Risk assessment method of water inrush in tunnel construction based on multi-source information fusion

Abstract

The present invention relates to the technical field of tunnel engineering disaster prediction, and specifically to a method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion, comprising the following steps: S1: visual inspection, monitoring and measurement, and comprehensive advanced geological forecasting of the tunnel construction site, establishing a corresponding single-information source risk assessment model based on the obtained data and conducting risk assessment to obtain a risk assessment value of the single information source; S2: evaluating the three single-information source risk assessment models obtained in S1 to obtain the importance weight and credibility of each single-information source risk assessment model; S3: fusing the risk assessment values, importance weights, and credibility of the three single-information source risk assessment models through the ER evidence theory to obtain the final probability of sudden water inrush risk. The risk assessment method proposed by the present invention fully considers the risk factors affecting sudden water inrush in tunnel construction, and has better robustness and accuracy than the single-information assessment method.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering disaster prediction, and in particular to a tunnel construction water inrush risk assessment method based on multi-source information fusion. Background Art

[0002] In recent years, with the continuous development of regional economies and the improvement of transportation networks, tunnel construction, as a key control project for regional economic development and transportation networks, has also seen rapid growth. Various major accidents and disasters can occur during tunnel construction. Water inrush is one of the more serious geological hazards that can occur. Existing tunnel construction often involves tunnels traversing karst areas with complex geological structures, resulting in a large number of high-risk, deep, and long tunnel projects. Tunnel excavation disturbs groundwater, making water inrush highly likely. Water inrush during tunnel construction not only causes casualties and economic losses, but also poses significant challenges to tunnel construction.

[0003] To ensure tunnel construction safety, avoid casualties, and reduce unnecessary economic losses, it is crucial to predict water inrush and select appropriate preventive measures. Currently, tunnel water inrush predictions focus on analyzing forecast information, geological conditions, and hydrological data. However, these methods are inaccurate and lack universal applicability. Tunnel water inrush is a complex, multifactorial process, making it difficult to assess and categorize based on a single piece of information. Therefore, it is necessary to establish a judgment standard based on comprehensive feedback from multiple information sources. It is difficult to establish an accurate tunnel water inrush prediction model based on a single piece of information. Summary of the Invention

[0004] Therefore, the technical problem to be solved by the present invention is to overcome the defect in the prior art that it is difficult to establish an accurate tunnel water inrush prediction model by relying on only a single piece of information, thereby providing a tunnel construction water inrush risk assessment method based on multi-source information fusion.

[0005] The tunnel construction water inrush risk assessment method based on multi-source information fusion includes the following steps:

[0006] S1: Conduct visual inspections, monitoring measurements, and comprehensive advanced geological forecasts at the tunnel construction site. Based on the obtained data, establish corresponding single-source risk assessment models and conduct risk assessments to obtain single-source risk assessment values.

[0007] S2: Evaluate the three single-source risk assessment models obtained in S1, obtain the importance weight of each single-source risk assessment model based on its accuracy, and obtain the credibility of each single-source risk assessment model based on its uncertainty.

[0008] S3: The risk assessment values, importance weights, and credibility of the three single-information-source risk assessment models are integrated through the ER evidence theory to obtain the final probability of sudden water inrush risk.

[0009] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the process of establishing the risk assessment model of the visual inspection single information source in S1 includes the following steps:

[0010] Based on the actual conditions of tunnel construction sites and existing cases of water inrush in tunnels, several risk factors for visual inspection were screened and divided into four risk levels according to their risk magnitude. Existing water inrush in tunnels were classified according to risk factors and risk levels to obtain a water inrush database. This database was used as the training set for a support vector machine (SVM). A visual inspection single-information source risk assessment model was obtained using the SVM. The discriminant function of the SVM is as follows:

[0011]

[0012] Where: m is the size of the training data set;

[0013] α i is the Lagrange multiplier;

[0014] K(x i , x) is the kernel function;

[0015] b is the threshold parameter based on the training set.

