Intelligent assessment method for pipe jacking construction safety risk based on multi-source data fusion

Abstract

The application discloses a pipe jacking construction safety risk intelligent evaluation method based on multi-source data fusion, and relates to the technical field of pipe jacking construction.The steps of the method comprise the following steps: obtaining historical monitoring data of pipe jacking construction, including geological radar images, engineering operation parameters and structure deformation parameters; based on the historical monitoring data, establishing a construction topology network, identifying risk driving factors in combination with a preset GAM model, and automatically adjusting the monitoring data collection frequency; introducing a machine learning model, iteratively training each batch of monitoring data, evaluating the abnormal fluctuation of the structure deformation parameters, and identifying the risk degree of geological mutation; wherein, the risk degree comprises effective risk and ineffective risk; based on the construction topology network, in combination with the effective risk, executing an operation safety strategy, and determining the out-of-bound behavior; the application realizes accurate association of node links, improves monitoring pertinence, and to a certain extent, refines multi-source data fusion, facilitates judgment of the core source of risk, and timely early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pipe jacking construction, in particular to a pipe jacking construction safety risk intelligent evaluation method based on multi-source data fusion. BACKGROUND

[0002] With the development of technology, the proportion of infrastructure such as water supply and power supply used in underground space construction has increased significantly, and pipe jacking construction, as a non-excavation underground pipeline laying technology, has been widely used. However, due to the underground operation environment, there are characteristics such as complex and variable geological conditions, difficult to intuitively control construction parameters and structural state, hidden and dangerous risk transmission, and the traditional pipe jacking construction safety risk evaluation technology has gradually exposed many deficiencies.

[0003] On the one hand, although some monitoring data will be collected during the construction process, these data are usually in isolated and discrete state, lacking effective systematic fusion, or simply staying at the data superposition level without deeply mining the influence degree or correlation between different parameters, for example: knowing that the settlement of a batch is 0.6mm, the structural stress is 28MPa, and the joint deformation is 1.2mm, but unable to determine whether the settlement leads to stress increase or the stress anomaly causes settlement. Such uncorrelated analysis of multi-source fusion is difficult to determine the core source of risk, and ultimately leads to rapid overrun of subsequent settlement due to the failure to deal with the core risk.

[0004] On the other hand, the data obtained by sensors are often limited, and personnel are usually arranged at the underground pipe jacking construction site. When a danger occurs, if the sensor fails or has low sensitivity or insufficient accuracy, false alarms are likely to occur, making the detection efficiency of out-of-bound behavior threatening the safety of construction personnel low. SUMMARY

[0005] (I) Technical problems solved

[0006] In view of the deficiencies of the prior art, the present application provides a pipe jacking construction safety risk intelligent evaluation method based on multi-source data fusion. Based on historical monitoring data, by establishing a construction topology network, combining a GAM model and machine learning iterative training, not only can the settlement risk curve be accurately predicted, but also the key risk driving factors can be accurately identified, the multi-source data fusion process can be indirectly refined, and the frequency of monitoring and collection can be automatically updated. The machine learning model is introduced to iteratively train each batch of monitoring data, evaluate the abnormal fluctuation of structural deformation parameters, identify effective risks, further determine out-of-bound behavior, and timely alarm, solving the problems raised in the background art.

[0007] (II) Technical solutions

[0008] To achieve the above purposes, the present application is implemented by the following technical solutions:

[0009] The application provides a pipe jacking construction safety risk intelligent evaluation method based on multi-source data fusion, which comprises the following steps:

[0010] Obtain historical monitoring data of pipe jacking construction, including ground penetrating radar images, engineering operation parameters and structural deformation parameters;

[0011] Based on the historical monitoring data, a construction topology network is established, the risk driving factors are identified in combination with a preset machine learning model, and the monitoring data acquisition frequency is automatically adjusted; a machine learning model is introduced, each batch of monitoring data is iteratively trained, the abnormal fluctuation of the structural deformation parameters is evaluated, and the risk degree of geological mutation is identified; wherein, the risk degree includes effective risk and ineffective risk;

[0012] Based on the construction topology network, in combination with the effective risk, the operation safety strategy is executed to determine the out-of-bound behavior.

[0013] Further, the engineering operation parameters at least include jacking speed, machine head posture, cutter head torque and construction process execution state code value; the structural deformation parameters at least include structural stress, joint deformation and subsidence depth.

[0014] Further, the construction topology network is established, the risk driving factors are identified in combination with a preset GAM model, including:

[0015] Based on the historical monitoring data, the three-dimensional space coordinate sequence of the construction axis is extracted, the B-spline curve fitting is used to generate the construction path; a plurality of monitoring points are arranged along the construction path at a preset interval, and the spatial adjacent monitoring point clusters are mapped to the topology nodes; the topology edges are constructed between the adjacent topology nodes to construct the construction topology network;

[0016] Initialize the GAM model for each topology node, analyze the linear correlation of the engineering operation parameters, the ground penetrating radar images and the settlement risk, and configure the topology-aware feature extractor; if it is linear correlation, predict the settlement risk curve after N times of future detection to obtain the settlement stage; if it is nonlinear correlation, perform topology-aware feature preprocessing, select features through multicollinearity diagnosis, divide the historical monitoring data into training set and validation set in time sequence, and use the topology-constrained regularization method to prevent overfitting; select the optimal hyperparameters through distributed cross-validation, identify the risk driving factors based on the feature importance ranking; wherein, the risk driving factors at least include the settlement rate and the stress growth rate.

