineering construction risk structure are coupled with the end project units of the highway engineering work decomposition structure, the operation process and the risk factors are effectively matched, and a highway engineering construction risk evaluation index system in the loess region is established;

[0041] Highway tunnel engineering WBS-RBS risk decomposition table, as shown in Table 1.

[0042] Table 1

[0043]

[0044] The embodiment takes construction activities as rows and risk factors as columns to establish a WBS-RBS risk matrix. Literature induction is used to determine whether a work unit involves a certain type of risk. Thus, high-risk associated units are extracted from the matrix to form a risk source index library, and an index hierarchical structure is established step by step.

[0045] S1.4, according to the risk probability level standard, the consequence loss level standard and the risk level standard, the risk evaluation index and the risk level contained in the risk source database of highway engineering construction are quantified, the data preprocessing of the database is carried out based on principal component analysis method, and the sample data set is determined.

[0046] Specifically, the risk probability level standard, the consequence loss level standard and the risk level standard refer to the "Guidelines for Safety Risk Assessment of Highway Tunnel Engineering Construction". In the embodiment, the risk probability level standard, the consequence loss level standard and the risk level standard are shown in Tables 2-6, and the interval values are determined according to the actual situation for interpolation scoring. Among them, the risk level quantization standard is shown in Table 2.

[0047] Table 2

[0048]

[0049] Combined with relevant specifications and literature research results, the specific safety risk classification standards and index evaluation criteria of the secondary indexes belonging to the first index engineering geology and natural environment are determined as shown in the table. The secondary indexes belonging to the first index design construction and construction management are all qualitative evaluation indexes, so their evaluation is divided into five levels for description in a qualitative way, and the risk level degree is given a quantitative score according to the table standard. For example: the design parameter index is scored in five levels according to the degree of meeting the specification requirements, using engineering analogy and theoretical analysis to review the parameters, and correcting the parameters according to the site monitoring conditions.

[0050] The engineering geology risk evaluation index quantization standard is shown in Table 3.

[0051] Table 3

[0052]

[0053] The natural environment risk evaluation index quantization standard is shown in Table 4.

[0054] Table 4

[0055]

[0056] The design construction risk evaluation index quantization standard is shown in Table 5.

[0057] Table 5

[0058]

[0059] The construction management risk evaluation index quantization standard is shown in Table 6.

[0060] Table 6

[0061]

[0062] In step S1, the end risk factors of the highway engineering construction risk structure consider the environmental characteristics of the loess area, and the corresponding secondary evaluation indexes include vegetation coverage, vegetation matching type, non-uniform settlement condition and engineering water conservation measures;

[0063] Further, as preferred, the step S1.4 includes the following steps:

[0064] The sample decentralization processing is performed on the risk source database obtained in step S1, the covariance matrix of the sample is calculated, and the first principal components with the cumulative contribution rate greater than 95% are taken as the result of principal component analysis to determine the sample data set.

[0065] S2, construction of a risk evaluation model based on a T-S fuzzy neural network. The sample data set is divided into a training set and a test set in proportion, a fuzzy neural network is designed, the training set is input to continuously train and adjust the parameters of the fuzzy neural network, the parameters of the neural network are optimized using a PSO algorithm, the number of neurons is adjusted, finally the test set is input to the trained neural network to predict the risk level, the effectiveness of the fuzzy neural network is tested, and a risk evaluation model based on a T-S fuzzy neural network is constructed;

[0066] Preferably, the division ratio of the training set and the test set in step S2 is .

[0067] Preferably, in step S2, the risk indicators in the training set and the test set are input parameters, and the risk level is an output parameter. The T-S fuzzy neural network model takes the risk indicators as the input layer and the risk level predicted by the neural network as the output layer. The model input parameters and the output parameters in the training set are used to build a fuzzy neural model, the membership function parameters and the network weights are dynamically optimized by a PSO algorithm, and the model training error is calculated. The test set data is used to verify the risk evaluation model based on the T-S fuzzy neural network, and the model verification error is calculated.

[0068] S3, real-time evaluation of the risk level of highway engineering construction. Real-time highway engineering construction risk data is input into the risk evaluation model established in step S2 to determine the risk level, and real-time evaluation of the risk level of highway engineering construction in the loess area is completed.