[0016] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the following steps are used to extract the relevant probability from the output value of the support vector machine:

[0017] Use the Sigmoid function to map the output of the support vector machine to the interval [0, 1]. The specific formula is as follows:

[0018]

[0019] Where: a and b are the parameters obtained by minimizing the negative log-likelihood function for a set of training examples;

[0020]

[0021] Where: t i A new label for the class;

[0022] t + t i +1;

[0023] t- is t i -1;

[0024] N + is the number of points belonging to class 1;

[0025] N - is the number of points belonging to class 2.

[0026] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the monitored and measured data in S1 include surrounding rock displacement, stress and seepage water pressure, which are divided into four risk levels according to the risk level of the obtained monitored and measured data.

[0027] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the process of establishing a risk assessment model for monitoring and measuring a single information source includes the following steps:

[0028] The cloud model is used to convert the data obtained from monitoring and measurement into a sudden water inrush risk probability value. The sudden water inrush risk probability value of the monitoring and measurement data is integrated using the ER evidence theory to obtain a monitoring and measurement single information source risk assessment model. The cloud model establishment process is as follows:

[0029] Consider various tunnel collapse risk factors in the decision-making process i Analyze and further divide each risk factor into different risk states B ij (i=1,2,…,M;j=1,2,…,N), each risk state corresponds to a specific double limit interval, denoted as [b ij (L), b ij (R)], the double limit interval [b ij (L), b ij (R)] converted to cloud model (Ex ij ,En ij ,He ij )

[0030]

[0031] Where: Ex ij For expectations;

[0032] En ij For Ex ij Entropy;

[0033] He ij is super entropy;

[0034] h is a constant, and its value range is 0~En ij ;

[0035] In formula (4), x should meet three requirements: first, x∈X, second, x is a random instantiation of concept B, and third, x satisfies the formula. The degree of certainty that x belongs to concept B can be obtained by the following formula:

[0036]

[0037] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the method for obtaining the comprehensive advanced geological prediction in S1 includes a long-distance detection method, a short-distance detection method, and an advanced drilling method, and the comprehensive advanced geological prediction is divided into four levels according to the TGP results.

[0038] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the process of establishing a comprehensive advanced geological forecast single information source risk assessment model includes the following steps:

[0039] The comprehensive advanced geological forecast was analyzed by the expert scoring method to obtain the sudden water inrush risk probability values ​​of different advanced geological forecast methods. The sudden water inrush risk probability values ​​of each advanced geological forecast method were integrated using the ER evidence theory to obtain a comprehensive advanced geological forecast single information source risk assessment model. The ER evidence theory formula is as follows:

[0040]

[0041] Among them, (θ,p θ,j ) is an element of evidence j , indicating that the probability that the evidence points to the proposition θ is p θ,j , θ is any subset of Θ.

[0042] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the importance weight in S2 is obtained by the assessment accuracy of the single information source risk assessment model, and its formula is as follows:

[0043]

[0044] Where: is the class-wise precision;

[0045] is the class-wise recall rate;

[0046] is the class-wise F1-score;

[0047] TP is the case of accurate prediction.

[0048] As a preferred method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion in the present invention, the credibility in S2 is obtained by aggregated uncertainty, and its formula is as follows:

[0049]

[0050]

[0051] Where: is the focal element θ i Trust

[0052] G is the normalization factor;

[0053] n is the total number of focus elements;

[0054] r j For evidence j credibility.

[0055] The technical solution of the present invention has the following advantages:

[0056] This paper proposes a water inrush risk assessment method that comprehensively considers on-site construction conditions, monitoring and measurement data, and comprehensive advanced geological forecasts. It uses different risk assessment models to evaluate multiple information sources and employs the ER evidence theory rules to fuse risk assessment values, credibility, and importance weights to derive a final water inrush risk probability. Compared to single-source risk assessment methods, this proposed risk assessment method fully considers the risk factors that influence water inrush in tunnel construction, resulting in greater robustness and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0059] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0060] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0061] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0062] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0063] Example 1

[0064] This embodiment provides a tunnel construction water inrush risk assessment method based on multi-source information fusion, which is used to provide construction suggestions for tunnel construction sites, reduce the occurrence of water inrush accidents, and thus reduce casualties and economic losses.

