A tunnel environment safety risk assessment method and system based on deep learning

Abstract

The application discloses a kind of based on deep learning's tunnel environment safety risk assessment method and system, comprising: obtaining the multi-source data of tunnel, if satisfying preset condition then obtaining environmental data;Image data is extracted to feature, and damage feature vector is obtained;Environmental data and damage feature vector are input into prediction model, to output damage propagation risk index;Based on stress distribution data, the structural stability coefficient of tunnel is calculated, and damage propagation risk index, structural stability coefficient and environmental data are input into risk assessment model, to generate the first influence weight of damage propagation risk index and the second influence weight of structural stability coefficient according to environmental data, and based on the first influence weight and the second influence weight respectively on damage propagation risk index and structural stability coefficient Fusion, obtain the environmental safety risk assessment result of tunnel, therefore, by implementing the application, it can improve the accurate evaluation of tunnel collapse risk.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of risk assessment, in particular to a tunnel environment safety risk assessment method and system based on deep learning. BACKGROUND

[0002] As a key node of the transportation network and an important carrier of underground space, the structural safety of the tunnel is directly related to the safety of people's lives and property and the stability of economic and social operation. Collapse is the most serious failure form of tunnel structure, which has the characteristics of strong burst, great destructive and difficult rescue. Once it happens, it may not only cause significant casualties and facility damage, but also cause long-term interruption of the transportation aorta, resulting in huge direct and indirect economic losses, and even adverse social impact. Therefore, it is necessary to conduct real-time, quantitative and forward-looking assessment on the health status of the tunnel structure, especially when stress anomaly or crack appears. If the risk of evolution into overall instability and collapse cannot be scientifically assessed, it may delay the warning due to misjudgment, or cause unnecessary operation and maintenance cost due to overreaction.

[0003] The existing tunnel safety risk assessment method often faces a bottleneck in accuracy when dealing with complex disaster-causing mechanisms. One method focuses on single-dimensional analysis, for example, relying only on structural mechanics model and stress data of limited measuring points to calculate safety factor, which is difficult to effectively capture and quantify the dynamic expansion threat of apparent crack damage. Another computer vision method based on image recognition mainly focuses on the static geometric features of cracks, lacks correlation analysis with internal stress state and inducing environmental factors, and ignores the "chain reaction" mechanism that local damage may be dramatically amplified and trigger overall instability under certain adverse environment, resulting in deviation between risk assessment results and real risk situation, and difficulty in accurately distinguishing high-risk and low-risk states, thereby affecting the timeliness and scientificity of early warning decision. SUMMARY

[0004] The present application provides a tunnel environment safety risk assessment method and system based on deep learning, which can improve the accurate assessment of tunnel collapse risk.

[0005] An embodiment of the present application provides a tunnel environment safety risk assessment method based on deep learning, comprising:

[0006] Obtaining multi-source data of the tunnel, if the multi-source data meet the preset condition, obtaining environmental data where the tunnel is located, wherein the multi-source data includes stress distribution data and image data of cracks;

[0007] Performing feature extraction on the image data to obtain a damage feature vector reflecting the change of crack morphology;

[0008] inputting the environmental data and the damage feature vector into a prediction model to output a damage propagation risk index representing deterioration of the damage in a future period of time;

[0009] a structure stability coefficient of the tunnel is calculated based on the stress distribution data, and the damage propagation risk index, the structure stability coefficient and the environmental data are input into a risk assessment model to generate a first influence weight of the damage propagation risk index and a second influence weight of the structure stability coefficient according to the environmental data through an attention weight mechanism in the risk assessment model, and the damage propagation risk index and the structure stability coefficient are fused based on the first influence weight and the second influence weight respectively to obtain an environmental safety risk assessment result of the tunnel.

[0010] The stress distribution data and crack image data are acquired in the embodiments of the application, risk signals can be preliminarily captured from two core dimensions of macro mechanical state and micro defect appearance, detailed environmental data is acquired by triggering a preset condition, so that the system can focus on external environmental factors inducing risk deterioration when discovering potential risks, redundancy and inefficiency caused by continuous collection of all environmental data are avoided, and timeliness and data pertinence of the assessment are ensured; feature extraction is performed on the crack image, aiming to capture the essence of the morphological change; the damage feature vector is input into the prediction model, which can quantify the possibility of development, expansion and accelerated deterioration of the existing damage in a future period of time under specific environmental load, directly revealing the time dimension and evolution path of the collapse risk, which is a key to improve the foresight and accuracy of the early warning; the structure stability coefficient is calculated based on the stress distribution data, which is an evaluation of the distance between the current overall bearing capacity of the tunnel and the failure threshold, representing the possibility of instability of the structure under instantaneous load even without considering damage evolution, which is a basic statics dimension of the collapse risk evaluation, and is complementary to the dynamic damage propagation prediction; the damage propagation risk index, the structure stability coefficient and the environmental data are input into the risk assessment model at the same time, which can identify which risk factor is dominant under different environments; the fusion based on the weights enables the final risk assessment result to be adaptive to the environmental context, dynamically highlighting the most urgent and dominant risk source, and then outputting a precise result with high context relevance, which is closer to the actual engineering judgment, thereby greatly improving the accuracy and reliability of the assessment of the complex collapse risk situation.

[0011] Further, the feature extraction on the image data to obtain the damage feature vector reflecting the morphological change of the crack comprises:

[0012] identifying a crack contour of the image data and calculating a geometric feature parameter of the crack contour;

[0013] based on the geometric feature parameter and a preset reference feature parameter, wherein the reference feature parameter is calculated according to reference image data;

[0014] calculate a morphological change amount of the crack profile relative to the reference feature parameter, and determine a damage feature vector reflecting the morphological change of the crack based on the morphological change amount.

[0015] In this way, feature extraction is performed on the crack image, aiming to capture the essence of the "morphological change".

