Deep foundation pit stability monitoring dynamic early warning method considering advanced prediction

Abstract

The invention belongs to the technical field of building engineering construction, and relates to a deep foundation pit stability monitoring dynamic early warning method considering advanced prediction. Comprising the following steps: S1, constructing a deep foundation pit monitoring index system and a measurement system under the apron beach landform; s2, determining subjective weights of monitoring indexes based on an improved G1 method; s3, correcting the index weight based on a CRITIC method; s4, determining the comprehensive weight of the monitoring indexes of the deep foundation pit; s5, obtaining a deep foundation pit stability evaluation score based on a TOPSIS method; s6, refining and grading safety levels under the special landform; and S7, predicting a deep foundation pit stability evaluation score based on the neural network model and setting early warning. According to the method, the engineering entity monitoring layout and the calculation method are optimized according to the geomorphic characteristics of the apron beach, and longitudinal scoring and early warning of the time sequence are carried out for deep foundation pit monitoring, so that the engineering analysis result has higher reliability and comprehensiveness.

Description

Technical Field

[0001] The invention belongs to the technical field of building engineering construction, and relates to a deep foundation pit stability monitoring dynamic early warning method considering advance prediction. Background Art

[0002] The stability of deep foundation pits under the floodplain is affected by a variety of factors, including geological conditions, groundwater distribution, and surrounding building environment. Due to the complexity and diversity of on-site construction conditions, actual projects often cannot fully meet the design requirements of the support structure, so there are certain deviations in actual construction. The site monitoring data can track and reflect the stress conditions of the rock and soil at the corresponding parts, which plays an important guiding role in guiding the design and adjustment of the foundation pit support structure. Therefore, the research on monitoring data analysis methods has important engineering value and scientific significance for the stability assessment of deep foundation pits, and can provide effective technical support for improving the safety and efficiency of foundation pit construction.

[0003] However, the following problems exist in the monitoring of deep foundation pit stability under special landforms: First, qualitative methods such as the fault tree method and the WBS method can identify the potential causes of system failures or accidents, clarify the logical relationship between the factors affecting the stability of deep foundation pits and give a concise description, but the analysis process is relatively complex and cumbersome, lacking more professional considerations of the stability of deep foundation pits, and lacking engineering experience accumulation for deep foundation pit monitoring under offshore floodplain landforms. Quantitative analysis methods such as the TOPSIS method systematically consider the factors affecting the stability of deep foundation pits, and determine objective scores by approaching the ideal solution, but the weights depend on pre-input. In order to obtain better weight value results, different combinations of weighting methods are often used for simulation output, which inevitably contains certain subjective components. Even objective weighting will produce results that are contrary to the actual results of indicator weight analysis. Second, in the previous classification of deep foundation pit construction safety status, the foundation pit stability safety level is often divided into 3 to 4 levels with low accuracy based on the experience of experts, which cannot meet the actual production and construction links. The needs of on-site safety; third, deep foundation pit monitoring and early warning focus on a single real cross-section, lacking long-term dynamic tracking and early warning of the foundation pit stability safety risk time series. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a dynamic early warning method for monitoring the stability of deep foundation pits taking into account advance prediction. The problem of low credibility of the weight calculation of deep foundation pit monitoring indicators under special construction landforms is solved, the dynamic tracking and early warning program of deep foundation pit monitoring is optimized, and the construction efficiency and project economy are improved.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] A dynamic early warning method for monitoring the stability of a deep foundation pit taking into account advance prediction is characterized by comprising the following steps:

[0007] S1. Construct a monitoring index system and measurement system for deep foundation pits under floodplain landforms;

[0008] S2, determine the subjective weights of monitoring indicators based on the improved G1 method;

[0009] S3, modify the indicator weights based on the CRITIC method;

[0010] S4. Determine the comprehensive weight of deep foundation pit monitoring indicators;

[0011] S5. Determine the deep foundation pit stability evaluation score based on the TOPSIS method;

[0012] S6. Detailed classification of safety levels under special landforms;

[0013] S7. Predict the deep foundation pit stability evaluation score based on the neural network model and set an early warning.

[0014] The step 1 specifically includes:

[0015] S11. Based on the construction characteristics of deep foundation pits under the floodplain landform, monitor the main deformation conditions, settlement observation data and geological structure characteristics during the construction process, including the deformation of the support structure and the surrounding environment and the monitoring of the deep-level changes of the foundation pit, and build a step-by-step hierarchical evaluation index system to determine the various index factors that affect the stability of deep foundation pits;

[0016] S12. Build a multi-level indicator evaluation model that affects the safety and stability of deep foundation pits. According to the monitoring needs of deep foundation pits in floodplain landforms, the first level includes engineering monitoring data of the main structure of the pit and the surrounding environment; the second level indicators include groundwater level, deep-level changes, horizontal and vertical displacements of structures, and measured data of pipeline settlement to evaluate the impact of the underground environment on the stability of the foundation pit;