[0069] The determination of the risk level in step S3 adopts dynamically adjusted evaluation criteria to evaluate the risk level in combination with changes in real-time data.

[0070] The above technical solutions will be described in detail in combination with the related drawings:

[0071] As shown in the figure, the loess area highway engineering construction risk evaluation method based on the T-S fuzzy neural network provided in the embodiment includes the following steps: Figure 1

[0072] S1, preparation of sample data set. For highway engineering construction projects in loess areas, based on static data and real-time data, a highway engineering construction risk source database is constructed, a loess area highway engineering construction risk evaluation index system is established, the risk source database is sorted and sample data set is obtained;

[0073] Specifically, in step S1, based on typical construction case data and real-time project unmanned aerial vehicle identification data, artificial data, etc., a highway engineering construction risk source database can be constructed as shown in the figure. Figure 2 The horizontal (top) represents the highway engineering work breakdown structure (WBS), i.e. the highway engineering is decomposed into unit engineering, sub-item engineering and specific construction activities, respectively denoted as The vertical (left side) represents the highway engineering construction risk breakdown structure (RBS), i.e. the risk is divided into different categories such as personnel risk, mechanical equipment risk, material risk, technical risk, environmental risk, etc., respectively denoted as The middle matrix area: formed by the intersection of work breakdown units and risk breakdown units, each intersection is represented by , representing a specific construction activity corresponding to a certain type of risk.

[0074] In the embodiment, 25 groups of highway tunnel construction risk source data in special geological areas are selected, considering the data characteristics and engineering characteristics, a loess area highway engineering construction risk evaluation index system is established as shown in the figure. Figure 3 The first-level indexes include engineering geology, natural environment, design construction and construction management, and the second-level factors include 18 items such as surrounding rock grade, water enrichment condition, geological structure, etc.

[0075] According to the risk probability level standard, the consequence loss level standard and the risk level standard, the risk evaluation index and risk level information contained in the highway engineering construction risk source database are quantified, the database is preprocessed based on the principal component analysis method, and the sample data set is determined.

[0076] ​​S2, risk evaluation model construction based on T-S fuzzy neural network. The sample data set is divided into training set and test set according to the proportion, the T-S fuzzy neural network is designed, the fuzzy neural network is continuously trained and the parameters and the number of neurons are optimized by using PSO algorithm, finally the test set is input into the trained neural network, the risk level is predicted, the effectiveness of the fuzzy neural network is tested, and the risk evaluation model based on the improved T-S fuzzy neural network is constructed;

[0077] In step S2, the risk evaluation model based on the improved T-S fuzzy neural network is constructed, as shown in Figure 4 The variable description of the fuzzy neural network risk evaluation model is shown in Table 7.

[0078] Table 7

[0079]

[0080] The specific modeling steps are as follows:

[0081] Data preprocessing. 17 groups of sample data are randomly selected as the training set, and the remaining 8 groups are the test set for the next modeling and verification. The training set input vector , the training set output vector ; the test set input vector , the test set output vector .

[0082] The input layer of the antecedent network. The number of input layer nodes is the same as the dimension of the input vector, and in this example, the number of input layer nodes is The value of m, each node is connected with the corresponding input vector, the input vector is transmitted to the next layer and the input vector is normalized.

[0083] Fuzzy layer of antecedent network. For input data, the membership degree of each input variable is calculated according to fuzzy rules. Considering the sensitivity of neural network and the characteristics of data, the bell-shaped function with Gaussian distribution is selected as the membership function, and the membership degree of input variable is:

[0084] (1)

[0085] Wherein, and are the center and width number of the membership function respectively; is the number of fuzzy subsets.

[0086] Fuzzy rule calculation layer of antecedent network. The membership degrees are calculated, the multiplication algorithm is used to calculate the weight value from the fuzzy reasoning layer neuron to the output layer neuron, and the multiplication operator is:

[0087] (2)

[0088] wherein, is a weighting coefficient; is the membership function of the lower fuzzy set; is the dimension of .

[0089] The antecedent network output layer. The fuzzy rules mapped by each node are used to calculate the corresponding output variable, realizing the nonlinear mapping from the input space to the output space. The antecedent network output result is defuzzified to calculate the weight of each attribute .