[0016] Further, the environmental data and the damage feature vector are input into a prediction model to output a damage propagation risk index representing the deterioration of the damage in a future period, comprising:

[0017] The environmental data and the damage feature vector are input into a prediction model to perform time series encoding on the environmental data through a first channel to obtain environmental pattern features, wherein the prediction model comprises the first channel, a second channel and a regression output layer;

[0018] The environmental pattern features are fused with the damage feature vector received by the second channel to obtain comprehensive state features;

[0019] The comprehensive state features are input into the regression output layer to calculate a damage propagation risk index for quantitatively representing the future set period.

[0020] In this way, by inputting the damage feature vector into the prediction model, the possibility of the existing damage developing, expanding and leading to accelerated deterioration of the structure performance in a future period under a specific environmental load can be quantified, directly revealing the time dimension and evolution path of the collapse risk, which is a key to improving the early warning foresight and accuracy.

[0021] Further, the structural stability coefficient of the tunnel is calculated based on the stress distribution data, comprising:

[0022] According to the stress distribution data, a plurality of stress concentration regions and corresponding maximum stress values in the tunnel structure are determined;

[0023] Based on each of the maximum stress values and a limit strength parameter of the tunnel, a local stability safety factor of each of the stress concentration regions is calculated;

[0024] Each of the local stability safety factors is normalized to obtain a plurality of processing results, and each of the processing results is weighted and fused to generate an overall structural stability coefficient of the tunnel.

[0025] Thus, calculating the structural stability coefficient based on the stress distribution data is an evaluation of the distance between the current overall bearing capacity of the tunnel and the failure threshold, which represents the possibility of instability of the structure under instantaneous load even without considering damage evolution, which is a basic static dimension for evaluating collapse risk, and is complementary to dynamic damage propagation prediction.

[0026] Further, the inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model to generate a first influence weight of the damage propagation risk index and a second influence weight of the structural stability coefficient according to the environmental data through an attention weight mechanism in the risk assessment model comprises:

[0027] Inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model, wherein the risk assessment model comprises an environmental encoder, a first fully connected network, and a second fully connected network;

[0028] Encoding the environmental data through the environmental encoder to obtain an environmental feature vector;

[0029] Concatenating the environmental feature vector with the damage propagation risk index and the structural stability coefficient respectively to form a first concatenated vector and a second concatenated vector;

[0030] Inputting the first concatenated vector and the second concatenated vector into the first fully connected network and the second fully connected network respectively to perform calculation and normalization processing to obtain the first influence weight of the damage propagation risk index and the second influence weight of the structural stability coefficient.

[0031] Thus, by simultaneously inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into the risk assessment model, it can be identified which risk factor is dominant under different environments;

[0032] Further, the fusion of the damage propagation risk index and the structural stability coefficient based on the first influence weight and the second influence weight respectively to obtain the environmental safety risk assessment result of the tunnel comprises:

[0033] Multiplying the first influence weight and the damage propagation risk index to obtain a weighted risk value;

[0034] Multiplying the second influence weight and the structural stability coefficient to obtain a weighted stability value;

[0035] Summing the weighted risk value and the weighted stability value to obtain the environmental safety risk assessment result of the tunnel.

[0036] In this way, the weight-based fusion enables the final risk assessment result to be adaptive to the environmental context, dynamically highlight the current most urgent and dominant risk source, and output a precise result with high context relevance, which is closer to the actual engineering judgment, thereby greatly improving the accuracy and reliability of the complex collapse risk situation assessment.

[0037] Further, if the multi-source data meets a preset condition, environmental data of the tunnel is acquired, including:

[0038] It is judged whether the stress distribution data and the image data meet a preset condition;

[0039] If the uniformity of the stress distribution data is lower than a preset threshold or the image data has a crack, it is determined that the multi-source data meets the preset condition, and environmental data of the tunnel is acquired.

[0040] In this way, the detailed environmental data is acquired by triggering the preset condition, so that the system can focus on the external environmental factors that induce risk deterioration when potential risks are found, avoiding redundancy and inefficiency caused by continuous acquisition of all environmental data, and ensuring the timeliness and data pertinence of the assessment.

[0041] Further, after obtaining the environmental safety risk assessment result of the tunnel, it further includes:

[0042] The environmental safety risk assessment result is compared with a preset risk threshold interval to determine a risk level of tunnel collapse;

[0043] If the risk level reaches a warning level, comprehensive warning information including the risk level, main risk source and recommended disposal measures is generated.

[0044] In this way, the environmental safety risk assessment result is calibrated and quantified by the preset threshold interval, and structured warning information is generated, which ensures that the assessment result is not only accurate but also usable, directly improving the accuracy and timeliness of risk decision-making.

[0045] Another embodiment of the application also provides a tunnel environmental safety risk assessment system based on deep learning, including:

[0046] An acquisition module is configured to acquire multi-source data of a tunnel, and if the multi-source data meets a preset condition, environmental data of the tunnel is acquired, wherein the multi-source data includes stress distribution data and image data of cracks;

[0047] An extraction module is configured to perform feature extraction on the image data to obtain a damage feature vector reflecting crack morphology change;

[0048] a prediction module configured to input the environmental data and the damage feature vector into a prediction model to output a damage propagation risk index representing a risk of the damage worsening in a future period;

[0049] an evaluation module configured to calculate a structure stability coefficient of the tunnel based on the stress distribution data, and input the damage propagation risk index, the structure stability coefficient and the environmental data into a risk evaluation model, so as to generate a first influence weight of the damage propagation risk index and a second influence weight of the structure stability coefficient according to the environmental data by an attention weight mechanism in the risk evaluation model, and fuse the damage propagation risk index and the structure stability coefficient based on the first influence weight and the second influence weight respectively to obtain an environmental safety risk evaluation result of the tunnel.

[0050] Further, the evaluation module comprises:

[0051] an input unit configured to input the damage propagation risk index, the structure stability coefficient and the environmental data into a risk evaluation model, wherein the risk evaluation model comprises an environmental encoder, a first fully connected network and a second fully connected network;

[0052] an encoding unit configured to encode the environmental data by the environmental encoder to obtain an environmental feature vector;

[0053] a splicing unit configured to splice the environmental feature vector with the damage propagation risk index and the structure stability coefficient respectively to form a first spliced vector and a second spliced vector;

[0054] a processing unit configured to input the first spliced vector and the second spliced vector into the first fully connected network and the second fully connected network respectively to perform calculation and normalization processing to obtain a first influence weight of the damage propagation risk index and a second influence weight of the structure stability coefficient. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative effort.