[0017] Further, the monitoring data acquisition frequency is automatically adjusted, including:

[0018] Calibrate the association relationship of nodes and links of the construction topology network; wherein, the nodes are monitoring points participating in the construction process in the construction topology network, including but not limited to the starting work well, the receiving work well, the pipe jacking machine operation tunneling end, and the pipe joint rigid connection node; the link is the actual pipe jacking construction path connecting two nodes, including the straight jacking section, the curve turning section, the branch angle jacking section, and the main and branch pipe connection section, and the link carries at least a risk driving factor;

[0019] Based on the construction topology network, the basic monitoring frequency of each topology node is obtained, the compensation monitoring frequency of each topology node caused by the construction advancing direction is calculated through link compensation analysis, the basic monitoring frequency and the compensation monitoring frequency are fused, and the monitoring data acquisition frequency is updated.

[0020] Further, a machine learning model is introduced to iteratively train each batch of monitoring data and evaluate the abnormal fluctuation of the structure deformation parameter to identify the risk degree of geological mutation, including:

[0021] Each batch of monitoring data is taken as the input data of the machine learning model, the risk driving factor and the abnormal fluctuation are taken as the input features, the real-time risk probability of each topology node of the pipe jacking construction is obtained through probability reasoning, and compared with the preset risk threshold in the model to obtain the risk degree of geological mutation: if the real-time risk probability is greater than or equal to the risk threshold, it is marked as effective risk, and if the real-time risk probability is less than the risk threshold, it is marked as invalid risk.

[0022] Further, the machine learning model adopts a Bayesian network architecture.

[0023] Further, the abnormal fluctuation of the structure deformation parameter is evaluated, including:

[0024] The structure deformation parameters of each batch of links are collected and grouped into a set P, a structure deformation parameter is selected from P based on the permutation and combination principle, a risk transmission task set C(b, a) is constructed, and the risk transmission task is a risk influencing behavior initiated on the monitoring link; wherein, b represents the structure deformation parameter at any time under a certain batch;

[0025] The risk driving direction and the influence degree are determined based on the risk transmission task set, and the influence degree is determined based on the cosine similarity; the mean and the fluctuation value of the structure parameter in any risk transmission task are analyzed, a three-dimensional monitoring vector is formed combined with the influence degree, and a judgment vector is constructed based on the three-dimensional monitoring vector; the judgment vector is introduced into a preset MLP judgment model, and a judgment result is output; if the judgment result obtained by any judgment vector is greater than a preset judgment threshold, the construction area corresponding to the risk transmission task is marked as an abnormal working condition area;

[0026] Identify the settlement stage where the abnormal working condition area is located, call the initial weight built-in in the database under this stage, including the first weight, the second weight and the third weight, and adaptively update the weight, combine the structure deformation parameters with the corresponding weight, including multiplying the subsidence depth and the first weight, multiplying the structure stress and the second weight, multiplying the joint deformation and the third weight, adding the multiplied results to obtain the fluctuation index, and determining the abnormal fluctuation degree of the structure deformation parameters.

[0027] Further, determining the abnormal fluctuation degree of the structure deformation parameters comprises:

[0028] Draw the time variation curve of the fluctuation index, and preset a standard curve, obtain the end point of the time variation curve deviating from the standard curve, take the end point as the center and wc% as the radius, and perform a circle drawing operation to form a circular determination range; extract a first line segment deviating from the end point, if all points on the first line segment are located within the circular determination range, determine that the abnormal fluctuation degree of the fluctuation index is small; if there are at least two points on the first line segment located outside the circular determination range, determine that the abnormal fluctuation degree of the fluctuation index is large.

[0029] Further, the initial first weight is greater than the initial second weight, which is greater than the initial third weight.

[0030] Further, the execution of the operation safety strategy comprises:

[0031] Obtain the real-time position of the construction personnel as a dynamic layer, through the construction topological network, bind a plurality of sensors of each topological node as a static layer, build an intelligent map of the pipe jacking construction site; count the topological nodes with effective risks, determine the risk position in combination with the intelligent map; by calculating the distance between each construction personnel and the risk position, determine whether the construction personnel has a border crossing behavior, and map the real-time position of the construction personnel to the pipe jacking construction topological network, and obtain the optimal path of the construction personnel from the current position to the safe position by using the A* algorithm.

[0032] (Three) beneficial effects

[0033] The present application provides a pipe jacking construction safety risk intelligent evaluation method based on multi-source data fusion, which has the following beneficial effects:

[0034] 1. This invention separates linear and nonlinear processing. For linear correlation, it predicts and outputs the settlement risk curve after N future detections. For nonlinear correlation, it performs topology-aware feature preprocessing, filters features through multicollinearity diagnosis and topology constraint regularization, and obtains risk driving factors by ranking feature importance, thereby improving model training efficiency. Then, by calibrating the association relationship between nodes and links in the construction topology network, and through link compensation analysis, it dynamically adjusts the monitoring frequency of the current topology nodes based on the construction progress direction, achieving precise association between nodes and links, avoiding indiscriminate adjustments, improving monitoring targeting, and laying a data foundation for subsequent refined analysis.

[0035] 2. This invention constructs a risk transmission task set through the principle of permutation and combination, introduces cosine similarity to calculate the influence, and constructs a comprehensive judgment vector by combining the mean and fluctuation values. It adopts MLP intelligent judgment to evaluate the abnormal fluctuation of structural deformation parameters, which greatly reduces the false alarm rate. In the process of model calculation, it avoids the traditional data superposition, refines the fusion of multi-source data, effectively identifies the core source of risk, greatly reduces the false alarm rate, and improves the efficiency of risk judgment.