[0065] like Figure 1 As shown, the specific steps include:

[0066] S1: Conduct visual inspections, monitoring measurements, and comprehensive advanced geological forecasts at the tunnel construction site. Based on the obtained data, establish corresponding single-source risk assessment models and conduct risk assessments to obtain single-source risk assessment values.

[0067] S2: Evaluate the three single-information source risk assessment models obtained in S1, obtain the importance weight of each single-information source risk assessment model through the accuracy of the single-information source risk assessment model, and obtain the credibility of each single-information source risk assessment model through the uncertainty of the single-information source risk assessment model.

[0068] S3: The risk assessment values, importance weights, and credibility of the three single-information-source risk assessment models are integrated through the ER evidence theory to obtain the final probability of sudden water inrush risk.

[0069] In the visual inspection of S1 in this embodiment, 11 risk factors for visual inspection were screened based on the actual conditions of the tunnel construction site and at least 80 cases of sudden water inrush in tunnels. These risk factors were then divided into four risk levels according to their risk magnitude, as shown in Table 1. Existing sudden water inrush cases in tunnels were classified according to the risk factors and risk levels to obtain a sudden water inrush database. The sudden water inrush database was used as a training set for a support vector machine, and the support vector machine was used for learning to obtain a visual inspection single information source risk assessment model. The discriminant function of the support vector machine is specifically in the following form:

[0070]

[0071] Where: m is the size of the training data set;

[0072] α i is the Lagrange multiplier;

[0073] K(x i , x) is the kernel function;

[0074] b is the threshold parameter based on the training set.

[0075] However, the linear support vector machine only gives a class prediction output of "yes" or "no". In order to extract the relevant probability from the support vector machine output, Platt's method is used in this embodiment. This method uses the Sigmoid function to map the output of the support vector machine to the interval [0, 1]. The specific formula is as follows:

[0076]

[0077] Where: a and b are the parameters obtained by minimizing the negative log-likelihood function for a set of training examples;

[0078]

[0079] Where: t i A new label for the class;

[0080] t + t i +1;

[0081] t- is t i -1;

[0082] N + is the number of points belonging to class 1;

[0083] N - is the number of points belonging to class 2.

[0084] Table 1 Visual inspection risk levels

[0085]

[0086]

[0087] Among them, BQ stands for basic rock mass quality index;

[0088] CRD stands for Cross-Septal Method;

[0089] CD stands for Single Sidewall Drilling;

[0090] Bench stands for step method;

[0091] Full face represents the full-section excavation method.

[0092] In the monitoring and measurement of this embodiment, the main monitored data include surrounding rock displacement, stress, and seepage water pressure. The obtained monitoring and measurement data are divided into four risk levels according to their risk level, as shown in Table 2. The process of establishing a risk assessment model for a single information source of monitoring and measurement includes the following steps:

[0093] The cloud model is used to convert the data obtained from monitoring and measurement into a sudden water inrush risk probability value. The sudden water inrush risk probability value of the monitoring and measurement data is integrated using the ER evidence theory to obtain a monitoring and measurement single information source risk assessment model. The cloud model is obtained by transforming the normal cloud formula. The normal cloud is defined as follows:

[0094] Given a quantitative domain X, if B is a qualitative concept on X, then x satisfies: 1. x∈X; 2. x is a random instantiation of concept B; 3. x satisfies the formula, and the degree of certainty that x belongs to concept B can be obtained by formula (6).

[0095]

[0096]

[0097] Various tunnel collapse risk factors were analyzed during the decision-making process. i In order to mine useful information from multiple sources, each risk factor should be further divided into different risk states B ij (i=1,2,…,M;j=1,2,…,N). Each risk state can correspond to a specific double limit interval, denoted as [b ij (L), b ij (R)]. The double limit interval [b ij (L), b ij (R)] into a normal cloud model (Ex ij ,En ij ,He ij ) formula (5).

