Tunnel water inrush amount prediction method based on multi-source information fusion

Abstract

The invention discloses a tunnel water inrush prediction method based on multi-source information fusion, and the method comprises the steps: obtaining the geological information along a tunnel through a plurality of methods, and extracting the feature information related to the lithology, the geological structure, the water pressure and the tunnel burial depth, and the position coordinates of the tunnel from the geological information; and carrying out data fusion on the extracted feature information, constructing a training set and a test set of a prediction model by using the fused output data, taking a PSO-RF algorithm model as a prediction model of the water inflow, training and testing the prediction model, and predicting the tunnel water inflow by using the prediction model which is qualified in training and testing. According to the method, the intelligent prediction model and multi-source geological information input are combined to predict the water inflow of the tunnel, and the water inflow prediction accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel water inrush prevention and control, and in particular to a method for predicting the amount of sudden water inrush in a tunnel. Background Art

[0002] Water inrush is the phenomenon that groundwater continuously flows into the site when mining, excavating foundation pits or underground caverns in rock (soil) below the groundwater surface. The sources of water inrush include groundwater (pore water source, fissure water source, karst water source) and surface water source.

[0003] Water inrush is a common engineering geological problem encountered in engineering construction. Water inrush not only brings difficulties to construction, but is also an important cause of instability of surrounding rock of underground caverns, seepage damage of dam foundation pits, or slope collapse. Therefore, it is very important to predict the amount of water inrush during tunnel excavation to ensure construction safety.

[0004] In the prior art, the prediction of water inflow only relies on the monitoring data of water inflow at the construction site, which makes it difficult to accurately predict the water inflow. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method for predicting sudden water inrush in a tunnel based on multi-source information fusion, so as to solve the technical problem of improving the accuracy of water inrush prediction during tunnel construction.

[0006] The method for predicting sudden water inrush in a tunnel based on multi-source information fusion of the present invention is characterized by comprising the following steps:

[0007] 1) Use geological drilling, geological survey, geological mapping, geological radar, seismic and advance drilling methods to obtain geological information along the tunnel;

[0008] 2) extracting characteristic information related to the four attributes of lithology, geological structure, water pressure and tunnel burial depth and the tunnel location coordinates corresponding to the characteristic information from the geological information obtained in step 1), and encoding the extracted characteristic information into characteristic values;

[0009] 3) For the characteristic values ​​obtained in step 2), data fusion is performed on the characteristic values ​​with the same attributes and the same tunnel position coordinates, including:

[0010] Establish recognition framework Θ={θ 1 ,θ 2 ,…,θ k}, θ j represents the specific identification result, that is, it is identified as belonging to a certain category among lithology, geological structure, water pressure and tunnel depth; the characteristic value obtained in step 2) is used as evidence, and each piece of evidence E 1 、E 2 ,…,E iThe corresponding probability distribution function is m 1 、m 2 ,…,m i , that is, m i (θ j )=P(E i ∣θ j ), each m i The probability distribution defined on the recognition frame Θ represents the evidence E i The support degree for the recognition result; A is the corresponding focal element, and The fused probability distribution function m is calculated by the following formula:

[0011]

[0012] Among them, K is a normalization constant used to ensure that the sum of the fused probability distribution function is 1:

[0013]

[0014] ④Calculate the average evidence:

[0015]

[0016] In the formula, k is the number of focal elements; n is the number of evidences;

[0017] ⑤Calculate the distance between a single piece of evidence and the average evidence:

[0018]

[0019] The larger the distance, the smaller the correlation between the two pieces of evidence. In order to make high-trust evidence obtain a high weight coefficient, the constructor makes the weight coefficient of the evidence equal to the distance d. i Inversely proportional, the mass function of different focal elements is:

[0020]

[0021] ③ In order to objectively and truly reflect the importance of each source of evidence, the entropy method is used to determine the weight of each evidence. The weight coefficient expression is:

[0022]

[0023] In the formula,

[0024] ⑥Use weight coefficient ω i The weight of the initial evidence source is modified, that is,

[0025]

[0026] The new evidence mnew As the evidence source, substitute it into formula (1) to perform pairwise fusion, and output a fused data each time;

[0027] The four attributes corresponding to the same tunnel location coordinates are fused into a sequence {a 1 , a 2 , a 3 , a 4}, where a 1 Indicates the lithology fusion information, a 2 represents the geological structure fusion information, a 3 represents water pressure fusion information, a 4 It represents the fusion information of tunnel depth, and then obtains a data set composed of several series;

[0028] 4) Using the PSO-RF algorithm model as the water inflow prediction model, the prediction model is trained and tested:

[0029] The data set obtained in step 3) is divided into a training set and a test set, the PSO-RF algorithm model is trained with the training set, and the trained model is tested with the test set. A sequence is used as an input of the PSO-RF algorithm model, and the goodness of fit and mean square error are selected as evaluation indicators of model training to obtain a prediction model that has passed the training and testing;

[0030] 5) Use the qualified prediction model obtained in step 4) to predict the tunnel water inflow.