[0017] S13. Monitoring points are set up in the submarine and landforms of the project site, and the monitoring data sets of underwater shoals and floodplains are expanded according to the requirements of on-site construction. The monitoring contents mainly include horizontal displacement of the pit top, vertical displacement of the pit top, deep horizontal displacement, axial force, groundwater level, and settlement of surrounding buildings. Laser displacement sensors are arranged at the top of the foundation pit slope within 10-20m around the foundation pit. At the same time, for the slope with strengthened support, an additional set of laser displacement sensors is arranged within 1m of the top of the slope to comprehensively monitor the The horizontal and vertical displacements of the top of the foundation pit slope are monitored; monitoring targets are set on the main load-bearing components of the buildings around the foundation pit, and laser prisms are set on the base plane of the building. The surrounding environmental settlement and other data are monitored through AI cameras; the monitoring system is densely arranged in the seawater cofferdam area of ​​the deep foundation pit, and different instrument combinations such as pressure gauges, static levels or flow meters are set to monitor the water level changes or flow rate changes on both sides of the seawater cofferdam. At the same time, a SwY-20 steel ruler water level meter with an observation accuracy of 5mm is set in the foundation pit to monitor the groundwater level.

[0018] The step 2 specifically includes:

[0019] S21. Starting from the bottom of the indicator hierarchy, the weight of each indicator factor at the same level relative to the indicator at the previous level is annotated, and a certain judgment is made on the relative importance of each indicator at each level;

[0020] Assuming that the indicator system consists of three levels: target layer, primary indicator and secondary indicator, firstly, the weight of the secondary indicator relative to the primary indicator is calculated, i.e., indicator X1, X2, ..., X m The weight relative to indicator X is denoted as Indicators Y1, Y2, ... Y n The weight relative to Y is denoted as Then, the weights of the first-level indicators X and Y relative to the target layer are calculated and recorded as

[0021] S22. Invite i experts to score the importance of adjacent indicators at the same level. The scores range from 1 to 10. The higher the corresponding value, the more important the indicator represented. The scoring criteria are shown in Table 1:

[0022] Table 1 Scoring criteria

[0023] Importance Fraction Especially important 9-10 More important 6-8 General Important 4-5 unimportant 1-3

[0024] S23, sort the indicators that have been evaluated for importance; according to the importance of the indicators scored by the jth expert, sort them according to the size of the scores, such as the importance score of indicator X1 Greater than index X n Importance score Then record it as X1≥X n , the order of importance is:

[0025] X1≥X2≥…≥X n ;

[0026] S24. According to the score value of the jth expert on the importance of the indicator, calculate the ratio of the weights of adjacent indicators at the same level, and evaluate the indicator X. k-1 With X k The weight ratio is recorded as:

[0027]

[0028] In the above formula: It is represented by the index X based on the score of the jth expert. k The weight of

[0029] S25. Based on the score of the jth expert, calculate the index X n Weight for:

[0030]

[0031] The weights of other indicators can be obtained by the recursive formula converted in step S24, as follows:

[0032]

[0033] The step 3 specifically includes:

[0034] S31, for index x j Calculate the standard deviation, denoted as σ j , x ij is the original value of the i-th indicator of the j-th evaluation object; is the average value of the indicators in m evaluation objects, and the calculation steps are as follows:

[0035]

[0036] S32, construct the correlation coefficient matrix R of n indicators = (r ij ) n×n , r ij is the vector x i With x j The Pearson correlation coefficient is calculated as follows:

[0037]

[0038] S33, set Q i is the amount of information contained in the i-th indicator, and its expression is:

[0039]

[0040] S34, using information quantity Q i Calculate the objective weight of the indicator

[0041]

[0042] The step 4 specifically includes:

[0043] The weights obtained by the improved G1 method and the CRITIC method are combined using the multiplication normalization method to obtain the combined weight w i , the calculation steps are as follows:

[0044]

[0045] The step 5 specifically includes:

[0046] S51, construct an initial matrix based on the sample, and perform vector normalization processing to obtain a standard decision matrix denoted as Y=(v ij ) m×n ;

[0047] S52, substitute the combined weights obtained in step S4 to construct a weighted standard matrix Z=(z ij ) m×n , the weighted canonical matrix element calculation formula is:

[0048] z ij =w i y ij , w i To improve the combined weights of G1 and CRITIC methods;

[0049] S53. Determine positive and negative ideal solutions and

[0050]

[0051] S54, according to the positive and negative ideal solution and Find the Euclidean distance and That is, the distance between the actual value of the evaluation indicator and the optimal and worst values ​​of the indicator:

[0052]

[0053] S55. Calculate the relative closeness C of each evaluation index to the optimal value i , C i The larger the value, the closer it is to the optimal level, and vice versa:

[0054]

[0055] The step 6 specifically includes:

[0056] S61. Define the corresponding relationship between relative closeness, evaluation level and evaluation score as shown in Table 2:

[0057] Table 2 defines the corresponding relationship between relative closeness, evaluation level and evaluation score

[0058]

[0059] Among them: EL is the ratio of direct economic loss to total investment; CL is the ratio of lost construction period to planned construction period; MW is the number of minor injuries; SW is the number of serious injuries; D is the number of deaths;

[0060] S62: In order to make the score evaluation more precise, the corresponding relationship between the relative closeness and the evaluation score defined in step S61 is expressed as follows:

[0061]

[0062] S63. Calculate the total evaluation score based on the weights of different indicators. The formula is as follows:

[0063]

[0064] Among them, w i represents the combined weight of the i-th indicator, p i It represents the evaluation score of the ith indicator, m is a selected positive real number, and P is the final score value.

[0065] The step 7 specifically includes:

[0066] S71, neural network modeling is performed according to the multi-level index evaluation model built in step S13, and a multi-level prediction system of input layer, intermediate layer and output layer is confirmed, and the final output layer, i.e., the prediction result, is the comprehensive evaluation score of the deep foundation pit stability obtained by the target in step S63;

[0067] Among them, the calculation process from the input layer to the middle layer and from the middle layer to the input layer is forward propagation, and the process of comparing the output result with the expected output result and using the network to feed back the generated error is back propagation;

[0068] S72. From the input layer to the middle layer, the Sigmoid activation function is introduced to normalize the output of each neuron:

[0069]

[0070] S73. From the middle layer to the output layer, a regression function is introduced to obtain the predicted value of deep foundation pit stability:

[0071] y=∑w i x i +b

[0072] Where: x i is the input value of the first-level indicator in the middle layer, w i is the learning weight of the neural network, and b is the intercept term;

[0073] S74. Back propagation during iterative calculation uses the gradient descent learning method to determine the loss function of the optimization target:

[0074]

[0075] S75. Through iterative model learning, the dynamically updated deep foundation pit stability monitoring data set is input, the comprehensive evaluation score of the deep foundation pit stability under the floodplain landform is predicted in advance, and the evaluation level is refined according to step S61 to achieve a dynamic cycle early warning effect.

[0076] The dynamic connection between step S1 and step S5, S7 data reuse is:

[0077] The data collected in step S1 are preprocessed and then a stability evaluation score is obtained in step S5;

[0078] The original data acquired by monitoring in step S1 is input again, and the evaluation score in step S5 is used as the output target, and the evaluation score of the next period is predicted through the neural network model in step S7;

[0079] When the data in step S1 is reused in step S7, it must be entered along with the update of the monitoring data of the whole process of deep foundation pit construction, that is, the data set in step S7 has dynamic update capability over step S5;

[0080] In step S7, the ratio of the training set, validation set, and test set in the data set entered is 7:2:1.

[0081] Beneficial effects of the present invention:

[0082] (1) The present invention uses improved G1-CRITC and TOPSIS evaluation methods to evaluate the stability of deep foundation pits under floodplain landforms. The improved part of the G1 method uses a ten-point system to replace the original method's ratio evaluation, which is more suitable for daily construction scoring methods, simplifies operations and can further increase the degree of differentiation. In addition, the multiplication normalization method is used to preprocess the indicator weight data. This tool is used to normalize the characteristic values ​​in the various indicator weight data sets calculated by the improved G1 method and the CRITIC method to ensure that the values ​​of different characteristics have similar scales, thereby improving the model performance of data analysis. Finally, the TOPSIS method is used to make full use of the original data information and objectively evaluate the monitoring input data. This invention solves the problem that the rationality of the establishment of the indicator system in the existing analysis method is affected by the level of expert experience. It can objectively and accurately assign weights to the evaluation indicators, and arrange them in a sequential relationship to objectively display the dynamic scoring trajectory of the stability of the deep foundation pit, thereby improving the reliability of the weight calculation and evaluation results.

[0083] (2) Based on the "Technical Standards for Monitoring of Construction Foundation Pit Engineering" and "General Specifications for Building and Municipal Foundations", combined with the content of foundation pit monitoring under special landforms, the present invention proposes a dynamic early warning method for monitoring the stability of deep foundation pits under floodplain landforms that takes into account advance prediction. By importing the measured data sequence of key indicators of the project deep foundation pit, the subjective and objective weighting methods are combined for weighting, and the TOPSIS method is used for comprehensive evaluation. The corresponding table of evaluation levels and evaluation scores is divided into 5 levels, which makes the evaluation of deep foundation pit stability more detailed in terms of score division and safety level assessment.