[0036] 3. By binding effective risks to construction topology network nodes, the system can accurately locate the specific location of risks in three-dimensional space. By constructing an intelligent map, static and dynamic layers are spatially superimposed to monitor the number, location, and dynamics of construction personnel in a limited space in real time. In case of danger, one-click emergency call can be made, avoiding reliance solely on alarms from sensors or devices, and improving the efficiency and success rate of emergency rescue. Attached Figure Description

[0037] Figure 1 This is a schematic diagram illustrating the steps of an intelligent assessment method for safety risks in pipe jacking construction, according to an exemplary embodiment. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example:

[0040] This invention provides an intelligent assessment method for safety risks in pipe jacking construction based on multi-source data fusion; Figure 1 This is a schematic diagram illustrating the steps of an intelligent assessment method for safety risks in pipe jacking construction according to an exemplary embodiment; please refer to [link / reference]. Figure 1 The method includes the following steps:

[0041] S1: Acquire historical monitoring data for pipe jacking construction, including ground-penetrating radar images, engineering operation parameters, and structural deformation parameters. Engineering operation parameters include, but are not limited to, jacking speed, cutterhead attitude, cutterhead torque, and construction procedure execution status codes. Structural deformation parameters include, but are not limited to, structural stress, joint deformation, and settlement depth. It should be noted that several types of sensors or devices are installed at key monitoring sections along the pipe jacking construction route, with the geological sensors at these sections aligned with key geological interfaces to capture data from each monitoring section. Acquisition of historical monitoring data: Ground-penetrating radar images: Based on the characteristics of ground-penetrating radar... Using a ground-penetrating radar detector, the detection frequency is selected to separate the effective signal from the background noise and capture ground-penetrating radar images; jacking speed: jacking displacement is recorded in real time by displacement sensors, and the rate is calculated by combining the timestamps. The sensors are installed on the piston rod of the main jacking cylinder or the propulsion guide rail, outputting displacement data every 0.1 to 1 second. The ratio of the displacement change to the time interval is marked as the jacking speed; jacking head attitude: the angular acceleration and linear acceleration of the jacking head are detected in real time by a three-axis accelerometer and a three-axis gyroscope to obtain the pitch angle, yaw angle, and roll angle. Laser guidance sensors are used for positioning, and the horizontal or vertical deviation of the jacking head axis from the design axis is calculated. The output shows complete headstock attitude parameters; cutterhead torque: the torsional deformation of the cutterhead drive shaft is monitored by a torque sensor, and the output torque value is calculated. A non-contact torque sensor is installed on the drive shaft of the cutterhead drive motor of the pipe jacking machine to capture torque fluctuations when the cutterhead cuts the strata; construction process execution status code value: the construction process includes, but is not limited to, cutterhead start, jacking start, and pipe section assembly. The PLC controller automatically triggers the coding, mapping the construction process to a preset digital code, and the generated code value is stored synchronously with the timestamp to form a process sequence chain; for example: 001: cutterhead start, 002: jacking; structural deformation parameters are associated with pipe sections. For surface safety, sensors need to be installed on the inner wall of the pipe sections and at surface monitoring points to capture stress, deformation, and settlement in real time. For structural stress, surface settlement observation points are set up at intervals of several times the pipe diameter on both sides of the pipe axis using a total station to record three-dimensional coordinates and calculate displacement. Resistance strain gauges are attached to key sections such as pipe section interfaces and the top / bottom of the pipe body to monitor static stress, or fiber optic grating sensors are installed for long-distance monitoring of dynamic stress. Additionally, structural stress reflects stress changes at the bottom of the caisson. As an auxiliary sequence, the structural stress sequence provides information on changes in bottom structural stress to improve the accuracy of caisson sinking attitude prediction, using q... j ∈R D The structural stress at D stress points at time j, and the structural stress parameter sequence Q at D stress points over J historical times, can be expressed as: Q = (q1, q2, ..., q J )∈R J*D In the formula, q1, q2, ... Jrespectively represent the structural stress parameter sequence monitored at the 1st time, the 2nd time, and the Jth time; joint deformation: a laser displacement sensor is installed on the inner wall of the joint, and a reflector plate is attached to the opposite pipe section to ensure that the laser is aligned, and to measure the relative displacement, and to measure the opening and misalignment; subsidence speed: a geological radar transmits high-frequency electromagnetic waves and receives reflected waves from different soil layer interfaces, and according to the wave speed and reflection characteristics, the soil layer type is inverted, for example: there is a difference between the reflection signals of sand and clay; between the monitoring point and the reference point, a static level gauge probe is pre-buried below the monitoring point, and a wire tension sensor directly measures the vertical displacement of the monitoring point relative to the reference point through the change in the length of the steel wire, a negative value indicates the subsidence depth; the static level gauge senses the liquid level difference through the probe, and outputs the subsidence of the monitoring point relative to the reference point; the data is compared with the initial elevation to obtain the cumulative subsidence depth; the historical monitoring data is integrated in the format of construction phase-parameter type-monitoring location to obtain a geological, construction and structural dimension data set; in the subsequent data processing process, the monitoring data is normalized to eliminate the dimensional influence, and the formula calculation process has physical significance.

[0042] S2: based on the historical monitoring data, a construction topology network is established, risk driving factors are identified in combination with a preset GAM model, and the monitoring data acquisition frequency is automatically adjusted; a machine learning model is introduced, each batch of monitoring data is iteratively trained, abnormal fluctuations of the structural deformation parameters are evaluated, and the risk degree of geological mutation is identified;