[0098]

[0099] Where: Ex ij For expectations;

[0100] En ij For Ex ij Entropy;

[0101] He ij is super entropy;

[0102] h is a constant, and its value range is 0~En ij , which is suitable for reflecting the uncertainty of these factors. In this embodiment, h is taken as 0.002.

[0103] In the CM framework, correlation can be measured by factor B i The observed value b ij With specific risk status B ij The relative membership between the cloud models. The BPA of the influencing factors under different risk states can be obtained by formula (14).

[0104]

[0105] Where: m i (B j ) is the BPA value of the influencing factor under different risk conditions;

[0106] En′ satisfies the conditions En′~N(En,He 2 )’s random number;

[0107] m i (Φ) is the BPA value under uncertainty, that is, i All elements are included if no focus element can be identified under the indicator.

[0108] Table 2 Risk levels of monitoring measurement data

[0109]

[0110] Here, σ represents the design tensile and compressive strength of the steel.

[0111] In this embodiment of the comprehensive advanced geological forecast, the comprehensive advanced geological forecast method includes long-distance detection (tunnel seismic wave exploration), short-distance detection, and advanced drilling. Based on the TGP results, the comprehensive advanced geological forecast is divided into four risk levels as shown in Table 3. The process of establishing the comprehensive advanced geological forecast single information source risk assessment model includes the following steps:

[0112] The comprehensive advanced geological forecast was analyzed by the expert scoring method to obtain the sudden water inrush risk probability values ​​of different advanced geological forecast methods. The sudden water inrush risk probability values ​​of each advanced geological forecast method were integrated using the ER evidence theory to obtain a comprehensive advanced geological forecast single information source risk assessment model. The ER evidence theory formula is as follows:

[0113]

[0114] Among them, (θ,p θ,j ) is an element of evidence j , indicating that the probability that the evidence points to the proposition θ is p θ,j , θ is any subset of Θ.

[0115] In the ER rule, the evidence e is defined j Credibility r j and weight w j Credibility j The ability to represent the source of information. The credibility of evidence is an inherent property of evidence. The weight of evidence w j It can be used to reflect its importance relative to other evidence. The importance of weighted confidence distribution to credibility is defined as shown in formula (15).

[0116]

[0117] in Measuring e j The degree of support for θ, taking into account both importance and credibility, is shown in formula (16).

[0118]

[0119] where m θ,j =w j p θ,j , c rw,j =1 / (1+w j -r j ) is the normalization factor and satisfies w j ∈[0,1] is evidence e j The importance weight, r j ∈[0,1] is evidence e j credibility.

[0120] If two pieces of evidence e1 and e2 are independent of each other, the combined belief degree of e1 and e2 supporting the proposition θ is p through the ER fusion rule. θ,e(2) , as shown in formula (17) and formula (18).

[0121]

[0122] Where D is all subsets; B and C are two independent pieces of evidence.

[0123] Table 3 Classification of comprehensive advanced geological prediction indicators

[0124]

[0125]

[0126] In the multi-source information fusion method of this embodiment, the assessment quality of three single-source risk assessment models is evaluated. The quality of each assessment model is evaluated by importance weight and credibility. The importance weight of each single-source risk assessment model is obtained by the accuracy of the single-source risk assessment model, and the credibility of each single-source risk assessment model is obtained by the uncertainty of the single-source risk assessment model. Then, the risk assessment values, importance weights, and credibility of the three single-source risk assessment models are fused using the ER evidence theory to obtain the final sudden water inrush risk probability. The formula for determining the importance weight and credibility of the assessment model is as follows:

[0127] Importance Weight: In this example, a threshold of 0.01 was chosen to maintain the presence of each selected piece of evidence. A minimum class-wise F1-score was set to limit the contribution of low-quality evidence. Finally, the importance weight of each piece of evidence in the evaluation model was set to max{min{class-wise F1-scores}, 0.01}. Class-wise F1-score is a metric that combines precision and recall.