[0084] (3) The present invention uses a neural network model to predict the deep foundation pit stability evaluation score through iterative learning of forward and backward propagation within the model. The neural network model performs iterative learning based on the captured engineering monitoring data, and has the advantages of a large data set sample size and continuous updating of samples during the whole process monitoring. It surpasses the current situation of traditional deep foundation pit monitoring and early warning focusing on a single real cross section, and realizes long-term dynamic tracking and early warning of construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 Flow chart of the method of the present invention.

[0086] Figure 2 This is a structural diagram of the multi-level evaluation indicators provided by the present invention.

[0087] Figure 3 A deep foundation pit engineering safety status evaluation index system is provided in an embodiment of the present invention.

[0088] Figure 4 This is the neural network hierarchy provided by the present invention.

[0089] Figure 5 It is a dynamic scoring diagram of the deep foundation pit structure within the period of the embodiment of the present invention.

[0090] Figure 6 It is a dynamic monitoring scoring diagram of deep foundation pit stability within the period of the embodiment of the present invention.

[0091] Figure 7 This is a graph showing the reduction in loss rate of the model according to an embodiment of the present invention.

[0092] Figure 8 It is a dynamic monitoring score of deep foundation pit stability and a prediction diagram of a neural network model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0093] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and implementation plans:

[0094] Example 1

[0095] like Figure 1 As shown, a dynamic early warning method for deep foundation pit stability monitoring considering advance prediction specifically includes the following steps:

[0096] S1. Construct a monitoring index system and measurement system for deep foundation pits under floodplain landforms; specifically include:

[0097] S11. Based on the construction characteristics of deep foundation pits under the floodplain landform, monitor the main deformation conditions, settlement observation data and geological structure characteristics during the construction process, including the deformation of the support structure and the surrounding environment and the monitoring of the deep-level changes of the foundation pit, and build a step-by-step hierarchical evaluation index system to determine the various index factors that affect the stability of deep foundation pits;

[0098] S12. Build a multi-level indicator evaluation model that affects the safety and stability of deep foundation pits. According to the monitoring needs of deep foundation pits in floodplain landforms, the first level includes engineering monitoring data of the main structure of the pit and the surrounding environment; the second level indicators include groundwater level, deep-level changes, horizontal and vertical displacements of structures, and measured data of pipeline settlement to evaluate the impact of the underground environment on the stability of the foundation pit;

[0099] S13. Monitoring points are set up in the submarine and landforms of the project site, and the monitoring data sets of underwater shoals and floodplains are expanded according to the requirements of on-site construction. The monitoring contents mainly include horizontal displacement of the pit top, vertical displacement of the pit top, deep horizontal displacement, axial force, groundwater level, and settlement of surrounding buildings. Laser displacement sensors are arranged at the top of the foundation pit slope within 10-20m around the foundation pit. At the same time, for the slope with strengthened support, an additional set of laser displacement sensors is arranged within 1m of the top of the slope to comprehensively monitor the The horizontal and vertical displacements of the top of the foundation pit slope are monitored; monitoring targets are set on the main load-bearing components of the buildings around the foundation pit, and laser prisms are set on the base plane of the building. The surrounding environmental settlement and other data are monitored through AI cameras; the monitoring system is densely arranged in the seawater cofferdam area of ​​the deep foundation pit, and different instrument combinations such as pressure gauges, static levels or flow meters are set to monitor the water level changes or flow rate changes on both sides of the seawater cofferdam. At the same time, a SwY-20 steel ruler water level meter with an observation accuracy of 5mm is set in the foundation pit to monitor the groundwater level.

[0100] S2. Determine the subjective weights of monitoring indicators based on the improved G1 method; specifically include:

[0101] S21. Starting from the bottom of the indicator hierarchy, the weight of each indicator factor at the same level relative to the indicator at the previous level is annotated, and a certain judgment is made on the relative importance of each indicator at each level;

[0102] Assuming that the indicator system consists of three levels: target layer, primary indicators and secondary indicators, firstly, the weight of the secondary indicators relative to the primary indicators is calculated, i.e., indicators X1, X2,…, X m The weight relative to indicator X is denoted as Indicators Y1, Y2, ... Y n The weight relative to Y is denoted as Then, the weights of the first-level indicators X and Y relative to the target layer are calculated and recorded as

[0103] S22. Invite i experts to score the importance of adjacent indicators at the same level. The scores range from 1 to 10. The higher the corresponding value, the more important the indicator represented. The scoring criteria are shown in Table 1:

[0104] Table 1 Scoring criteria

[0105] Importance Fraction Especially important 9-10 More important 6-8 General Important 4-5 unimportant 1-3

[0106] S23, sort the indicators that have been evaluated for importance; according to the importance of the indicators scored by the jth expert, sort them according to the size of the scores, such as the importance score of indicator X1 Greater than index X nImportance score Then record it as X1≥X n , the order of importance is:

[0107] X1≥X2≥…≥X n ;

[0108] S24. According to the score value of the jth expert on the importance of the indicator, calculate the ratio of the weights of adjacent indicators at the same level, and evaluate the indicator X. k-1 With X k The weight ratio is recorded as:

[0109]

[0110] In the above formula: It is represented by the index X based on the score of the jth expert. k The weight of

[0111] S25. Based on the score of the jth expert, calculate the index X n Weight for:

[0112]

[0113] The weights of other indicators can be obtained by the recursive formula converted in step S24, as follows:

[0114] S3. Modify the indicator weights based on the CRITIC method; specifically:

[0115] S31, for index x j Calculate the standard deviation, denoted as σ j , x ij is the original value of the i-th indicator of the j-th evaluation object; is the average value of the indicators in m evaluation objects, and the calculation steps are as follows:

[0116]

[0117] S32, construct the correlation coefficient matrix R of n indicators = (r ij ) n×n , r ij is the vector x i With x j The Pearson correlation coefficient is calculated as follows:

[0118]

[0119] S33, set Q i is the amount of information contained in the i-th indicator, and its expression is:

[0120]

[0121] S34, using information quantity Q i Calculate the objective weight of the indicator

[0122]

[0123] S4. Determine the comprehensive weight of deep foundation pit monitoring indicators; specifically include:

[0124] The weights obtained by the improved G1 method and the CRITIC method are combined using the multiplication normalization method to obtain the combined weight w i , the calculation steps are as follows:

[0125]

[0126] S5. Determine the deep foundation pit stability evaluation score based on the TOPSIS method; specifically:

[0127] S51, construct an initial matrix based on the sample, and perform vector normalization processing to obtain a standard decision matrix denoted as Y=(y ij ) m×n ;

[0128] S52, substitute the combined weights obtained in step S4 to construct a weighted standard matrix Z=(z ij ) m×n , the weighted canonical matrix element calculation formula is:

[0129] z ij =w i y ij , w i To improve the combined weights of G1 and CRITIC methods;

[0130] S53. Determine positive and negative ideal solutions and

[0131]

[0132] S54, according to the positive and negative ideal solution and Find the Euclidean distance and That is, the distance between the actual value of the evaluation indicator and the optimal and worst values ​​of the indicator:

[0133]

[0134] S55. Calculate the relative closeness C of each evaluation index to the optimal value i , C iThe larger the value, the closer it is to the optimal level, and vice versa:

[0135]

[0136] S6. Detailed classification of safety levels under special landforms; specifically including:

[0137] S61. Define the corresponding relationship between relative closeness, evaluation level and evaluation score as shown in Table 2:

[0138] Table 2 defines the corresponding relationship between relative closeness, evaluation level and evaluation score

[0139]

[0140] Among them: EL is the ratio of direct economic loss to total investment; CL is the ratio of lost construction period to planned construction period; MW is the number of minor injuries; SW is the number of serious injuries; D is the number of deaths;

[0141] S62: In order to make the score evaluation more precise, the corresponding relationship between the relative closeness and the evaluation score defined in step S61 is expressed as follows:

[0142]

[0143] S63. Calculate the total evaluation score based on the weights of different indicators. The formula is as follows:

[0144]

[0145] Among them, w i represents the combined weight of the i-th indicator, p i It represents the evaluation score of the ith indicator, m is a selected positive real number, and P is the final score value.

[0146] S7. Predict the deep foundation pit stability evaluation score based on the neural network model and set early warning, including:

[0147] S71, neural network modeling is performed according to the multi-level index evaluation model built in step S13, and a multi-level prediction system of input layer, intermediate layer and output layer is confirmed, and the final output layer, i.e., the prediction result, is the comprehensive evaluation score of the deep foundation pit stability obtained by the target in step S63;

[0148] Among them, the calculation process from the input layer to the middle layer and from the middle layer to the input layer is forward propagation, and the process of comparing the output result with the expected output result and using the network to feed back the generated error is back propagation;

[0149] S72. From the input layer to the middle layer, the Sigmoid activation function is introduced to normalize the output of each neuron:

[0150]

[0151] S73. From the middle layer to the output layer, a regression function is introduced to obtain the predicted value of deep foundation pit stability:

[0152] y=∑w i x i +b

[0153] Where: x i is the input value of the first-level indicator in the middle layer, w i is the learning weight of the neural network, and b is the intercept term;

[0154] S74. Back propagation during iterative calculation uses the gradient descent learning method to determine the loss function of the optimization target:

[0155]

[0156] S75. Through iterative model learning, the dynamically updated deep foundation pit stability monitoring data set is input, the comprehensive evaluation score of the deep foundation pit stability under the floodplain landform is predicted in advance, and the evaluation level is refined according to step S61 to achieve a dynamic cycle early warning effect.