[0056] Figure 1 is a flowchart of an embodiment of the tunnel environmental safety risk evaluation method based on deep learning provided by the present application;

[0057] Figure 2is a flowchart of one embodiment of steps S201 to S203 provided by the present application;

[0058] Figure 3 is a flowchart of one embodiment of steps S301 to S304 provided by the present application;

[0059] Figure 4 is a structural diagram of one embodiment of the tunnel environment safety risk assessment system based on deep learning provided by the present application. DETAILED DESCRIPTION

[0060] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms “include” and “have” and any variations thereof in the specification and claims of the present application and the above description of drawings are intended to cover non-exclusive inclusion.

[0062] In the description of the embodiments of the present application, the technical terms “first”, “second”, etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of “a plurality of” is two or more, unless otherwise explicitly and specifically limited.

[0063] Reference herein to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0064] In the description of the embodiments of the present application, the term “and / or” is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character “ / ” herein generally represents an “or” relationship between the associated objects.

[0065] In the description of the embodiments of the present application, the term "a plurality of" refers to two or more (including two), and similarly, "a plurality of groups" refers to two or more groups (including two groups), and "a plurality of pieces" refers to two or more pieces (including two pieces).

[0066] In the description of the embodiments of the present application, unless otherwise explicitly specified and limited, the technical terms "mounting", "connecting", "connecting", "fixing" and the like should be understood in a broad sense, for example, can be fixedly connected, or can be detachably connected, or can be integrated; can be mechanically connected, or can be electrically connected; can be directly connected, or can be indirectly connected through an intermediate medium; can be the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0067] As a key node of the traffic network and an important carrier of underground space, the structure safety of the tunnel is directly related to the safety of people's life and property and the stability of economic and social operation. Collapse is the most serious failure form of tunnel structure, and once it occurs, it will cause serious consequences. The existing tunnel safety risk assessment method focuses on single dimension analysis or computer vision method based on image recognition, but both methods cannot accurately assess the collapse risk.

[0068] Referring to Figure 1 To improve the accurate assessment of the tunnel collapse risk, an embodiment of the present application provides a tunnel environment safety risk assessment method based on deep learning, comprising steps S101 to S104.

[0069] Step S101, acquiring multi-source data of the tunnel, if the multi-source data meets a preset condition, acquiring environment data where the tunnel is located, wherein the multi-source data comprises stress distribution data and image data of cracks;

[0070] In some embodiments, the multi-source data of the tunnel is acquired, specifically: fiber grating sensor arrays are pre-buried or installed at key structural sections (such as vaults, side walls, inverted arches) of the tunnel to continuously monitor the strain and stress inside the concrete, forming the stress distribution data. At the same time, high-definition industrial cameras or visual sensors carried by inspection robots are deployed at the corresponding positions to collect high-definition images of the tunnel surface according to a preset period or a trigger condition (such as stress anomaly), to constitute the original crack image data.

[0071] It should be noted that after obtaining the initial multi-source data, the initial multi-source data needs to be preprocessed (such as cleaning, filtering and denoising, coordinate conversion) to form stress distribution data reflecting the stress size and spatial distribution of the tunnel structure inside, to facilitate subsequent judgment.

[0072] In some embodiments, if the multi-source data meets a preset condition, environment data of the tunnel is acquired, including: determining whether the stress distribution data and the image data meet a preset condition; if uniformity of the stress distribution data is lower than a preset threshold or cracks appear in the image data, it is determined that the multi-source data meets the preset condition, and environment data of the tunnel is triggered to be acquired. Specifically, first, since uneven stress distribution or over-limit is the fundamental mechanical cause of cracks in the lining structure, and cracks are the most intuitive manifestation of the above-mentioned mechanical deterioration, after the stress distribution data and the image data are acquired, the stress distribution data needs to be analyzed, and the uniformity index thereof is calculated (for example, the standard deviation, the coefficient of variation of the stress values of all measuring points, or the stress field gradient is evaluated), and then the uniformity index is compared with a preset empirical threshold to determine whether the stress distribution is abnormal. At the same time, a crack recognition algorithm (such as an image segmentation model based on deep learning) is called to process the image data to detect whether the image contains a crack profile. If cracks are identified in the image, then if the uniformity of the stress distribution data is lower than the preset threshold, or cracks appear in the image data, as long as any of the above conditions is met, the system determines that the current multi-source data has met the preset condition for starting the deep risk assessment. Once it is determined that the preset condition is met, it means that there are cracks or other reasons that may cause collapse risk, i.e. risk signs, so the environment data of the tunnel needs to be immediately triggered and acquired by the sensor, wherein the environment data includes but is not limited to temperature, humidity, vibration, underground water pressure, etc.

[0073] In this way, the preset condition is used to trigger the acquisition of detailed environment data, so that the system can focus on the external environmental factors that induce risk deterioration when potential risks are found, avoiding redundancy and inefficiency caused by continuous acquisition of all environment data, and ensuring the timeliness and data pertinence of the assessment.

[0074] It should be noted that tunnel collapse is the result of the combined action of overall structural instability (reflected by stress data) and local damage expansion (reflected by crack image) under adverse environment (reflected by environment data). If abnormal tunnel stress data or cracks are found during the inspection, it does not necessarily lead to collapse, so the environment data needs to be combined to evaluate the collapse risk.