[0128]

[0129] Where: is the class-wise precision;

[0130] is the class-wise recall rate;

[0131] is the class-wise F1-score;

[0132] TP is the case of accurate prediction.

[0133] Credibility: In this embodiment, aggregate-uncertainty (AU) is used to calculate the credibility level of each body of evidence learned from the probabilistic classification model.

[0134]

[0135] Where: is the focal element θ i Trust

[0136] G is the normalization factor;

[0137] n is the total number of focus elements;

[0138] r j For evidence j credibility.

[0139] In this embodiment, the ER rule is used to fuse the different risk assessment values, importance graph weights, and credibility obtained from the three single-source risk assessment models mentioned above. The rule (min{class-wise F1-scores}) selects importance weights based on the accuracy of each model. Since there were no cases at the beginning, the importance weights of each model were consistent (w = [0.333; 0.333; 0.333]). As tunnel construction progresses, cases continue to accumulate. The importance weight of each model is continuously modified according to the rule (min{class-wise F1-scores}). This optimizes the fusion model and improves the accuracy of the evaluation.

[0140] Example 2

[0141] In this example, five tunnel end faces were selected as test samples and the fusion results of ER rule and DS theory were compared (e.g. Figure 1 As shown in Table 4, E1, E2, and E3 are the confidence levels of the visual inspection model, monitoring and measurement model, and comprehensive advanced geological prediction model, respectively. The bold numbers in the table are the risk levels obtained by the corresponding prediction models. The following conclusions can be drawn

[0142] (1) Taking tunnel section DK203+050 as an example, the basic probability distribution of the three models is E1 = [0.43, 0.57, 0, 0]; E2 = [0.01, 0.99, 0, 0]; E3 = [0, 0.72, 0.28, 0]. The importance weight is w = [0.314; 0.433; 0.627]; the credibility of the model is r = [0.507; 0.960; 0.572]. The reduced probability value is The corresponding power set is The normalization factor is According to formula (18), the probability distribution considering the credibility and importance weights is obtained

[0143] The probability distribution of the three models is fused using the ER rule, and the fused probability value is , where e(3) represents the combination of three pieces of evidence. Finally, after removing the power set term, the collapse risk probability is The collapse risk level of this section is "II" (water seepage).

[0144] (2) The multi-information fusion method proposed in this embodiment can improve the accuracy of sudden water inrush risk assessment and reduce uncertainty. Multi-source information fusion assessment technology has higher accuracy than single-source information risk assessment technology.

[0145] (3) As cases accumulate, the fusion model continuously improves its accuracy by revising the importance weights. Taking the No. 3 tunnel section as an example, different information sources give different evaluation results. Initially, the results of the fusion model do not match the actual situation. As cases accumulate, the values ​​of the importance weights are revised (w = [0.314; 0.433; 0.627]). The calculation results of the improved fusion model are basically consistent with the actual situation on site.

[0146] (4) When the evaluation results of three different information sources are inconsistent (such as tunnel section No. 2), the fusion result of the ER rule is better than that of the DS theory. The DS theory only accumulates consensus support. If a proposition is opposed by any evidence, even if it is supported by other evidence, it will be completely rejected. Therefore, when the evaluation results of three single information sources are different, the ordinary DS theory will give a fusion result that is contrary to common sense. However, the ER rule fully considers the importance weight and credibility of the model when merging high-conflict information sources. Therefore, it can judge which model is more likely to represent the actual situation, and the results given can be closer to reality.