[0031] Beneficial effects of the present invention:

[0032] The present invention is a method for predicting sudden water inrush in a tunnel based on multi-source information fusion. A variety of geological information along the tunnel is obtained through a variety of detection methods, and the various detection methods or geological information are fused through a multi-source information fusion method to obtain a fused data set that can reflect the water inrush in the tunnel. Then, a water inrush prediction model with a good fit between the predicted value and the true value is trained through the fused data set. The water inrush prediction is performed using the prediction model, and combined with multi-source geological information input, the accuracy of water inrush prediction can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is the overall flow chart of tunnel water inrush prediction based on multi-source information fusion.

[0034] Figure 2 This is the flow chart of water inflow prediction model training based on PSO-RF algorithm.

[0035] Figure 3 Prediction results for the test set. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0037] The method for predicting sudden water inrush in a tunnel based on multi-source information fusion in this embodiment includes the following steps:

[0038] 1) Use geological drilling, geological survey, geological mapping, geological radar, seismic and advance drilling methods to obtain geological information along the tunnel. The geological information includes geological profile information, topographic information, face exposure information and geophysical information.

[0039] 2) Extract characteristic information related to the four attributes of lithology, geological structure, water pressure and tunnel burial depth and the tunnel location coordinates corresponding to the characteristic information from the geological information obtained in step 1), and encode the extracted characteristic information into characteristic values.

[0040] For example: the depth of the bottom of the rock layer, the thickness of the rock layer, and the geological properties of the rock layer included in the geological profile information obtained by various detection methods are characteristic information related to the lithology, the terrain information includes the terrain undulations and the distribution of surface water systems, which are information related to the geological structure, the surface water level and groundwater level included in the terrain information are information related to water pressure, the face exposure information also includes information related to lithology, geological structure and water pressure, the geophysical information includes the detected fault information, fracture distribution information, and water-rich area information, which belongs to the geological structure information, and the detected water level in the water-rich area belongs to different water pressure information. An example of extracting characteristic information from the address information obtained by the drilling method is shown in Table 1 below:

[0041] Table 1 LSSK51 drilling information extraction results

[0042]

[0043] The tunnel location coordinates can be extracted with the help of CAD software. An example of tunnel coordinate location extraction is shown in Table 2 below:

[0044] Table 2 Three-dimensional coordinates of a tunnel route

[0045]

[0046] 3) For the characteristic values ​​obtained in step 2), data fusion is performed on the characteristic values ​​with the same attributes and the same tunnel position coordinates, including:

[0047] Establish recognition framework Θ={θ 1 ,θ 2 ,…,θ k}, θ j represents the specific identification result, that is, it is identified as belonging to a certain category among lithology, geological structure, water pressure and tunnel depth; the characteristic value obtained in step 2) is used as evidence, and each piece of evidence E1 、E 2 ,…,E i The corresponding probability distribution function is m 1 、m 2 ,…,m i , that is, m i (θ j )=P(E i ∣θ j ), each m i The probability distribution defined on the recognition frame Θ represents the evidence E i The support degree for the recognition result; A is the corresponding focal element, and The fused probability distribution function m is calculated by the following formula:

[0048]

[0049] Among them, K is a normalization constant used to ensure that the sum of the fused probability distribution function is 1:

[0050]

[0051] ⑦Calculate the average evidence:

[0052]

[0053] In the formula, k is the number of focal elements; n is the number of evidences;

[0054] ⑧Calculate the distance between a single piece of evidence and the average evidence:

[0055]

[0056] The larger the distance, the smaller the correlation between the two pieces of evidence. In order to make high-trust evidence obtain a high weight coefficient, the constructor makes the weight coefficient of the evidence equal to the distance d. i Inversely proportional, the mass function of different focal elements is:

[0057]

[0058] ③ In order to objectively and truly reflect the importance of each source of evidence, the entropy method is used to determine the weight of each evidence. The weight coefficient expression is:

[0059]

[0060] In the formula,

[0061] ⑨Use weight coefficient ω i The weight of the initial evidence source is modified, that is,

[0062]

[0063] The new evidence m new As the evidence source, substitute it into formula (1) to perform pairwise fusion, and output a fused data each time;

[0064] The four attributes corresponding to the same tunnel location coordinates are fused into a sequence {a 1 , a 2 , a 3 , a 4}, where a 1 Indicates the lithology fusion information, a 2 represents the geological structure fusion information, a 3 represents water pressure fusion information, a 4 Indicates the tunnel depth fusion information, and then obtains a data set composed of several series. The data samples are described in Table 3 below:

[0065] Table 3 Fusion data samples

[0066]

[0067] 4) Using the PSO-RF algorithm model as the water inflow prediction model, the prediction model is trained and tested:

[0068] The data set obtained in step 3) is divided into a training set and a test set. The training set is used to train the PSO-RF algorithm model, and the test set is used to test the trained model. A sequence is used as an input of the PSO-RF algorithm model. The goodness of fit and mean square error are selected as evaluation indicators for model training to obtain a prediction model that has passed the training and testing. The model training process is as follows: Figure 2 The test results are shown in Figure 3 As shown in the figure, the goodness of fit between the predicted value and the true value (R 2 ) and mean square error (MSE) are 0.9711 and 4.3244 respectively, and the fitting effect is good, which shows that the model has achieved a high prediction accuracy, thus verifying the feasibility of the model for predicting water inflow at the tunnel face.

[0069] 5) Use the trained prediction model to predict the tunnel water inflow.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting sudden water inrush in tunnels based on multi-source information fusion, characterized by: The following steps are involved: 1) Use geological drilling, geological survey, geological mapping, geological radar, seismic and advance drilling methods to obtain geological information along the tunnel; 2) extracting characteristic information related to the four attributes of lithology, geological structure, water pressure and tunnel burial depth and the tunnel location coordinates corresponding to the characteristic information from the geological information obtained in step 1), and encoding the extracted characteristic information into characteristic values; 3) For the characteristic values ​​obtained in step 2), data fusion is performed on the characteristic values ​​with the same attributes and the same tunnel position coordinates, including: Establish the recognition framework Θ={θ1,θ2,…,θ k }, θ j represents the specific identification result, that is, it is identified as belonging to a certain category among lithology, geological structure, water pressure and tunnel depth; the characteristic value obtained in step 2) is used as evidence, and each piece of evidence E1, E2, ..., E i The corresponding probability distribution functions are m1, m2, ..., m i , that is, m i (θ j )=P(E i ∣θ j ), each m i The probability distribution defined on the recognition frame Θ represents the evidence E i The support degree for the recognition result; A is the corresponding focal element, and The fused probability distribution function m is calculated by the following formula: Among them, K is a normalization constant used to ensure that the sum of the fused probability distribution function is 1: ① Calculate the average evidence: In the formula, k is the number of focal elements; n is the number of evidences; ② Calculate the distance between a single piece of evidence and the average evidence: The larger the distance, the smaller the correlation between the two pieces of evidence. In order to make high-trust evidence obtain a high weight coefficient, the constructor makes the weight coefficient of the evidence equal to the distance d. i Inversely proportional, the mass function of different focal elements is: ③ In order to objectively and truly reflect the importance of each source of evidence, the entropy method is used to determine the weight of each evidence. The weight coefficient expression is: In the formula, ③Use weight coefficient ω i The weight of the initial evidence source is modified, that is, The new evidence m new As the evidence source, substitute it into formula (1) to perform pairwise fusion, and output a fused data each time; The four attributes corresponding to the same tunnel location coordinates are fused to form a sequence {a1, a2, a3, a4}, where a1 represents lithology fusion information, a2 represents geological structure fusion information, a3 represents water pressure fusion information, and a4 represents tunnel depth fusion information, thereby obtaining a data set consisting of several sequences; 4) Using the PSO-RF algorithm model as the water inflow prediction model, the prediction model is trained and tested: The data set obtained in step 3) is divided into a training set and a test set, the PSO-RF algorithm model is trained with the training set, and the trained model is tested with the test set. A sequence is used as an input of the PSO-RF algorithm model, and the goodness of fit and mean square error are selected as evaluation indicators of model training to obtain a prediction model that has passed the training and testing; 5) Use the qualified prediction model obtained in step 4) to predict the tunnel water inflow.

Images