[0090] (3)

[0091] The consequent network input layer. That is, the input in the network, which functions to transmit the input variable to the fully connected layer.

[0092] The consequent network fully connected layer. The input variable of the input layer is accepted to realize the learning of the input variable and obtain the output result , wherein the calculation method of .

[0093] (4)

[0094] wherein, is the fuzzy neural network coefficient.

[0095] The consequent network output layer. The weight transmitted from the antecedent network and the output result of the consequent network are received to obtain the final output result , wherein the calculation method of .

[0096] (5)

[0097] In step S2, the particle swarm algorithm is used to optimize the T-S fuzzy neural network parameter configuration, and an improved particle swarm algorithm is designed to optimize the T-S fuzzy neural network algorithm. The specific algorithm design steps are as follows:

[0098] Parameter initialization. The particle swarm size , the maximum number of iterations , the inertia weight and the learning factor are input. The parameters of the T-S fuzzy neural network are randomly initialized, including the membership function center, the width and the network weight.

[0099] Define the fitness function of particles. The mean square error is used as the fitness function, and the optimization goal is to minimize the training error. Calculate the network output of each particle under the current parameter configuration, and calculate the error value.

[0100] Improve the particle update strategy. Adopt adaptive inertia weight to enhance the global search ability. Adopt mutation strategy to avoid falling into local optimum, and introduce Gaussian noise disturbance for particles with poor local optimal solution. Adopt particle hierarchical update strategy to improve convergence efficiency.

[0101] 1) Adaptive inertia weight ω (enhance global search). Introduce adaptive inertia weight to control the exploration intensity of individuals based on relative fitness. Take large weight for early and poor particles, and take small weight for late and good particles, considering global search and refinement.

[0102] For the first particle, let its fitness be , . Thus, the adaptive inertia weight of the particle with low fitness is larger, and the adaptive inertia weight of the particle with high fitness is smaller.

[0103] 2) Mutation strategy (Gaussian noise disturbance to avoid local optimum). When the individual stagnates for a long time or is significantly worse than the global optimum, fine-tune the position (or velocity) with decreasing intensity Gaussian disturbance to restore diversity and avoid falling into local optimum, while setting a very low mutation probability for elite particles to protect convergence.

[0104] For the first particle, if has not improved for consecutive generations, or , or or is lower than the threshold, the particle position and velocity are mutated as follows:

[0105] (6)

[0106] (7)

[0107] (8)

[0108] Where, the particle position is clipped to ensure it is between the lower bound and the upper bound , , is the variable search width, decreases with the number of generations.

[0109] For the triggered particle, set Perform mutation probability, such as setting a very low probability (0-1%) for excellent particles to protect convergence.

[0110] 3) Particle stratified update (differentiate parameters within layers, improve efficiency). The population is divided into elite, development, and exploration layers according to fitness, respectively configured (ω, , , , mutation rate), and the local optimal topology is used for the exploration layer to search different basins in parallel, thereby improving the overall convergence efficiency.

[0111] Among them, the stratification rule (sort by fitness every generation) can be set as follows:

[0112] (1) Elite layer: small inertia, small speed upper limit, weak or disabled mutation, focusing on local convergence.

[0113] (2) Development layer: medium inertia, standard update, a small amount of mutation, and fine search around the optimal solution.

[0114] (3) Exploration layer: large inertia, large , higher mutation, and local neighborhood optimal instead to encourage multi-peak search.

[0115] velocity and position update. Particle swarm optimization uses a speed-position search model, and the standard PSO formula is used to update the speed and position:

[0116] (9)

[0117] where, and represent the individual historical optimal solution and the global optimal solution, , is a random number. The termination condition of the algorithm is when the maximum number of iterations is reached or the fitness function converges, the optimal parameters are output.

[0118] S3, real-time evaluation of highway engineering construction risk level. Input the real-time highway engineering construction risk data into the risk evaluation model established in S2 to determine the risk level and complete the real-time evaluation of the highway engineering construction risk level in the loess area.

[0119] In step S3, for the highway engineering construction risk of the tunnel part, the model training mean square error is 0.02 and the test mean square error is 0.05, both of which are within the acceptable range, indicating that the model can effectively evaluate the safety risk level of highway tunnel engineering. The input vector , the output vector If 18 known indexes are used to predict 1 risk level, the risk level can be determined, and the real-time evaluation of the risk level of highway engineering construction in loess area can be completed.