[0075] In step S102, feature extraction is performed on the image data to obtain a damage feature vector reflecting crack morphology changes;

[0076] In some embodiments, the feature extraction on the image data to obtain a damage feature vector reflecting the change in crack morphology includes: identifying a crack contour of the image data, and calculating a geometric feature parameter of the crack contour; calculating a change in morphology of the crack contour relative to a preset reference feature parameter, wherein the reference feature parameter is calculated according to reference image data; and determining a damage feature vector reflecting the change in crack morphology based on the change in morphology. Specifically, first, when it is determined that the tunnel has cracks, the contour of the cracks is accurately identified from the acquired tunnel surface image, and the geometric feature parameter of the contour is calculated, wherein the geometric feature parameter generally includes but is not limited to: crack length, maximum width, average width, strike (angle with the principal stress direction), number of branches, tortuosity (ratio of actual length to straight-line distance between the beginning and end), and area. These parameters digitally describe the current spatial morphology of the cracks from different dimensions, and the specific calculation method can be realized by the prior art, which is not limited in the present application. Then, the change in morphology of the geometric feature parameter of the current crack contour relative to the reference feature parameter is calculated, for example, the growth rate of the length, the expansion of the width, the increase in the tortuosity, or the number of newly added branches, etc., wherein the change can be an absolute difference or a relative proportion. Finally, the plurality of calculated changes in morphology are taken as an ordered set to jointly constitute a damage feature vector, so as to dynamically reflect the evolution process and severity of the cracks from the reference state to the current state.

[0077] It should be noted that the reference feature parameter is calculated according to reference image data, wherein the reference image data is typically selected from one of the following two cases: one, baseline image collected at the initial stage of the tunnel construction or after the last repair when the tunnel is in a stable and healthy state; and two, in continuous monitoring, the image at a selected historical time (such as when there is no obvious sign of deterioration) is selected as a dynamic reference. By performing the same contour identification and parameter calculation process on the reference image, a corresponding set of reference feature parameters can be obtained.

[0078] In this way, the feature extraction on the crack image aims to capture the essence of the "change in morphology".

[0079] Step S103: inputting the environmental data and the damage feature vector into a prediction model to output a damage propagation risk index representing deterioration of damage in a future period;

[0080] Please refer to Figure 2 In some embodiments, the inputting of the environmental data and the damage feature vector into the prediction model to output a damage propagation risk index representing deterioration of damage in a future period includes steps S201 to S203:

[0081] Step S201, inputting the environmental data and the damage feature vector into a prediction model to perform time series encoding on the environmental data through a first channel to obtain an environmental pattern feature, wherein the prediction model comprises the first channel, a second channel, and a regression output layer;

[0082] In some embodiments, the environmental data and the damage feature vector are input into two independent channels of the prediction model in parallel to perform feature encoding to obtain the environmental pattern feature, wherein the first channel is dedicated to processing environmental data with time series characteristics (such as periodic temperature, humidity, vibration, or water pressure data).

[0083] It should be noted that the first channel is usually composed of a time series encoding network, such as a recurrent neural network (RNN) or a long short-term memory network (LSTM), which reads the environmental data sequence in time step order, captures the dynamic pattern, trend, and periodicity of the data in the time dimension through its internal memory unit, and finally outputs a fixed-dimensional environmental pattern feature vector through the processing of the last time step to reflect the historical influence and potential driving pattern of the environmental conditions on the structural damage evolution.

[0084] Step S202, performing nonlinear transformation on the damage feature vector through the second channel to obtain a transformation result, and performing feature fusion on the environmental pattern feature and the transformation result to obtain a comprehensive state feature;

[0085] In some embodiments, the second channel comprises a plurality of cascaded fully connected layers, which perform nonlinear transformation on the damage feature vector after receiving the damage feature vector reflecting the current crack static morphology obtained by image analysis to obtain a transformation result suitable for fusion with the environmental feature; then, the environmental pattern feature output by the first channel and the damage feature vector output by the second channel are spliced through the feature fusion layer of the prediction model to form a more comprehensive comprehensive state feature, which contains both the current state (geometric morphology) of the damage and the dynamic environmental background in which the damage is located.

[0086] Step S203, inputting the comprehensive state feature into the regression output layer to calculate a damage propagation risk index for quantitatively representing the damage propagation risk in a future set period.

[0087] In some embodiments, the fused comprehensive state feature is input into the regression output layer of the prediction model to perform a series of weighted calculations and dimension reduction on the comprehensive state feature, and finally output a continuous, quantitative scalar value, i.e., a damage propagation risk index, wherein the damage propagation risk index specifically represents an estimated value of the probability or severity of significant deterioration of the damage within a future preset period (such as the next 24 hours or the next week) under the combined action of the current damage state and the existing environmental pattern.

[0088] Note that the regression output layer is usually composed of one or more fully connected layers, with the last layer using a linear activation function. The role of this layer is to perform a non-linear mapping from the high-dimensional feature space to a one-dimensional risk score.

[0089] Note that the training process of the prediction model includes, first, the training sample contains: 1) environmental data of a time series segment; 2) crack damage feature vector observed and extracted at the end of the time period; 3) as a supervised label, the actual observed damage deterioration amount (such as crack length growth rate, area expansion amount) in the subsequent one fixed future period (such as the next week). The training goal is to enable the model to learn to predict the future damage development degree based on the current environmental sequence and the current damage state. In terms of structure, the prediction model is an end-to-end neural network, and the core components and information flow are as follows: the first channel (time series encoding channel): usually composed of a long short-term memory network (LSTM) layer or a time series convolutional network (TCN). The input of this channel is a fixed time window (such as the past 48 hours) of environmental data sequence. The LSTM layer learns the long-term dependencies in the sequence through its gating mechanism, and the hidden state at the last time step is extracted as a fixed-dimensional environmental pattern feature vector, which encodes the dynamic trend and potential periodic pattern of environmental changes. The second channel (damage state channel): usually composed of two to three layers of fully connected layers (FC), each followed by a nonlinear activation function such as ReLU. This channel receives the damage feature vector representing the current geometry of the crack (such as length, width, orientation, fractal dimension), and maps it to a hidden space matching the output dimension of the first channel through nonlinear transformation, forming a damage state feature for subsequent fusion. Feature fusion layer: concatenates the environmental pattern features output by the first channel and the damage state features output by the second channel to form a higher-dimensional comprehensive state feature. This operation ensures that the "static present situation" of the damage and the "dynamic background" information of the environment are preserved and jointly represented. Regression output layer: composed of one or more fully connected layers, with the last layer using a linear activation function. This layer receives the comprehensive state feature, and through a series of nonlinear transformations and final linear mapping, outputs a continuous scalar value, which is the damage propagation risk index. This index directly corresponds to the prediction of the future damage deterioration amount defined in the training label. During training, the mean square error is used as the loss function, and the Adam optimizer is used for backpropagation to minimize the error between the model's predicted risk index and the actual observed future damage deterioration amount. At the same time, techniques such as Dropout are usually used during training to prevent overfitting, and the model's generalization performance is monitored through a validation set. In this way, the trained prediction model can be obtained.