[0147] Table 4 Fusion results of five test samples

[0148]

[0149]

[0150] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A tunnel construction water inrush risk assessment method based on multi-source information fusion is characterized by: The steps include: S1: Conduct visual inspections, monitoring measurements, and comprehensive advanced geological forecasts at the tunnel construction site. Based on the obtained data, establish corresponding single-source risk assessment models and conduct risk assessments to obtain single-source risk assessment values. S2: Evaluate the three single-source risk assessment models obtained in S1, obtain the importance weight of each single-source risk assessment model based on its accuracy, and obtain the credibility of each single-source risk assessment model based on its uncertainty. S3: The risk assessment values, importance weights, and credibility of the three single-source risk assessment models are integrated through the ER evidence theory to obtain the final water inrush risk probability; The methods for obtaining the comprehensive advanced geological prediction in S1 include long-distance detection method, short-distance detection method and advanced drilling method, and the comprehensive advanced geological prediction is divided into four levels according to the TGP results; The process of establishing the risk assessment model of comprehensive advanced geological forecast information source includes the following steps: The comprehensive advanced geological forecast was analyzed by the expert scoring method to obtain the sudden water inrush risk probability values ​​of different advanced geological forecast methods. The sudden water inrush risk probability values ​​of each advanced geological forecast method were integrated using the ER evidence theory to obtain a comprehensive advanced geological forecast single information source risk assessment model. The ER evidence theory formula is as follows: Among them, (θ,p θ,j ) is an element of evidence j , indicating that the probability that the evidence points to the proposition θ is p θ,j , θ is any subset of Θ; The importance weight in S2 is obtained through the evaluation accuracy of the single information source risk assessment model, and its formula is as follows: Where: is the class-wise precision; is the class-wise recall rate; is the class-wise F1-score; TP is the case of accurate prediction; The credibility in S2 is obtained by the aggregated uncertainty, and its formula is as follows: Where: is the focal element θ i Trust G is the normalization factor; n is the total number of focus elements; r j For evidence j credibility.

2. The method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion according to claim 1 is characterized in that: The process of establishing the risk assessment model for the visual inspection information source in S1 includes the following steps: Based on the actual conditions of tunnel construction sites and existing cases of water inrush in tunnels, several risk factors for visual inspection were screened and divided into four risk levels according to their risk magnitude. Existing water inrush in tunnels were classified according to risk factors and risk levels to obtain a water inrush database. This database was used as the training set for a support vector machine (SVM). A visual inspection single-information source risk assessment model was obtained using the SVM. The discriminant function of the SVM is as follows: Where: m is the size of the training data set; α i is the Lagrange multiplier; K(x i , x) is the kernel function; b is the threshold parameter based on the training set.

3. The method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion according to claim 2 is characterized in that: The following steps are used to extract the relevant probabilities from the support vector machine output: Use the Sigmoid function to map the output of the support vector machine to the interval [0, 1]. The specific formula is as follows: Where: a and b are the parameters obtained by minimizing the negative log-likelihood function for a set of training examples; Where: t i A new label for the class; t + t i +1; t- is t i -1; N + is the number of points belonging to class 1; N - is the number of points belonging to class 2.

4. The method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion according to claim 1 is characterized in that: The monitored and measured data in S1 include surrounding rock displacement, stress and seepage water pressure, which are divided into four risk levels according to the risk level of the obtained monitored and measured data.

5. The method for assessing the risk of sudden water inrush in tunnel construction based on multi-source information fusion according to claim 4 is characterized in that: The process of establishing a risk assessment model for a single monitoring and measurement information source includes the following steps: The cloud model is used to convert the data obtained from monitoring and measurement into a sudden water inrush risk probability value. The sudden water inrush risk probability value of the monitoring and measurement data is integrated using the ER evidence theory to obtain a monitoring and measurement single information source risk assessment model. The cloud model establishment process is as follows: Consider various tunnel collapse risk factors in the decision-making process i Analyze and further divide each risk factor into different risk states B ij (i=1, 2, ..., M; j=1,2,…,N), each risk state corresponds to a specific double limit interval, denoted as [b ij (L), b ij (R)], the double limit interval [b ij (L), b ij (R)] converted to cloud model (Ex ij ,En ij ,He ij ), Where: Ex ij For expectations; En ij For Ex ij Entropy; He ij is super entropy; h is a constant, and its value range is 0~En ij ; In formula (4), x should meet three requirements: first, x∈X, second, x is a random instantiation of concept B, and third, x satisfies the formula. The degree of certainty that x belongs to concept B can be obtained by the following formula:

Images