[0043] The construction topology network is established, and the risk driving factors are identified in combination with the preset GAM model, including: based on the historical monitoring data, the three-dimensional spatial coordinate sequence of the construction axis is extracted, which is essentially the spatial point record of the actual pipe jacking path. Before extracting the three-dimensional spatial coordinate sequence, the monitoring data needs to be preprocessed to ensure the consistency and accuracy of the extracted coordinates. In this embodiment, different coordinate systems may be used for different equipment in pipe jacking construction, such as the machine head local coordinate system for the machine head attitude sensor and the city independent coordinate system for the total station. The coordinate system needs to be unified, and the origin is set at the center of the starting well, the X axis is the jacking direction, the Y axis is the vertical jacking direction, the Z axis is the elevation direction, and the upward is positive. At the same time, the abnormal values are removed, and the data is operated incompletely. Then, based on the preprocessed data, the construction axis coordinates corresponding to different mileages are extracted according to the principle of using high-frequency real-time data first and low-frequency high-precision data for calibration, such as extracting continuous coordinates from the monitoring data of the machine head attitude. Example 1: the initial mileage is 0 m, the coordinate is (1000.0, 500.0, -8.0), and after jacking 5 cm, the machine head coordinate is (1000.05, 500.0, -8.0), so the construction axis coordinate is (1000.05, 500.0, -8.0). The extraction of the three-dimensional coordinate sequence of the construction axis is essentially to select, calibrate and connect the spatial points that can reflect the actual jacking path from the multi-source historical monitoring data to form a coordinate set that can reflect the real-time jacking trajectory and meet the engineering accuracy requirements. The sequence can be directly used for subsequent analysis. Based on the three-dimensional spatial coordinate sequence, the B-spline curve fitting is used to generate the construction path. A plurality of monitoring points are arranged along the construction path at a preset interval, and the spatial adjacent monitoring point clusters are mapped as topology nodes. The topology edges are constructed between adjacent topology nodes to construct the construction topology network, and its representation form is: G=(V, E), wherein V={v1, v2, …, vn} represents the topology node set, v1, v2, …, vn represent the 1st, 2nd, and n-th topology nodes, respectively, E={e1, e2, …, en} represents the connection relationship between nodes, e1, e2, …, en represent the connection relationship between the 1st, 2nd, and n-th topology nodes, respectively. n} indicates the topology node set, v1, v2, v n represent the 1st, 2nd, and n-th topology nodes, respectively, E={e 12 ,e 23 ,…,e (n-1)n} represents the connection relationship between nodes, e 12 , e 23 , e (n-1)nrespectively represent the edges formed between the first and second topological nodes, the edges formed between the second and third topological nodes, and the edges formed between the n-1th and nth topological nodes; in addition, new activated construction partitions can be identified based on the construction progress, corresponding topological nodes can be added, and topological edges with adjacent nodes can be established; completed construction partitions are detected, and corresponding nodes are removed from the topological network; the construction path is fitted through a B-spline curve, and the actual three-dimensional trend of the pipeline is accurately restored, providing a real spatial reference for risk analysis; the spatially adjacent monitoring points are clustered and mapped as topological nodes, avoiding the randomness of single-point data, enhancing the robustness and representativeness of the data, and better reflecting the overall state of a section; through the construction of topological edges, a foundation is laid for risk tracing and diffusion early warning; then, a GAM model is initialized for each topological node, the linear correlation between the engineering operation parameters, the ground penetrating radar image and the settlement risk is analyzed, and a topologically aware feature extractor is configured; in this embodiment, first, the ground penetrating radar image is subjected to Gaussian filtering for noise reduction, the Sobel operator is used to extract the geological stratification profile of the construction area, and edge detection is performed; the pixel gradient amplitude and direction are calculated, the local maximum gradient is retained through non-maximum suppression, the geological stratification edge is refined, and based on a double threshold, a continuous geological profile is generated to locate the key stratigraphic interface; a lightweight U-Net model is used to obtain image features, including but not limited to texture roughness, which is used to reflect soil uniformity, interface clarity, which is used to reflect the degree of distinctness of different stratigraphic boundaries, and abnormal area size, which is used to reflect the size of cavities or weak zones; then, the image features are taken as one set of inputs of the GAM model, the engineering operation parameters are taken as another set of inputs of the GAM model, the surface settlement is taken as the objective function, and the influence relationship diagram is formed by independently fitting each input variable; if the influence relationship diagram is roughly a straight line, for example, upward or downward, it is judged as a linear relationship, representing that the influence of the parameter on the risk is uniform and constant; if the influence relationship diagram is a curved line, for example, an S-shaped curve, a parabolic curve or a curve with a platform, it is judged as a nonlinear relationship, representing that the influence degree or direction of the parameter on the risk will change in different value intervals; if it is linearly correlated, the settlement risk curve after N times of future detection is predicted; it should be noted that the settlement risk curve has three stages: the initial settlement stage, the middle settlement stage and the final settlement stage; the initial settlement stage, the middle settlement stage and the final settlement stage can be dynamically adjusted according to the soil properties and the settlement depth and rate: the initial settlement stage, for example, the settlement depth is less than or equal to 2m: mainly slow sinking, the settlement rate is less than or equal to 0.5m / d, it is necessary to ensure the stability of the construction track, symmetrical soil taking and prevention of deviation; the middle settlement stage: the settlement rate is less than or equal to 1m / d, it is necessary to maintain uniform soil taking; the final settlement stage: for example, the settlement rate is less than or equal to 0.2m / d, mainly for rectification, strictly control the depth of soil covering under the blade foot; the above-mentioned values are all for example, the specific value is set according to the actual situation; if it is nonlinear correlation, perform feature preprocessing of topological perception, screen features through multiple collinearity diagnosis, divide historical monitoring data into training set and validation set according to time sequence, use regularization method of topological constraint to prevent overfitting; select the optimal hyperparameter through distributed cross-validation, identify risk driving factors based on feature importance ranking; wherein, the risk driving factors include but are not limited to settlement rate; it should be pointed out that the settlement rate reflects whether the stratum is stable, directly warning the safety risk of the structure; the significance of the above analysis lies in: separating linear and nonlinear through linear and nonlinear separation processing: distinguishing linear and nonlinear relationship, directly predicting linear part and finely preprocessing nonlinear part; this not only guarantees the efficiency, but also ensures that the model can capture the complex real world law; through multiple collinearity diagnosis and regularization of topological constraint, the selected features are truly independent and have a significant impact on risk, and finally through feature importance ranking, a direct and reliable decision basis is provided for adjusting construction parameters.