[0090] Thus, by inputting the damage feature vector into the prediction model, the possibility of the existing damage developing, expanding and leading to accelerated deterioration of the structural performance in the future period under a specific environmental load can be quantified, directly revealing the time dimension and evolution path of the collapse risk, which is a key to improving the foresight and accuracy of early warning.

[0091] In step S104, the structural stability coefficient of the tunnel is calculated based on the stress distribution data, and the damage propagation risk index, the structural stability coefficient and the environmental data are input into a risk assessment model to generate a first influence weight of the damage propagation risk index and a second influence weight of the structural stability coefficient according to the environmental data through an attention weight mechanism in the risk assessment model, and the damage propagation risk index and the structural stability coefficient are fused based on the first influence weight and the second influence weight respectively to obtain an environmental safety risk assessment result of the tunnel.

[0092] In some embodiments, the calculation of the structural stability coefficient of the tunnel based on the stress distribution data comprises: determining a plurality of stress concentration regions and corresponding maximum stress values in the tunnel structure according to the stress distribution data; calculating a local stability safety coefficient of each stress concentration region based on each maximum stress value and a limit strength parameter of the tunnel; and performing normalization processing on each local stability safety coefficient to obtain a plurality of processing results, and performing weighted fusion on each processing result to generate an overall structural stability coefficient of the tunnel. Specifically, first, a spatial gradient algorithm or a peak detection algorithm is used to locate points or regions in the stress distribution data whose stress values are significantly higher than those of the surrounding regions and exceed a preset stress threshold, and these regions are marked as stress concentration regions, and the maximum stress value in each region is recorded as the maximum stress value of the region , wherein represents the th stress concentration region. Then, based on the maximum stress value of each stress concentration region and the limit strength parameter of the tunnel structure, the local stability safety coefficient of each region is calculated, and the calculation formula is: wherein the limit strength parameter is determined according to the design strength of the tunnel lining material (such as concrete, steel) or the limit compressive / tensile strength measured in the laboratory. Then, since the safety coefficients of different regions may have different dimensions and initial ranges, in order to facilitate subsequent fusion, the local stability safety coefficients are normalized, wherein the Min-Max normalization method can be used to map them to the interval [0, 1], thereby obtaining a plurality of normalized processing results . Finally, the normalized processing results are fused with a preset fusion weight The weighted fusion is performed to generate the overall structural stability coefficient S of the tunnel, and the related calculation formula is: wherein n is the total number of stress concentration regions.

[0093] It should be noted that the fusion weight of each stress concentration region is dynamically determined according to the area size of the region, the importance of the structure position (such as the vault, side wall), and the stress concentration degree, and can be specifically obtained through a preset weight distribution function or a weight configuration table, which is not limited in the present application.

[0094] It should be noted that the local stability safety coefficient reflects the safety margin of a single stress concentration region under the current load, and the local stability safety coefficient The greater the value, the more stable the region. The overall structural stability coefficient S obtained by calculation is used to represent the instantaneous stability state of the tunnel structure as a whole, and the higher the value, the better the overall stability of the structure.

[0095] Thus, the structural stability coefficient calculated based on the stress distribution data is an evaluation of the distance between the current overall bearing capacity of the tunnel and the failure threshold, which represents the possibility of instability of the structure under the instantaneous load even without considering damage evolution, which is a basic statics dimension for evaluating the collapse risk, and is complementary to the dynamic damage propagation prediction.

[0096] Please refer to Figure 3 In some embodiments, the inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model to generate, by an attention weight mechanism in the risk assessment model, a first influence weight of the damage propagation risk index and a second influence weight of the structural stability coefficient according to the environmental data includes steps S301 to S304:

[0097] Step S301, inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model, wherein the risk assessment model includes an environmental encoder, a first fully connected network, and a second fully connected network;

[0098] In some embodiments, the damage propagation risk index, the structural stability coefficient S, and the environmental data (usually a time series or instantaneous vector containing temperature, humidity, vibration, groundwater level, etc. dimensions) calculated above are input into a risk assessment model. It should be noted that the risk assessment model is designed to include three core components: an environmental encoder, a first fully connected network, and a second fully connected network.

[0099] It should be noted that the training process of the risk assessment model is as follows: first, a historical data set is collected and constructed, which should contain multi-source data samples in the historical monitoring period of the tunnel project, each sample is composed of synchronous historical stress distribution data, historical crack image sequence, historical environmental monitoring data (temperature, humidity, vibration, etc.) and corresponding real risk level label marked by expert experience or subsequent accident record. After data preprocessing, the corresponding historical damage propagation risk index and historical structure stability coefficient are calculated based on the historical stress distribution data and historical image data offline, and the two are used as training features together with the historical environmental data. Then, a risk assessment model (including an environmental encoder, two fully connected networks, etc.) is initialized. During training, the historical risk index, historical stability coefficient and historical environmental data calculated as described above are used as model input, and the risk level marked by experts is used as supervision target. Through the back propagation algorithm, a composite loss function combining regression loss (such as mean square error MSE) and classification loss (such as cross entropy) is optimized, which aims to make the final output of the model as close as possible to the real risk label. In this process, the parameters of the environmental encoder and the two fully connected networks in the model are optimized synchronously, so that the attention weight mechanism can automatically learn how to dynamically and reasonably allocate the weights of damage propagation risk and structure stability in the final judgment under different environmental feature patterns (for example, humid environment vs. strong vibration environment). Among them, the training adopts an iterative method, uses a validation set to monitor the performance to prevent overfitting, until the model converges, thereby obtaining a trained risk assessment model that can automatically evaluate the environmental safety risk of the tunnel according to real-time data.

