*3.2. Establish BP Neural Network Model*

The samples of this case came from a total of 216 high cutting slopes of the HK, HF, YG, and HSH high-speed railway. By collecting the geological situation, special construction plan, and minutes of each high cutting slope and considering the relevant indicator rating methods in risk guidance combined with engineering practices and expert interviews, the indicators in each sample were assigned values according to Table 7, and all data were normalized. The details of the data processing are shown in the Appendix A. The Ordered Weighted Averaging is used to determine the weights of indicators at all levels, and the weight determination method is shown in Formula (6).

$$w = \overline{w}\_{\text{i}} \Big/ \sum\_{i=1}^{j} \overline{w}\_{\text{i}} , \text{i} = 1, 2, \dots, n \text{ j} = \text{i} = 1, 2, \dots, m \tag{6}$$

The weight values of CT, ME, P, E, and CM were 0.328, 0.104, 0.189, 0.258, and 0.121 respectively, as shown in Figure 3. From Figure 3, it can be seen that the risks of construction technology and environment have a relatively strong influence on the construction safety of high cutting slopes in HSR, and the risk of material and equipment, personnel, and construction management have a relatively weak influence on it.

**Figure 3.** The weight values of the secondary risk indicators.

#### *3.3. Training Simulation*

According to the BP neural network algorithm, Matlab was used to create the neural network and run the results. After 11 iterations, the training MSE value of the simulation training was 0.000176, which is less than the target value of 0.0003. This met the predetermined accuracy requirements, and the BP neural network model achieved convergence. The simulation results are shown in Figure 4; it can be seen that the R-values of the training set and test set are 0.947 and 0.808, which indicate that the model fits the observed values well.

**Figure 4.** Schematic diagram of BP Neural network training fit degree.

The results of the error analysis for the test set data are shown in Figure 5. It can be seen that the maximum absolute value of the error is 8.22%; therefore, the training effect of the model is satisfactory.

#### *3.4. Discussion*

The trained BP artificial neural network was used to predict the risk of an HF grade eight high cutting slope construction project. After normalizing the data for each risk indicator (0.788, 0.394, 0.303, 0.303, 0.273, 0.455, 0.303, 0.303, 0.485, 0.273, 0.485, 0.273, 0.364, 0.485, 0.424, 0.212, 0.576, 0.278, 0.364, 0.389, 0.212, 0.697, 0.303, 0.818, 0.576, 0.273, 0.485, 0.424, 0.636, 0.394, 0.455, 0.364, 0.364, 0.394, 0.212, 0.394, 0.333, 0.424, 0.758), they were entered into the network to obtain the predicted value of 0.479. According to Table 7, the construction safety risk of the high cutting slope is predicted to be a medium risk, which is consistent with the risk level of the project; thus, the model fit is excellent.

According to the input value of each risk factor, the risk index greater than 0.5 is the main risk. It is known that the main risks of this high cutting slope construction are earthwork excavation method, scaffolding equipment, slope height, slope rate, groundwater, personnel safety awareness, and construction safety risk management system, and these influencing factors have a greater impact on the construction safety management of the project. Therefore, the safety control of this high cutting slope construction focuses on slope excavation risk control, slope reinforcement, waterproof measures, and construc-

tion safety management measures. The following control measures are proposed for the construction safety risks of this high cutting slope project: (1) choosing a reasonable earth excavation method; (2) setting up reinforcement protection measures; (3) strengthening of waterproof design; (4) conducting pre-reinforcement treatment; (5) monitoring stability; (6) strengthening construction management measures.

**Figure 5.** Error analysis for the test set data.

#### **4. Conclusions**

In this study, the safety risks of high cutting slope construction in HSR were identified in all aspects from three dimensions, risk technical specification, literature analysis, and case statistical analysis, and a list of risk influencing factors was formed. The evaluation indicator system was constructed by designing questionnaires and analyzing them with SPSS data statistical software. The assessment model was established by a BP neural network, and the pre-control measures were proposed for the risk factors. The construction safety risks of a high cutting slope of HF high-speed railway was analyzed and evaluated. The main findings of this study are (1) a list of construction safety risks of high cutting slopes in HSR was formed; (2) a risk assessment indicator system of high cutting slopes in HSR was constructed; (3) a construction safety risk assessment model based on a BP neural network was established; and (4) the feasibility of the assessment model was verified.

The limitation of this study is the identification and analysis of construction risk factors with a certain one-sidedness and subjectivity. Combined with the dynamic and difficult quantitative nature of construction risks, it needs to be further combined with engineering practice to refine and improve the construction safety impact factors. In addition, the number and authenticity of the learning samples directly ensure the feasibility of the trained neural network, and more samples need to be collected to improve the sample credibility. Finally, the BP neural network training process is related to set parameters, which will be combined with more intelligent algorithms for improvement to improve the accuracy of the training results in the future.

**Author Contributions:** Conceptualization, J.H. and H.C.; methodology, X.Z. and J.F.; software, X.Z.; validation, X.Z. and J.F.; formal analysis, X.Z.; investigation, J.H., X.Z., J.F., Y.H. and H.C.; resources, J.H. and H.C.; data curation, Y.H.; writing—original draft preparation, X.Z. and J.F.; writing—review and editing, X.Z.; visualization, X.Z.; supervision, J.H. and H.C.; project administration, J.H. and H.C.; funding acquisition, J.H. and H.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was funded by the National Key R&D Program of China (Grant Number 2021YFB2301802).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All datasets generated for this study have been included in the article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**

For the construction technology risk (CT), the collected assessment data are shown in Table A1.


**Table A1.** Raw data of construction technical risk (CT).

By using Formula (1), data processing is shown in Table A2.

**Table A2.** Construction technical risk (CT) data processing.


The output values were normalized according to Formula (2), and the results are shown in Table A3.

**Table A3.** Normalized data of construction technical risk (CT).


For material and equipment risk (ME), the assessment data collected are shown in Table A4.


**Table A4.** Raw data of material and equipment risk (ME).

By using Formula (1), data processing is shown in Table A5.

**Table A5.** Data processing of material and equipment risk (ME).


The output values are normalized according to Formula (2), and the results are shown in Table A6.

**Table A6.** Normalized data of material and equipment risk (ME).


For environmental risk (E), the assessment data collected are shown in Table A7.


**Table A7.** Raw data of environmental risk (E).

By using the Formula (1), data processing is shown in Table A8.

E7 3 3 3 23 2 2 3 2 2 2

**Table A8.** Data processing of environmental risk (E).


The output values were normalized according to Formula (2), and the results are shown in Table A9.

**Table A9.** Normalized data of environmental risk (E).


For personnel risk (P), the assessment data collected are shown in Table A10.


**Table A10.** Raw data of personnel risk (P).

By using Formula (1), data processing is shown in Table A11.

**Table A11.** Data processing of Personnel risk (P).


The output values were normalized according to formula (2), and the results are shown in Table A12.

**Table A12.** Normalized data of personnel risk (P).


For construction management risk (CM), the assessment data collected are shown in Table A13.


**Table A13.** Raw data of construction management risk (CM).

By using Formula (1), data processing is shown in Table A14.

**Table A14.** Data processing of construction management risk (CM).


The output values are normalized according to Formula (2), and the results are shown in Table A15.


**Table A15.** Normalized data of construction management risk (CM).

Due to space issues, only the first ten groups of case data processing results are listed in this study, as shown in Table A16.


**Table A16.** Input data for the first ten samples.