[0044] The automatic adjustment monitoring data collection frequency comprises: calibrating the correlation of nodes and links of the construction topology network; wherein the nodes are monitoring points participating in the construction process in the construction topology network, including but not limited to the starting work well, the receiving work well, the pipe jacking machine operation tunneling end, and the pipe joint rigid connection node; the link is the actual pipe jacking construction path connecting two nodes, including the straight jacking section, the curve turning section, the branch angle jacking section, and the main branch pipe connection section, and the link carries at least a risk driving factor; the basic monitoring frequency of each topology node is obtained based on the construction topology network, the compensation monitoring frequency of each topology node due to the construction advancing direction is calculated through link compensation analysis, including: obtaining the pipe jacking construction path type where the link section is located or adjacent, at least including one of the straight jacking section, the curve turning section, the branch angle jacking section, and the main branch pipe connection section, and assigning the corresponding correction weight of each path type, and the correction weight is updated adaptively based on the genetic algorithm; wherein if multiple pipe jacking construction path types are associated, the largest correction weight is selected to reflect that the high-risk type should dominate the monitoring frequency adjustment, and the initial correction weight is set as: curve turning section = correction weight 0.4 > branch angle jacking section = correction weight 0.3 > main branch pipe connection section = correction weight 0.2 > straight jacking section = correction weight 0.1;obtain the correction weight of the path type involved in the construction advancing direction, multiply the basic monitoring frequency by the correction weight to obtain the compensation monitoring frequency; fuse the basic monitoring frequency and the compensation monitoring frequency to update the monitoring data acquisition frequency; wherein, the fusion mode adopts addition combination to ensure that the monitoring frequency of the area with high risk is increased; in addition, a loss function can also be defined to measure the difference between the corrected monitoring data accuracy and the theoretical accuracy; minimize the loss function to find the optimal acquisition frequency parameter, and stabilize the update period of each monitoring parameter within a specific target range; in addition, the risk propagation direction can be determined by the change sequence of the monitoring data of the front-end monitoring node and the rear-end monitoring node in the construction advancing direction; when stationary, the data of the front-end monitoring node and the rear-end monitoring node are both in the stable threshold range; when advancing forward, the data of the front-end monitoring node fluctuates first, and then the data of the rear-end monitoring node triggers a response; when adjusting in reverse, the data of the rear-end monitoring node recovers to stable first, and then the data of the front-end monitoring node returns to the threshold range; at the same time, the correction weight is updated adaptively based on the genetic algorithm, including: randomly generating an initial population in a high-dimensional target space, the population is composed of a plurality of particles, and a certain number of individuals are randomly generated at initialization, marked as weight combination, to ensure that all weight values are within a suitable range, and the range is between [0, 1], and the sum of the weight values is 1; based on the individual fitness, select individuals with high fitness as parents, use the roulette selection strategy to select particles from the population in turn and put them into the mating pool until the number of particles in the mating pool reaches the total number of particles; new individuals are obtained by selecting, crossing and mutating the population and added to the population to update the composition of the population; in the crossing process: select a crossing point to divide the parent's gene (weight vector) from this point, there are 3 potential crossing points, for example: crossing point = 1: cut between w1 and w2, the first half is [w1], and the second half is [w2, w3, w4]; crossing point = 2: cut between w2 and w3, the first half is [w1, w2], and the second half is [w3, w4]; crossing point = 3: cut between w3 and w4, the first half is [w1, w2, w3], and the second half is [w4]; wherein, w1, w2, w3, w4 are correction weights; at the same time, the crossing operation is not performed on every pair of parents, usually a crossover probability is set, for example: with a probability of 70%, the crossing is performed, and with a probability of 30%, the parent is kept unchanged; after updating the population, the average value or optimal solution of the population fitness no longer changes significantly, then the result is output and the training is stopped.

[0045] The machine learning model is introduced, each batch of monitoring data is iteratively trained, and the abnormal fluctuation of the structure deformation parameter is evaluated to identify the risk degree of geological mutation, including: taking each batch of monitoring data as the input data of the machine learning model, combining the risk driving factors and the abnormal fluctuation as the input features, obtaining the real-time risk probability of each topology node of the pipe jacking construction through probability reasoning, and comparing with the preset risk threshold in the model to obtain the risk degree of geological mutation: if the real-time risk probability is greater than or equal to the risk threshold, it is marked as effective risk, and if the real-time risk probability is less than the risk threshold, it is marked as invalid risk; wherein the machine learning model adopts a Bayesian network architecture; the risk threshold is set based on historical statistics, and the value is the sum of the average value of the historical risk probability and 2-3 times the standard deviation; the abnormal fluctuation of the structure deformation parameter is evaluated, including: collecting the structure deformation parameters of each batch of link segments and collecting them into a set P, selecting a structure deformation parameter from P based on the permutation and combination principle, constructing a risk transmission task set C(b, a), and the risk transmission task is a risk influencing behavior initiated on the monitoring link; wherein b represents the structure deformation parameter at any time under a batch; for example: obtaining the node number and link position of a link segment, such as K0+120 pipe joint connection and K0+150 ground settlement point; there are three types of core parameters: settlement depth, structure stress and joint deformation, a0 is 3, b0=2 is taken, permutation and combination is performed, and the risk transmission task set is {settlement depth+structure stress, settlement depth+joint deformation, structure stress+joint deformation} three risk transmission tasks, {settlement depth+structure stress} represents the change of settlement depth affecting the structure stress or the change of structure stress affecting the settlement depth, {settlement depth+joint deformation} represents the change of settlement depth affecting the joint deformation or the change of joint deformation affecting the settlement depth, and {structure stress+joint deformation} represents the change of structure stress affecting the joint deformation or the change of joint deformation affecting the structure stress. The change reflects the risk driving direction; the risk driving direction and the influence degree are determined based on the risk transmission task set, and the influence degree is determined based on the cosine similarity; it should be noted that the settlement depth reflects the vertical displacement of the ground or the pipe joint, the structure stress reflects the mechanical load borne by the pipe joint, and the joint deformation reflects the gap or fault amount of the pipe joint connection; using NLP technology, the settlement depth, structure stress and joint deformation are converted into embedding vectors, and the formula is:

[0046] , wherein, 、 represents the time series data vector of two structure deformation parameters; for example, assuming is [1.2, 1.5, 1.8, 2.1, 2.4, 2.7, 3.0], is [5.0, 5.5, 6.0, 6.5, 7.0, 7.5, 8.0], the influence degree (cosine similarity) is 0.99, if it is {settlement depth + structural stress}, it represents that the settlement depth increases synchronously when the structural stress increases, for example: the jacking force is too large, causing the pipe joint stress to exceed the limit, and at the same time, the stratum is extruded to cause settlement; if it is {settlement depth + joint deformation}, if the joint deformation increases, the settlement depth increases synchronously, for example: the joint sealing failure causes soil loss, which intensifies deformation and causes settlement; if it is {structural stress + joint deformation}, it represents that the joint deformation increases synchronously when the stress increases, for example: the pipe joint stress exceeds the limit and extrudes the joint, causing the joint gap to open; the mean value me and the fluctuation value fl of the structural parameters in any risk transmission task are analyzed, a three-dimensional monitoring vector is formed in combination with the influence degree Y, and a judgment vector is constructed based on the three-dimensional monitoring vector, including: [me + fl, Y] and [me - fl, Y]; the judgment vector is introduced into the preset MLP judgment model, and a judgment result is output; if the judgment result obtained by any judgment vector is greater than a preset judgment threshold, the construction area corresponding to the risk transmission task is marked as an abnormal working condition area; wherein the fluctuation value is one-half of the difference between the maximum value and the minimum value of the structural parameters in the time period; it should be noted that the MLP judgment model adopts a 4-layer architecture, which is input layer, hidden layer 1, hidden layer 2 and output layer in turn; the input layer includes 2 neurons for receiving 2 dimensions of the judgment vector, for example: me + fl and Y in [me + fl, Y]; in order to avoid weight deviation caused by different dimension values, the input vector needs to be processed by min-max normalization to map me ± fl and Y to the [0, 1] interval, eliminate the dimension influence, and ensure that the influence weight of each feature on the result is balanced during model training; the hidden layer 1 includes 8 neurons and adopts ReLU as the activation function, which is used to capture the first-order nonlinear relationship between me ± fl and Y, for example: when me ± fl increases, the increase rate of abnormal probability accelerates with the increase of Y; in order to prevent overfitting in small sample training, a dropout rate of 0.2 is set in the network; the hidden layer 2 includes 4 neurons and also adopts ReLU as the activation function, which is used to further fit the high-order coupling relationship between me ± fl and Y; the output layer includes 1 neuron and adopts Sigmoid as the activation function, which is used to output the judgment result, and its value range is 0 to 1, the closer to 1, the higher the confidence that the construction area corresponding to the risk transmission task is an abnormal working condition area;

[0047] The significance of the above analysis is: improving the risk transmission task set constructed by permutation and combination principle, introducing cosine similarity to calculate the influence degree, constructing a comprehensive judgment vector combining the mean value and the fluctuation value, using MLP intelligent judgment to evaluate the abnormal fluctuation of structural deformation parameters, avoiding traditional data superposition in the process of model calculation, refining multi-source data fusion, effectively identifying the core source of risk, greatly reducing the false positive rate, and improving the efficiency of risk judgment;

[0048] The abnormal condition area is identified to be in a settlement stage, initial weights including a first weight, a second weight and a third weight in the database at the stage are called, and the weights are adaptively updated. The structural deformation parameters are combined with the corresponding weights, including multiplying the settlement depth by the first weight, multiplying the structural stress by the second weight, and multiplying the joint deformation by the third weight. The results of the multiplications are added to obtain a fluctuation index to determine the abnormal fluctuation degree of the structural deformation parameters. The initial first weight is greater than the initial second weight, which is greater than the initial third weight. It should be noted that the abnormal settlement depth reflects the overall and rapidly spreading fatal risk, the abnormal structural stress reflects the local and controllable damage risk, and the abnormal joint deformation reflects the local and slowly developing leakage risk. Therefore, the initial first weight is set to be greater than the initial second weight, which is greater than the initial third weight. Then, the weights are adaptively updated. The initial settlement stage, the middle settlement stage and the final settlement stage of the pipe jacking construction are identified. The historical structural deformation parameters and the fluctuation index at each stage are called and marked as a set G. The abnormal fluctuation degrees of the elements in the G set are sequentially sorted from low to high. Every three values are taken as g test data sets for model training. The linear relationship between the abnormal fluctuation degree and the weight is measured by the mean square error method, and the weight is trained by the gradient descent method. After the training is completed, the initial adaptive weight of each settlement stage is obtained to realize the adaptive update of the weight. The abnormal fluctuation degree of the structural deformation parameters is determined. The time variation curve of the fluctuation index is drawn, and a standard curve is preset. The endpoints of the time variation curve deviating from the standard curve are obtained. The endpoints are taken as the center and wc% is taken as the radius to perform a circle drawing operation to form a circular determination range. The first line segment deviating from the endpoint is extracted. If all points on the first line segment are located within the circular determination range, it is determined that the abnormal fluctuation degree of the fluctuation index is small. If at least two points on the first line segment are located outside the circular determination range, it is determined that the abnormal fluctuation degree of the fluctuation index is large. The first line segment represents the connection line between the endpoint and the next data point. It should be noted that wc% is set based on the upper limit of normal fluctuation, not a fixed value, which is suitable for different strata. For example, if the ground is soft soil, the radius increases synchronously. The endpoint deviates. A minimum fluctuation threshold is preset. For any data point, if the corresponding fluctuation index is greater than the preset minimum fluctuation threshold, and the fluctuation index of the next consecutive data point is greater than the preset minimum fluctuation threshold, it is indicated that the endpoint deviates. The standard curve is obtained by acquiring the historical data of the fluctuation index under normal construction conditions in the pipe jacking construction. The mean value and the upper limit of the normal fluctuation of the historical data are calculated. The upper limit of the fluctuation is 1.96 times the standard deviation of the historical data verified to conform to the normal distribution by K-S test. The standard variation curve includes an upper boundary: mean value + fluctuation upper limit, and a lower boundary: mean value - fluctuation upper limit.