[0100] Step S302, encoding the environmental data through the environmental encoder to obtain an environmental feature vector;

[0101] In some embodiments, to realize the dynamic modulation of environmental conditions on risk factors, high-dimensional feature extraction of environmental data is performed through the environmental encoder to obtain an environmental feature vector. Among them, the environmental encoder is usually a shallow neural network (such as a multi-layer perception MLP) or a time series encoding network (such as LSTM, GRU, suitable for time series environmental data), which receives the original environmental data vector as input. The encoder maps the original environmental data to a more representative and compact environmental feature vector through a series of nonlinear transformations (such as fully connected layers with ReLU activation function).

[0102] Step S303, splicing the environmental feature vector with the damage propagation risk index and the structure stability coefficient respectively to form a first splicing vector and a second splicing vector;

[0103] In some embodiments, after obtaining the environmental feature vector, in order to establish the association between the environmental state and each risk factor, first, the environmental feature vector is spliced with the damage propagation risk index to form a first spliced vector. Then, the same environmental feature vector is spliced with the structural stability coefficient to form a second spliced vector.

[0104] It should be noted that if the damage propagation risk index and the structural stability coefficient are scalars, they are directly connected with the environmental feature vector at the beginning and the end; if the damage propagation risk index and the structural stability coefficient are both vectors, the corresponding dimensions are connected.

[0105] In step S304, the first spliced vector and the second spliced vector are respectively input into the first fully connected network and the second fully connected network for calculation and normalization processing to obtain the first influence weight of the damage propagation risk index and the second influence weight of the structural stability coefficient.

[0106] In some embodiments, after obtaining the first spliced vector and the second spliced vector, the two vectors are respectively sent into independent first fully connected network and second fully connected network for processing, wherein each fully connected network is usually composed of one to more hidden layers, which receives the spliced vector, and then outputs a scalar value representing the initial importance of the risk factor through weight matrix calculation and activation function transformation. For example, the first fully connected network processes the first spliced vector to output the initial weight of the damage propagation risk index; the second fully connected network processes the second spliced vector to output the initial weight of the structural stability coefficient. In order to ensure that the sum of the two weights is 1 and form a reasonable distribution ratio, the two initial weights are sent into a Softmax normalization layer (which can be integrated at the end of the fully connected network) together to calculate the two initial weights and output the final normalized first influence weight (a) and second influence weight (b), and satisfy a + b = 1.

[0107] In this way, by simultaneously inputting the damage propagation risk index, the structural stability coefficient and the environmental data into the risk assessment model, it can be identified which risk factor is dominant under different environments.

[0108] In some embodiments, the fusing of the damage propagation risk index and the structure stability coefficient based on the first influence weight and the second influence weight respectively to obtain the environmental safety risk assessment result of the tunnel comprises: multiplying the first influence weight by the damage propagation risk index to obtain a weighted risk value; multiplying the second influence weight by the structure stability coefficient to obtain a weighted stability value; and summing the weighted risk value and the weighted stability value to obtain the environmental safety risk assessment result of the tunnel. Specifically, first, after obtaining the first influence weight and the second influence weight, the damage propagation risk index and the corresponding first influence weight are subjected to scalar multiplication operation to obtain a weighted risk value, so as to dynamically adjust the importance of the future deterioration risk of the crack according to the current environmental condition (determined by the attention mechanism), for example, in an environment with high humidity and easy to accelerate damage, the weight is adjusted to be high, so that the risk index occupies a larger proportion in the final evaluation. Then, the structure stability coefficient and the corresponding second influence weight are subjected to scalar multiplication operation to obtain a weighted stability value, so as to adjust the sensitivity of the structure stability to the environmental condition, for example, in the environment of seismic motion or large temperature difference, the influence of the structure stability on the overall safety may be more prominent, and the weight is increased accordingly. Finally, the weighted risk value and the weighted stability value obtained by calculation are subjected to algebraic summation to obtain the final environmental safety risk assessment result of the tunnel.

[0109] It should be noted that the environmental safety risk assessment result is not a simple average of the two factors, but a dynamic trade-off and integration of "future potential threat" and "current static defense capability" under the perception of environmental context. The value can be directly used for subsequent risk level division and early warning decision. The higher the value, the greater the comprehensive safety risk.

[0110] Such weight-based fusion enables the final risk assessment result to be self-adaptive to the environmental context, dynamically highlighting the current most urgent and dominant risk source, and further outputting a precise result with high context relevance that is closer to the actual engineering judgment, thereby greatly improving the accuracy and reliability of the complex collapse risk situation assessment.

[0111] In some embodiments, after obtaining the environmental safety risk assessment result of the tunnel, the method further comprises: comparing the environmental safety risk assessment result with a preset risk threshold interval to determine a risk level of tunnel collapse; and if the risk level reaches a warning level, generating comprehensive warning information including the risk level, a main risk source, and a recommended treatment measure. Specifically, first, a plurality of "risk threshold intervals" are pre-set based on historical accident data, engineering specifications, and expert experience, wherein each interval corresponds to a specific "risk level of tunnel collapse" (e.g., low risk, medium risk, high risk, and emergency risk). Then, the calculated environmental safety risk assessment result is automatically compared with these preset intervals to determine its specific risk level. When it is determined that the risk level reaches or exceeds a preset warning level (e.g., high risk or emergency risk), the system triggers a warning generation module to automatically generate a comprehensive warning information based on the intermediate data and weight analysis in the risk assessment process.

[0112] It should be noted that the comprehensive warning information includes a risk level indicating the severity of the current risk, a main risk source, and a recommended treatment measure, wherein the main risk source is determined by backtracking the first influence weight and the second influence weight to identify the factor that contributes most to the current high risk. For example, if the first influence weight is significantly higher than the second influence weight, the warning information can explicitly indicate that damage dynamic deterioration is the main risk source; otherwise, it may indicate that structural stress exceeding the limit is the main cause. The recommended treatment measure is obtained by matching the risk level and the identified main risk source with a preset treatment measure knowledge base, wherein the treatment measure knowledge base includes different risk levels and different risk source combinations. For example, for the case of high risk and damage dynamic deterioration as the main source, it may be recommended to immediately conduct a close-range detailed detection of the section and consider temporary reinforcement and increased monitoring frequency.