[0120] The above merely describes the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for assessing construction risks in highway engineering in loess areas based on TS fuzzy neural networks, characterized in that, include: Obtain a sample dataset of construction risk sources for highway engineering in loess areas; A risk assessment model is constructed, wherein the risk assessment model adopts a TS fuzzy neural network. Based on the sample dataset, the risk assessment model is optimized by a particle swarm optimization algorithm to obtain an optimized risk assessment model. Real-time acquisition of highway construction risk data, followed by risk assessment using an optimized risk evaluation model to determine the highway construction risk level.

2. The method according to claim 1, characterized in that, The process of obtaining the sample dataset includes: Obtain static and real-time data on highway construction in loess areas; Based on the static and real-time data, a risk source database for highway construction is constructed, and a risk assessment index system for highway construction in loess areas is established. Based on the risk assessment index system, the risk source database is organized to obtain a sample dataset.

3. The method according to claim 2, characterized in that, The process of constructing the risk assessment indicator system includes: The WBS-RBS method is used to decompose the data in the risk source database step by step, wherein the highway engineering work in the risk source database is decomposed step by step to obtain the highway engineering work decomposition structure. By decomposing the construction risks of highway engineering in the loess area from the risk source database, the construction risk structure of highway engineering is obtained. The risk factors at the end of the highway construction risk structure are coupled and matched with the project units at the end of the highway engineering work breakdown structure. Based on the matching results, a risk assessment index system is constructed.

4. The method according to claim 3, characterized in that, The specific process of obtaining the sample dataset includes: Based on the risk assessment indicator system, obtain the risk probability level standard, the consequence loss level standard, and the risk level standard; The risk assessment indicators and risk levels in the risk source database are quantified based on the risk probability level standard, the consequence loss level standard, and the risk level standard. Principal component analysis is then used to process the quantified data to obtain a sample dataset.

5. The method according to claim 4, characterized in that, The risk assessment index system includes primary indicators and secondary factors. The primary indicators include engineering geology, natural environment, design and construction, and construction management. The secondary factors of engineering geology include surrounding rock grade, water abundance, geological structure, and non-uniform settlement. The secondary factors of natural environment include average annual precipitation, topography, vegetation cover, vegetation matching type, and surface water conditions. The secondary factors of design and construction include parameter design, geological survey, excavation method, timeliness of support, and construction defects. The secondary factors of construction management include on-site management, safety training, emergency plans, and engineering water conservation measures.

6. The method according to claim 1, characterized in that, In the risk assessment model, risk indicators from the risk assessment indicator system are used as input data. The membership degree of the risk indicators is calculated according to fuzzy rules. The membership degree is then fuzzily calculated using a multiplication algorithm to obtain weighting coefficients. The weighting coefficients are then defuzzified to obtain weights for different attributes. The input data is then processed through a fully connected layer to obtain the output of the consequent network. Finally, a weighted calculation is performed based on the weights of the different attributes and the output of the consequent network to obtain the risk level of highway construction.

7. The method according to claim 1, characterized in that, The process of optimizing the risk assessment model using the particle swarm optimization algorithm includes: The algorithm parameters of the particle swarm optimization algorithm are initialized, and the network parameters of the risk assessment model are initialized. The network parameters of the initialized risk assessment model are used as the initial particle swarm of the particle swarm optimization algorithm, wherein the network parameters include membership function centers, width and network weights. The mean squared error of the risk assessment model is used as the fitness function, and the minimization of the training error of training the risk assessment model based on the sample dataset is taken as the objective. Based on the objective and fitness function, the initial particle swarm is iteratively optimized using a particle swarm optimization algorithm until the maximum number of iterations is reached or the fitness function converges, thus obtaining the final particle swarm values, which are the network parameters of the optimized risk assessment model.

8. The method according to claim 1, characterized in that, During the iterative optimization process, an adaptive inertial weight is introduced into the particle swarm update process, and a mutation strategy and a particle hierarchical update strategy are adopted. Gaussian noise perturbation is introduced into particles whose local optimal solution is less than a certain threshold.

Images