[0049] The significance of the above analysis is that: by constructing the risk transmission task set through the permutation and combination principle, the possible propagation path of the risk on the topology edge is actively simulated, the influence degree is introduced by the cosine similarity calculation, the influence strength and direction of the parameter change of one node on another node can be accurately quantified, the risk tracing and propagation path prediction are realized; the mean, volatility and influence degree are fused into a comprehensive judgment vector in three different dimensions, the MLP intelligent judgment is adopted, the complex mapping relationship between the high-dimensional, nonlinear characteristics and the abnormal is learned, the judgment result is more stable and accurate than any single rule or linear model, and the false positive rate is greatly reduced.

[0050] S3: based on the construction topology network, combined with the effective risk, executing the operation safety strategy to determine the out-of-bound behavior; wherein, the operation safety strategy includes: obtaining the real-time position of the construction personnel through the positioning device adapted to the pipe jacking construction, and the positioning device adopts the UWB ultra-wideband positioning technology, the positioning accuracy is ≤1m, and the data update frequency is ≤5s; statistics the proportion of the effective risk of each topology node, and taking the real-time position of the construction personnel as the dynamic layer of the intelligent map; calling a plurality of sensors bound to each topology node as the static layer of the intelligent map, and the static layer constitutes a background risk field reflecting the inherent risk distribution of the construction environment itself, and the data is relatively stable under the condition that the equipment layout is unchanged; building the intelligent map of the pipe jacking construction site based on the static layer and the dynamic layer; statistics the topology node where the effective risk exists, and determines the risk position combined with the intelligent map; judging whether each construction personnel has out-of-bound behavior by calculating the distance between each construction personnel and the risk position, if the distance is less than the distance threshold, it is determined that there is out-of-bound behavior, and an alarm signal is sent, and the A* algorithm is used to calculate the optimal path from the current position of the personnel to the safe area, including: mapping the real-time position of the construction personnel to the pipe jacking construction topology network, if the position of the construction personnel is at the coordinate boundary of two topology nodes, the target mapping node is determined according to the straight line distance or the residence time of the construction personnel to the center of the two nodes; specifically: assuming that the three-dimensional coordinate range of each topology node is [X min ,X max ]×[Y min ,Y max ]×[Z min ,Z max ], wherein, the X axis is the jacking direction, representing the horizontal extension direction of the pipe jacking construction, then X min represents the minimum boundary value of the topology node in the X axis direction, X max represents the maximum boundary value of the topology node in the X axis direction, the Y axis is perpendicular to the jacking direction, which is horizontal to the X axis, then Y min represents the minimum boundary value of the topology node in the Y axis direction, Y maxrepresents the maximum boundary value of the topological node in the Y-axis direction, the Z-axis is the elevation direction, and represents the direction of pipe burial depth, and Z min represents the minimum boundary value of the topological node in the Z-axis direction, Z max represents the maximum boundary value of the topological node in the Z-axis direction; if the real-time coordinates (Xs, Ys, Zs) of the construction personnel fall within the coordinate range of a single topological node, the real-time coordinates are directly mapped to the topological node; wherein Xs represents the real-time coordinate value of the current position of the construction personnel in the X-axis, Ys represents the real-time coordinate value of the current position of the construction personnel in the X-axis, and Zs represents the real-time coordinate value of the current position of the construction personnel in the X-axis; if the real-time coordinates of the construction personnel are at the coordinate boundaries of two nodes, the distance from the construction personnel to the center of a topological node and the distance from the construction personnel to the center of another topological node are calculated; if the two distances are not equal, the topological node with the closer distance is mapped; if the two distances are equal, the length of time that the construction personnel stays within the coordinate range of the two topological nodes is counted, and the topological node with the longer length of time is mapped; if the distance is greater than or equal to the distance threshold, it is determined that there is no out-of-bound behavior;

[0051] The significance of the above analysis is that by binding the effective risk to the construction topological network node, the system can accurately locate the specific position of the risk in the three-dimensional space, construct an intelligent map by spatially superimposing the static layer and the dynamic layer, and real-time master the number, position and dynamics of the construction personnel in the limited space, and can call for help at one key in case of danger; avoid relying only on the alarm of sensors or devices, so that the control strategy is more intelligent and reasonable, unnecessary interference is reduced, and the efficiency and success rate of emergency rescue are improved.

[0052] In the application, the several formulas involved are calculated by taking the values after de-dimensioning, and the formula is obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and the formula is set by a person skilled in the art according to the actual situation.

[0053] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially. A person of ordinary skill in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions.

[0054] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, and may be located in one place, or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment of the present application according to actual needs.