[0113] In this way, the environmental safety risk assessment result is calibrated and quantified by the preset threshold interval, and structured warning information is generated, ensuring that the assessment result is not only accurate but also useful, directly improving the accuracy and timeliness of risk decision-making.

[0114] The embodiment of the present application obtains stress distribution data and crack image data, can preliminarily capture risk signals from two core dimensions of macro mechanical state and micro defect appearance, triggers detailed environment data through a preset condition, so that the system can focus on external environmental factors inducing risk deterioration when discovering potential risk, avoids redundancy and inefficiency caused by continuous collection of all environment data, and ensures timeliness and data pertinence of evaluation; the essence of the'morphological change' of the crack image is captured through feature extraction; the possibility that the existing damage develops, expands and leads to accelerated degradation of the structure performance in the future period under a specific environmental load is quantified by inputting the damage feature vector into the prediction model, and the time dimension and evolution path of the collapse risk are directly revealed, which is the key to improving the foresight and accuracy of the early warning; the structure stability coefficient is calculated based on the stress distribution data, which is an evaluation of the distance between the current overall bearing capacity of the tunnel and the failure threshold, and represents the possibility of instability of the structure under instantaneous load even without considering damage evolution, which is the basic statics dimension of the evaluation of the collapse risk, and is complementary to the dynamic damage propagation prediction; the risk assessment model is inputted by the damage propagation risk index, the structure stability coefficient and the environment data, so that the dominant risk factor in different environments can be identified; the fusion based on the weight makes the final risk assessment result adapt to the environmental context, dynamically highlights the current most urgent and dominant risk source, and then outputs a precise result with high context relevance, which is closer to the actual engineering judgment, thereby greatly improving the accuracy and reliability of the assessment of the complex collapse risk situation.

[0115] As shown in the above method embodiment, corresponding device embodiments are provided; Figure 4

[0116] An embodiment of the present application provides a tunnel environment safety risk assessment system based on deep learning, which comprises:

[0117] An acquisition module 100 is used for acquiring multi-source data of a tunnel, and if the multi-source data meets a preset condition, environment data of the tunnel is acquired, wherein the multi-source data comprises stress distribution data and image data of cracks.

[0118] An extraction module 200 is used for performing feature extraction on the image data to obtain a damage feature vector reflecting morphological changes of the cracks.

[0119] A prediction module 300 is used for inputting the environment data and the damage feature vector into a prediction model to output a damage propagation risk index representing damage deterioration in a future period.

[0120] ​The assessment module 400 is used to calculate the structural stability coefficient of the tunnel based on the stress distribution data, and input the damage propagation risk index, the structural stability coefficient, and the environmental data into the risk assessment model. The model uses an attention weighting mechanism to generate a first influence weight for the damage propagation risk index and a second influence weight for the structural stability coefficient based on the environmental data. The model then fuses the damage propagation risk index and the structural stability coefficient based on the first influence weight and the second influence weight to obtain the environmental safety risk assessment result of the tunnel.

[0121] In some embodiments, the evaluation module 400 includes: an input unit for inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model, wherein the risk assessment model includes an environmental encoder, a first fully connected network, and a second fully connected network; and an encoding unit for encoding the environmental data through the environmental encoder to obtain an environmental feature vector.

[0122] The splicing unit is used to splice the environmental feature vector with the damage propagation risk index and the structural stability coefficient respectively to form a first splicing vector and a second splicing vector; the processing unit is used to input the first splicing vector and the second splicing vector into the first fully connected network and the second fully connected network respectively for calculation and normalization processing to obtain the first influence weight of the damage propagation risk index and the second influence weight of the structural stability coefficient.

[0123] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the tunnel environment safety risk assessment method based on deep learning provided by any of the above-described method embodiments of the present invention.

[0124] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0125] Based on the above embodiments of the tunnel environment safety risk assessment method based on deep learning, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the tunnel environment safety risk assessment method based on deep learning of any embodiment of the present invention.

[0126] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0127] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The terminal device can include, but is not limited to, a processor and a memory.

[0128] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and connects all parts of the terminal device through various interfaces and lines.

[0129] On the basis of the above-mentioned method embodiment, another embodiment of the present application provides a computer readable storage medium, including a stored computer program, wherein when the computer program runs, the device where the computer readable storage medium is located executes the deep learning-based tunnel environment safety risk assessment method described in any one of the above-mentioned method embodiments of the present application.

[0130] The modules / units integrated in the device / terminal equipment, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0131] The above is the preferred embodiment of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which are also considered within the scope of protection of the present application.

Claims

1. A deep learning-based method for assessing tunnel environmental safety risks, wherein the method is used for tunnel collapse risk assessment, characterized in that... include: Acquire multi-source data of the tunnel. If the multi-source data meets preset conditions, acquire environmental data of the tunnel. The multi-source data includes stress distribution data and crack image data. Feature extraction is performed on the image data to obtain a damage feature vector reflecting the changes in crack morphology; The environmental data and the damage feature vector are input into the prediction model to output a damage propagation risk index that characterizes the deterioration of the damage in the future. The structural stability coefficient of the tunnel is calculated based on the stress distribution data. The damage propagation risk index, the structural stability coefficient, and the environmental data are then input into a risk assessment model. The model uses an attention weighting mechanism to generate a first influence weight for the damage propagation risk index and a second influence weight for the structural stability coefficient based on the environmental data. The damage propagation risk index and the structural stability coefficient are then fused based on the first and second influence weights to obtain the environmental safety risk assessment result for the tunnel. The step of extracting features from the image data to obtain a damage feature vector reflecting crack morphological changes includes: Identify the crack contour in the image data and calculate the geometric feature parameters of the crack contour; Based on the geometric feature parameters and the preset reference feature parameters, wherein the reference feature parameters are calculated based on reference image data; Calculate the morphological change of the crack profile relative to the reference feature parameters, and determine a damage feature vector reflecting the crack morphological change based on the morphological change. The step of inputting the environmental data and the damage feature vector into the prediction model to output a damage propagation risk index characterizing the deterioration of damage in the future includes: The environmental data and the damage feature vector are input into the prediction model to perform time-series encoding on the environmental data through the first channel to obtain environmental pattern features. The prediction model includes the first channel, the second channel, and a regression output layer. The environmental pattern features are fused with the damage feature vector received by the second channel to obtain the comprehensive state features; The comprehensive state characteristics are input into the regression output layer to calculate an index that quantitatively represents the risk of damage propagation within a set future time period. If the multi-source data meets preset conditions, then the environmental data of the tunnel is obtained, including: Determine whether the stress distribution data and the image data meet preset conditions; If the uniformity of the stress distribution data is lower than a preset threshold or cracks appear in the image data, it is determined that the multi-source data meets the preset conditions, and the acquisition of environmental data of the tunnel is triggered.