[0157] The dynamic connection between step S1 and step S5, S7 data reuse is:

[0158] The data collected in step S1 are preprocessed and then a stability evaluation score is obtained in step S5;

[0159] The original data acquired by monitoring in step S1 is input again, and the evaluation score in step S5 is used as the output target, and the evaluation score of the next period is predicted through the neural network model in step S7;

[0160] When the data in step S1 is reused in step S7, it must be entered along with the update of the monitoring data of the whole process of deep foundation pit construction, that is, the data set in step S7 has dynamic update capability over step S5;

[0161] In step S7, the ratio of the training set, validation set, and test set in the data set entered is 7:2:1.

[0162] Example 2

[0163] This embodiment adopts the method described in Example 1, and is described by taking a submarine tunnel construction project. The project has a buried section of 380m on the north side and an open section of 50m; a buried section of 1192m on the south side and an open section of 200m. The landform of the project site is a floodplain, and the site environmental conditions and geological conditions are relatively complex.

[0164] Step 1: Construct a monitoring index system and measurement system for deep foundation pits under floodplain landforms:

[0165] During the construction of deep foundation pit, the cumulative settlement of the surrounding surface and pipe trench changes greatly. Figure 2 The indicator system construction principle in the multi-level evaluation indicator structure diagram shown in the figure constructs a four-level safety status evaluation indicator system, such as Figure 3 shown.

[0166] First layer: U = [B1, B2];

[0167] Second layer: B1 = [C1, C2, C3, C4, C5, C6]; B2 = [C7, C8, C9];

[0168] The third layer: C1 = [D1, D2]; C2 = [D3, D4]; C3 = [D5, D6]; C4 = [D7, D8]; C5 = [D9, D 10 ]; C6 = [D 11 , D 12 ]; C7 = [D 13 , D 14 ]; C8 = [D 15 , D 16 ]; C9 = [D 17 , D 18 ].

[0169] Step 2: Determine the subjective weight of monitoring indicators based on the improved G1 method;

[0170] Taking the foundation pit structure as an example, experts were invited to score the importance of the secondary indicators using the improved G1 method, and the scores were given in the following table:

[0171]

[0172] According to the formula listed in step S24, the ratio r of the weights between adjacent indicators is calculated. k , the output is as follows:

[0173]

[0174] According to the formula listed in step S25, the subjective judgment weight value of each indicator is calculated As shown in the following table:

[0175]

[0176] Step 3: Modify the indicator weights based on the CRITIC method;

[0177] First, D1 to D 12 The input data is dimensionless and non-negative, and then the weights of the D level are merged to the upper level, and finally the subjective evaluation weight results of each indicator under the CRITIC method are obtained. As shown in the following table:

[0178]

[0179] Step 4: Use the multiplication normalization method to combine the subjective and objective weight values ​​of the detection indicators to obtain the combined weight w i , The calculation results are shown in the following table:

[0180]

[0181] Step 5: Obtain the deep foundation pit stability evaluation score based on the TOPSIS method;

[0182] Obtain the relative closeness C of the deep foundation pit stability at each time period during the monitoring period i The results are as follows:

[0183]

[0184] Step 6: The detailed classification of safety levels under special landforms will be shown in the following table;

[0185]

[0186] The relative closeness C i Bring it into step S62 to obtain the scores of each indicator, and take the average value at the secondary indicator level to obtain the dynamic score of the deep foundation pit structure within the period as follows: Figure 5 :

[0187] Finally, combined with the environmental monitoring and evaluation scores of the deep foundation pit surrounding area, the total score P of the deep foundation pit stability is calculated by the formula shown in step S63, and the comprehensive grade is evaluated according to the evaluation score. The dynamic monitoring score of the deep foundation pit stability within the period is as follows: Figure 6 :

[0188] Step 7: Predict the deep foundation pit stability evaluation score based on the neural network model;

[0189] The secondary indicators C1~C9 monitoring indicators and time are set as the input layer, the deep foundation pit stability evaluation score is set as the output layer, and an implicit intermediate layer is set. The hierarchical structure is as follows: Figure 4 shown.

[0190] The previous monitoring and evaluation data of deep foundation pit stability is used as the training set, and the dynamic monitoring score of deep foundation pit stability is used as the output layer. From the middle layer to the output layer, the regression function is introduced to obtain the predicted value of deep foundation pit stability: the regression function model is:

[0191] y=∑w i x i +b;

[0192] Substitute the deep foundation pit monitoring data into the formula of step S74 to obtain the decrease of the model loss rate under the learning rate condition as follows: Figure 7 ;

[0193] According to the grade subdivision table shown in step S61, the prediction results of the neural network model are converted into the deep foundation pit stability prediction grade, thereby achieving a dynamic cycle early warning effect. The specific scores and prediction grades are as follows Figure 8 ;

[0194] According to the above data monitoring and prediction results, it can be concluded that the deep foundation pit structure score of the project took appropriate measures in time after the warning, and the overall stability score also has a dynamic monitoring effect and gives a warning value. At the same time, the advance prediction of the deep foundation pit stability evaluation score based on the neural network also has a regulatory effect on the construction progress and key tasks of the project, and its cyclic warning mode can also continuously expand and update the neural network model training data set. The effectiveness of the foundation pit monitoring and dynamic warning has been confirmed because its evaluation results are consistent with the actual engineering situation and can provide good guidance for foundation pit construction.