[0055] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for intelligent assessment of safety risks in pipe jacking construction based on multi-source data fusion, characterized in that, The method comprises the following steps: Obtain historical monitoring data of pipe jacking construction, including ground penetrating radar images, engineering operation parameters and structural deformation parameters; Based on the historical monitoring data, establish a construction topology network, identify risk driving factors in combination with a preset GAM model, and automatically adjust the monitoring data collection frequency; introduce a machine learning model, iteratively train each batch of monitoring data, evaluate the abnormal fluctuation of the structural deformation parameters, and identify the risk degree of geological mutations; Wherein, the risk degree includes effective risk and ineffective risk; Wherein, in combination with the preset GAM model to identify risk driving factors, including: Based on the construction topology network, initialize the GAM model for each topology node, analyze the linear correlation of engineering operation parameters, ground penetrating radar images and settlement risk, and configure a topology-aware feature extractor; if it is linearly correlated, predict the settlement risk curve after N times of future detection to obtain the settlement stage; if it is nonlinearly correlated, perform topology-aware feature preprocessing, select features through multicollinearity diagnosis, divide historical monitoring data into training set and validation set in chronological order, and use topology-constrained regularization method to prevent overfitting; select the optimal hyperparameters through distributed cross-validation, identify risk driving factors based on feature importance ranking; wherein, the risk driving factors at least include settlement rate; Wherein, evaluating the abnormal fluctuation of the structural deformation parameters, including: Collect the structural deformation parameters of each batch of link segments and collect them into set P, select a structural deformation parameter from P based on the permutation combination principle, construct risk transmission task set C(b, a), and the risk transmission task is a risk influencing behavior initiated on the monitoring link; wherein, b represents the structural deformation parameter at any time under any batch; Determine the risk driving direction and influence degree based on the risk transmission task set, and the influence degree is determined based on the cosine similarity; analyze the mean and fluctuation value of the structural parameters in any risk transmission task, form a three-dimensional monitoring vector based on the influence degree, and construct a judgment vector based on the three-dimensional monitoring vector; import the judgment vector into the preset MLP judgment model, output the judgment result, if the judgment result obtained by any judgment vector is greater than the preset judgment threshold, mark the construction area corresponding to the risk transmission task as an abnormal working condition area; Identify the settlement stage of the abnormal working condition area, retrieve the initial weights built in the database at this stage, including the first weight, the second weight and the third weight, and adaptively update the weights, combine the structural deformation parameters with the corresponding weights, including multiplying the subsidence depth by the first weight, multiplying the structural stress by the second weight, and multiplying the joint deformation by the third weight, adding the multiplied results to obtain the fluctuation index, and determining the abnormal fluctuation degree of the structural deformation parameters; Based on the construction topology network, in combination with the effective risk, execute the operation safety strategy, and determine the out-of-bound behavior.

2. The method of claim 1, wherein, The engineering operation parameters include jacking speed, machine head posture, cutter torque and construction process execution state code value; the structural deformation parameters include structural stress, joint deformation and subsidence depth.

3. The method of claim 1, wherein, The construction topology network is established, including: Based on historical monitoring data, a three-dimensional spatial coordinate sequence of the construction axis is extracted, and a B-spline curve fitting is used to generate a construction path; a plurality of monitoring points are deployed along the construction path at preset intervals, and a cluster of spatially adjacent monitoring points is mapped as a topological node; and a topological edge is constructed between adjacent topological nodes to construct a construction topological network.

4. The method of claim 1, wherein, The automatic adjustment of the monitoring data collection frequency comprises: The correlation between the nodes and links of the construction topological network is calibrated; wherein the nodes are monitoring points participating in the construction process in the construction topological network, including the originating work well, the receiving work well, the pipe jacking machine operation driving end, and the pipe joint rigid connection node; the links are actual pipe jacking construction paths connecting two nodes, including the straight jacking section, the curved turning section, the branch angle jacking section, and the main and branch pipe connection section, and the links carry at least a risk driving factor; Based on the construction topological network, the basic monitoring frequency of each topological node is obtained, the compensation monitoring frequency of each topological node due to the construction advancing direction is calculated through link compensation analysis, the basic monitoring frequency and the compensation monitoring frequency are fused, and the monitoring data collection frequency is updated.

5. The method of claim 1, wherein, A machine learning model is introduced, and each batch of monitoring data is iteratively trained to evaluate the abnormal fluctuation of the structure deformation parameter to identify the risk degree of geological mutation, comprising: Each batch of monitoring data is taken as the input data of the machine learning model, the risk driving factor and the abnormal fluctuation are taken as the input features, the real-time risk probability of each topological node of the pipe jacking construction is obtained through probability reasoning, and compared with the preset risk threshold in the model to obtain the risk degree of geological mutation: if the real-time risk probability is greater than or equal to the risk threshold, it is marked as an effective risk, and if the real-time risk probability is less than the risk threshold, it is marked as an invalid risk.

6. The method of claim 5, wherein the method further comprises: The machine learning model adopts a Bayesian network architecture.

7. The method of claim 1, wherein the method further comprises: The determination of the abnormal fluctuation degree of the structure deformation parameter comprises: A time variation curve of the fluctuation index is drawn, a standard curve is preset, the endpoints of the time variation curve deviating from the standard curve are obtained, the endpoints are taken as the center of a circle with wc% as the radius, and a circle operation is performed to form a circular determination range; a first line segment deviating from the endpoints is extracted, if all points on the first line segment are located within the circular determination range, it is determined that the abnormal fluctuation degree of the fluctuation index is small; if there are at least two points on the first line segment located outside the circular determination range, it is determined that the abnormal fluctuation degree of the fluctuation index is large.

8. The method of claim 1, wherein the method further comprises: The initial first weight is greater than the initial second weight, which is greater than the initial third weight.

9. The method of claim 1, wherein, The execution of the operation safety strategy comprises: The real-time position of the construction personnel is obtained and taken as a dynamic layer, a plurality of sensors bound to each topological node are taken as a static layer through the construction topological network, and an intelligent map of the pipe jacking construction site is built; the topological nodes with effective risks are counted, and the risk position is determined in combination with the intelligent map; the distance between each construction personnel and the risk position is calculated to determine whether the construction personnel has a border crossing behavior, and the real-time position of the construction personnel is mapped to the pipe jacking construction topological network to obtain the optimal path of the construction personnel from the current position to the safe position by using the A* algorithm.

Images