2. The tunnel environment safety risk assessment method based on deep learning according to claim 1, characterized in that, The structural stability coefficient of the tunnel calculated based on the stress distribution data includes: Based on the stress distribution data, several stress concentration areas and their corresponding maximum stress values ​​in the tunnel structure are determined. Based on the maximum stress values ​​and the ultimate strength parameters of the tunnel, the local stability safety factor of each stress concentration region is calculated. The local stability safety factors are normalized to obtain several processing results, and the processing results are weighted and fused to generate the overall structural stability coefficient of the tunnel.

3. The tunnel environment safety risk assessment method based on deep learning according to claim 1, characterized in that, The step of inputting the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model, and generating a first influence weight for the damage propagation risk index and a second influence weight for the structural stability coefficient based on the environmental data through an attention weight mechanism in the risk assessment model, includes: The damage propagation risk index, the structural stability coefficient, and the environmental data are input into the risk assessment model, wherein the risk assessment model includes an environmental encoder, a first fully connected network, and a second fully connected network. The environmental data is encoded using the environmental encoder to obtain an environmental feature vector; The environmental feature vector is concatenated with the damage propagation risk index and the structural stability coefficient to form a first concatenated vector and a second concatenated vector. The first splicing vector and the second splicing vector are respectively input into the first fully connected network and the second fully connected network for calculation and normalization processing to obtain the first influence weight of the damage propagation risk index and the second influence weight of the structural stability coefficient.

4. The tunnel environment safety risk assessment method based on deep learning according to claim 1, characterized in that, The environmental safety risk assessment result of the tunnel is obtained by fusing the damage propagation risk index and the structural stability coefficient based on the first influence weight and the second influence weight, respectively, including: Multiply the first influence weight by the damage propagation risk index to obtain the weighted risk value; Multiply the second influence weight by the structural stability coefficient to obtain the weighted stability value; The environmental safety risk assessment result of the tunnel is obtained by summing the weighted risk value and the weighted stability value.

5. The tunnel environment safety risk assessment method based on deep learning according to any one of claims 1-4, characterized in that, After obtaining the environmental safety risk assessment results for the tunnel, the following is also included: The environmental safety risk assessment results are compared with a preset risk threshold range to determine the risk level of tunnel collapse; If the risk level reaches the warning level, a comprehensive warning message is generated that includes the risk level, the main sources of risk, and recommended measures.

6. A tunnel environmental safety risk assessment system based on deep learning, the assessment system being used for tunnel collapse risk assessment, characterized in that, include; The acquisition module is used to acquire multi-source data of the tunnel. If the multi-source data meets the preset conditions, the environmental data of the tunnel is acquired. The multi-source data includes stress distribution data and crack image data. The extraction module is used to extract features from the image data to obtain a damage feature vector that reflects the changes in crack morphology; The prediction module inputs the environmental data and the damage feature vector into the prediction model to output a damage propagation risk index that characterizes the deterioration of the damage in the future period. The assessment module is used to calculate the structural stability coefficient of the tunnel based on the stress distribution data, and input the damage propagation risk index, the structural stability coefficient, and the environmental data into the risk assessment model. The model uses an attention weighting mechanism to generate a first influence weight for the damage propagation risk index and a second influence weight for the structural stability coefficient based on the environmental data. The damage propagation risk index and the structural stability coefficient are then fused based on the first and second influence weights to obtain the environmental safety risk assessment result of the tunnel. The step of extracting features from the image data to obtain a damage feature vector reflecting crack morphological changes includes: Identify the crack contour in the image data and calculate the geometric feature parameters of the crack contour; Based on the geometric feature parameters and the preset reference feature parameters, wherein the reference feature parameters are calculated based on reference image data; Calculate the morphological change of the crack profile relative to the reference feature parameters, and determine a damage feature vector reflecting the crack morphological change based on the morphological change. The step of inputting the environmental data and the damage feature vector into the prediction model to output a damage propagation risk index characterizing the deterioration of damage in the future includes: The environmental data and the damage feature vector are input into the prediction model to perform time-series encoding on the environmental data through the first channel to obtain environmental pattern features. The prediction model includes the first channel, the second channel, and a regression output layer. The environmental pattern features are fused with the damage feature vector received by the second channel to obtain the comprehensive state features; The comprehensive state characteristics are input into the regression output layer to calculate an index that quantitatively represents the risk of damage propagation within a set future time period. If the multi-source data meets preset conditions, then the environmental data of the tunnel is obtained, including: Determine whether the stress distribution data and the image data meet preset conditions; If the uniformity of the stress distribution data is lower than a preset threshold or cracks appear in the image data, it is determined that the multi-source data meets the preset conditions, and the acquisition of environmental data of the tunnel is triggered.

7. The tunnel environment safety risk assessment system based on deep learning according to claim 6, characterized in that, The evaluation module includes: An input unit is used to input the damage propagation risk index, the structural stability coefficient, and the environmental data into a risk assessment model, wherein the risk assessment model includes an environmental encoder, a first fully connected network, and a second fully connected network. The encoding unit is used to encode the environmental data through the environmental encoder to obtain an environmental feature vector; The splicing unit is used to splice the environmental feature vector with the damage propagation risk index and the structural stability coefficient respectively to form a first splicing vector and a second splicing vector; The processing unit is used to input the first splicing vector and the second splicing vector into the first fully connected network and the second fully connected network, respectively, for calculation and normalization processing to obtain the first influence weight of the damage propagation risk index and the second influence weight of the structural stability coefficient.

Images