[0195] Finally, it should be emphasized that the above-listed embodiments are only for illustrating the technical solution of the present invention, rather than limiting it. The above description is only a preferred embodiment of the present invention. It should be noted that

[0196] The above is only a preferred embodiment of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should all be included in the scope of the claims of this application.

Claims

1. A dynamic early warning method for monitoring the stability of a deep foundation pit taking into account advance prediction, characterized in that: The specific steps include: S1. Construct a monitoring index system and measurement system for deep foundation pits under floodplain landforms; S2. Determine the subjective weight of monitoring indicators based on the improved G1 method; S3. Modify the indicator weights based on the CRITIC method; S4. Determine the comprehensive weight of deep foundation pit monitoring indicators; S5. Determine the deep foundation pit stability evaluation score based on the TOPSIS method; S6. Detailed classification of safety levels under special landforms; S7. Predict the deep foundation pit stability evaluation score based on the neural network model and set an early warning.

2. A dynamic early warning method for monitoring stability of deep foundation pits taking into account advance prediction according to claim 1, characterized in that: The step 1 specifically includes: S11. According to the construction characteristics of deep foundation pits under the floodplain landform, the main deformation conditions, settlement observation data and geological structure characteristics during the construction process are monitored, including the deformation of the support structure and the surrounding environment and the monitoring of the deep-level changes of the foundation pit, and a step-by-step evaluation index system is constructed to determine the various index factors that affect the stability of the deep foundation pit; S12. Build a multi-level indicator evaluation model that affects the safety and stability of deep foundation pits. According to the monitoring needs of deep foundation pits in floodplain landforms, the first level includes engineering monitoring data of the main structure of the pit and the surrounding environment; the second level indicators include groundwater level, deep-level changes, horizontal and vertical displacements of structures, and measured data of pipeline settlement to evaluate the impact of the underground environment on the stability of the foundation pit; S13. Monitoring points are set up in the submarine and landforms of the project site, and the monitoring data sets of underwater shoals and floodplains are expanded according to the requirements of on-site construction. The monitoring content mainly includes the horizontal displacement of the pit top, the vertical displacement of the pit top, the deep horizontal displacement, the axial force, the groundwater level, and the settlement of surrounding buildings. Laser displacement sensors are arranged at the top of the foundation pit slope within 10-20m around the foundation pit. At the same time, for the slope with strengthened support, an additional set of laser displacement sensors is arranged within 1m of the top of the slope to comprehensively monitor the The horizontal and vertical displacements of the top of the foundation pit slope are monitored; monitoring targets are set on the main load-bearing components of the buildings around the foundation pit, and laser prisms are set on the base plane of the building. The surrounding environmental settlement and other data are monitored through AI cameras; the monitoring system is densely arranged in the seawater cofferdam area of ​​the deep foundation pit, and different instrument combinations such as pressure gauges, static levels or flow meters are set to monitor the water level changes or flow rate changes on both sides of the seawater cofferdam. At the same time, a SwY-20 steel ruler water level meter with an observation accuracy of 5mm is set in the foundation pit to monitor the groundwater level.

3. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 2 specifically includes: S21. Starting from the bottom of the indicator hierarchy, annotate the weight of each indicator factor at the same level relative to the indicator at the previous level, and make a certain judgment on the relative importance of each indicator at each level; Assuming that the indicator system consists of three levels: target layer, primary indicators and secondary indicators, firstly, the weight of the secondary indicators relative to the primary indicators is calculated, i.e., indicators X1, X2,…, X m The weight relative to indicator X is denoted as Indicators Y1, Y2, ... Y n The weight relative to Y is denoted as Then, the weights of the first-level indicators X and Y relative to the target layer are calculated and recorded as S22. Invite i experts to score the importance of adjacent indicators at the same level. The scores range from 1 to 10. The higher the corresponding value, the more important the indicator represented. The scoring criteria are shown in Table 1: S23. Sort the indicators that have been evaluated for importance; according to the importance of the indicators scored by the jth expert, sort them according to the size of the scores, such as the importance score of indicator X1 Greater than index X n Importance score Then record it as X1≥X n , the order of importance is: X1≥X2≥…≥X n ; S24. According to the score value of the jth expert on the importance of the indicator, calculate the ratio of the weights of adjacent indicators at the same level, and evaluate the indicator X. k-1 With X k The weight ratio is recorded as: In the above formula: It is represented by the index X based on the score of the jth expert. k The weight of S25. Based on the score of the jth expert, calculate the index X n Weight for: The weights of other indicators can be obtained by the recursive formula converted in step S24, as follows:

4. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 3 specifically includes: S31. For index x j Calculate the standard deviation, denoted as σ j , x ij is the original value of the i-th indicator of the j-th evaluation object; is the average value of the indicators in m evaluation objects, and the calculation steps are as follows: S32. Construct the correlation coefficient matrix R of n indicators = (r ij ) n×n , r ij is the vector x i With x j The Pearson correlation coefficient is calculated as follows: S33. Let Q i is the amount of information contained in the i-th indicator, and its expression is: S34. Utilize information quantity Q i Calculate the objective weight of the indicator 5. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 4 specifically includes: The weights obtained by the improved G1 method and the CRITIC method are combined using the multiplication normalization method to obtain the combined weight w i , the calculation steps are as follows:

6. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 5 specifically includes: S51. Construct an initial matrix based on the sample and perform vector normalization to obtain a standard decision matrix denoted as Y = (y ij ) m×n ; S52. Substitute the combined weights obtained in step S4 to construct a weighted standard matrix Z = (z ij ) m×n , the weighted canonical matrix element calculation formula is: z ij =w i y ij , w i To improve the combined weights of G1 and CRITIC methods; S53. Determine positive and negative ideal solutions and S54. Based on positive and negative ideal solutions and Find the Euclidean distance and That is, the distance between the actual value of the evaluation indicator and the optimal and worst values ​​of the indicator: S55. Calculate the relative closeness C of each evaluation index to the optimal value i , C i The larger the value, the closer it is to the optimal level, and vice versa:

7. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 6 specifically includes: S61. Define the corresponding relationship between relative closeness, evaluation level and evaluation score as shown in Table 2: Table 2 defines the corresponding relationship between relative closeness, evaluation level and evaluation score Among them: EL is the ratio of direct economic loss to total investment; CL is the ratio of lost construction period to planned construction period; MW is the number of minor injuries; SW is the number of serious injuries; D is the number of deaths; S62. In order to make the score evaluation more precise, the corresponding relationship between the relative closeness and the evaluation score defined in step S61 is expressed as follows: S63. Calculate the total evaluation score based on the weights of different indicators. The formula is as follows: Among them, w i represents the combined weight of the i-th indicator, p i It represents the evaluation score of the ith indicator, m is a selected positive real number, and P is the final score value.

8. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The step 7 specifically includes: S71. Perform neural network modeling according to the multi-level indicator evaluation model built in step S13, confirm the multi-level prediction system of the input layer, the intermediate layer and the output layer, and the final output layer, i.e., the prediction result, is the comprehensive evaluation score of the deep foundation pit stability obtained by the step S63 target; Among them, the calculation process from the input layer to the middle layer and from the middle layer to the input layer is forward propagation, and the process of comparing the output result with the expected output result and using the network to feed back the generated error is back propagation; S72. From the input layer to the middle layer, the Sigmoid activation function is introduced to normalize the output of each neuron: S73. From the middle layer to the output layer, the regression function is introduced to obtain the predicted value of deep foundation pit stability: y=∑w i x i +b; Where: x i is the input value of the first-level indicator in the middle layer, w i is the learning weight of the neural network, and b is the intercept term; S74. The back propagation during iterative calculation uses the gradient descent learning method to determine the loss function of the optimization target: S75. Through iterative model learning, the dynamically updated deep foundation pit stability monitoring data set is input, the comprehensive evaluation score of the deep foundation pit stability under the floodplain landform is predicted in advance, and the evaluation level is refined according to the table of step S61 to achieve a dynamic cycle early warning effect.

9. The method for dynamic early warning of deep foundation pit stability monitoring considering advance prediction according to claim 1 is characterized in that: The dynamic connection between step S1 and step S5, S7 data reuse is: The data collected in step S1 are preprocessed and then a stability evaluation score is obtained in step S5; The original data acquired by monitoring in step S1 is input again, and the evaluation score in step S5 is used as the output target, and the evaluation score of the next period is predicted through the neural network model in step S7; When the data in step S1 is reused in step S7, it must be entered along with the update of the monitoring data of the whole process of deep foundation pit construction, that is, the data set in step S7 has dynamic update capability over step S5; In step S7, the ratio of the training set, validation set, and test set in the data set entered is 7:2:

1.

10. A dynamic early warning method for monitoring stability of deep foundation pits taking into account advance prediction according to any one of claims 1 to 9, characterized in that: The early warning method is applicable to floodplain landforms, which refer to the ground portion where the land slopes gently toward the sea and naturally extends, with the bottom material being gravel, clay and silty sediments.

Images