Intelligent risk early warning method and system for prefabricated box girder production in bridge engineering

Abstract

The invention relates to the technical field of risk early warning, in particular to an intelligent risk early warning method and system for prefabricated box girder production in bridge engineering. The method comprises the following steps: constructing a multi-dimensional data acquisition network for data acquisition; carrying out data preprocessing on the collected multi-dimensional data; constructing a multi-source data fusion engine and establishing a data quality scoring mechanism to realize data fusion; performing risk prediction on the fused data by using the multi-modal fusion analysis model; and performing intelligent linkage response based on the predicted risk. Through reinforcement learning scheduling and equipment linkage control, intelligent collaboration of production elements such as personnel, equipment and process is realized, the key process efficiency is improved by 20%-30%, and the resource waste is reduced by more than 15%.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk early warning, and in particular to a method and system for intelligent risk early warning in prefabricated box girder production in bridge engineering. BACKGROUND

[0002] In the current field of bridge engineering, prefabricated box girder production technology has made significant progress. From the process point of view, advanced processes such as high-level prefabrication have been applied. For example, in the construction of some large sea-crossing bridges, by raising the prefabrication pedestal by one beam transport trolley height, and cooperating with the movable bottom mold, the efficient transfer of the box girder between the prefabrication and beam storage pedestal is realized, which reduces the dependence on large lifting equipment, saves engineering cost and site space. In terms of equipment intelligence, some beam yards introduce intelligent batching monitoring systems, which use intelligent weighing devices to strictly control the proportion of raw material feeding, trace and monitor the raw materials, greatly improving the batching accuracy and the quality of prefabricated box girders, and changing the labor-intensive heavy physical work into easy computer operation, reducing labor input. At the same time, intelligent tensioning, intelligent grouting and other technologies are also being gradually promoted, which real-time supervision of production links through digital means ensures that key parameters such as prestressed tensioning value meet the specification requirements.

[0003] However, there are still some key problems to be solved in the prior art: Most of the beam yard data collection types are relatively single, mainly focusing on conventional concrete strength, appearance size and other parameters, such as the micro-performance changes of raw materials in the production process, the micro-strain of the structure under complex working conditions, and other key data lack effective collection means, making it difficult to fully reflect the real quality status of prefabricated box girders. And in terms of data transmission, a large number of traditional wired transmission or unstable wireless networks are used. In the face of harsh construction environment (such as heavy rain, strong electromagnetic interference), data packet loss and delay are serious, which leads to the real-time data in the production process cannot be timely and accurately fed back to the management system, affecting the timely judgment and treatment of production risks.

[0004] Current risk assessment mainly relies on experience judgment and simple index threshold setting, and cannot fully consider the complex correlation between multiple factors. For example, when judging the quality risk of concrete pouring, only a single index such as slump is used, while the comprehensive influence of factors such as temperature, vibration time, and raw material mix proportion fluctuation is ignored, resulting in low warning accuracy, and often false positives or false negatives. At the same time, risk early warning is often lagging, when an anomaly is detected, the problem may have had a substantial impact on the quality of the prefabricated box girder, and effective measures cannot be taken to intervene in the early stage.

[0005] Although some links are intelligent, there is a lack of effective coordination between links. The data of steel processing, concrete pouring and prestressed tensioning are independent of each other, and cannot form an organic whole, so it is difficult to optimize the production process from the overall point of view. Moreover, in terms of decision support, the existing data processing and analysis system cannot provide managers with in-depth decision-making suggestions. For example, when facing complex situations such as fluctuation of raw material prices and tight construction period, it cannot quickly generate a scientific and reasonable production scheduling plan, resulting in low production efficiency and serious resource waste. SUMMARY

[0006] In order to solve the above-mentioned problems, the present application provides a precast box girder production intelligent risk early warning method and system in bridge engineering.

[0007] In the first aspect, the present application provides a precast box girder production intelligent risk early warning method in bridge engineering, which adopts the following technical scheme: A precast box girder production intelligent risk early warning method in bridge engineering, comprising: Constructing a multi-dimensional data acquisition network for data acquisition; Data preprocessing of the collected multi-dimensional data; Constructing a multi-source data fusion engine and establishing a data quality scoring mechanism to realize data fusion; Using a multi-modal fusion analysis model to predict the risk of the fused data; Intelligent linkage response based on predicted risk.

[0008] Further, the construction of the multi-dimensional data acquisition network for data acquisition includes collecting raw material traceability data based on a consortium chain architecture through a blockchain; real-time monitoring of the micro-strain changes of the structure during concrete pouring and tensioning through a grating strain sensor; and laying noise monitors and dust concentration sensors to collect noise and dust data; laying multi-spectral cameras and laser radars in a spatially layered manner, and using fixed sensors and mobile inspection robots on the ground layer for equipment abnormal state inspection; using a dynamic sampling strategy to regularly collect concrete curing temperature data.

[0009] Further, the data preprocessing of the collected multi-dimensional data includes data preprocessing through a three-level preprocessing mechanism of sliding window filtering, outlier detection and linear interpolation before temperature data sampling, wherein the sliding window mean filtering method is used to filter out high-frequency noise collected by the temperature sensor; the temperature jump value exceeding the normal range is identified and marked based on the 3σ principle to realize outlier detection; and the marked outliers are smoothed and repaired by the linear interpolation method to avoid affecting the calculation of the change rate.

[0010] Further, the construction of the multi-source data fusion engine and the establishment of the data quality scoring mechanism include the integration of the BIM model lightweight engine, the three-dimensional binding of the prefabricated box girder BIM model and real-time monitoring data, the realization of the BIM visualization display of the temperature field and stress field data, the support of the corresponding sensor data historical curve through model node query; the establishment of the data quality scoring mechanism, the real-time scoring of the data from the four dimensions of integrity, accuracy, consistency and timeliness, and the triggering of the data repair process by low-score data.

[0011] Further, the risk prediction of the fused data by the multi-modal fusion analysis model includes the construction of a time-space process three-dimensional analysis model, the time-space alignment of the concrete pouring temperature, formwork strain and tension stress data of the same box girder, the identification of multi-parameter coupling abnormal patterns through tensor decomposition algorithm, and the detection of the mismatch phenomenon of strain and stress during tension at a certain time period and in a certain pedestal area; at the same time, the knowledge graph method is introduced to construct a prefabricated box girder production field knowledge graph containing 100+ entities and 500+ relationships for semantic reasoning of risk tracing.

[0012] Further, the risk prediction of the fused data by the multi-modal fusion analysis model also includes the use of LSTMAttention neural network to predict the concrete pouring defect probability risk of the subsequent process according to the reinforcement binding quality and formwork assembly precision data of the previous process data; at the same time, a double model of concrete strength growth is constructed and developed, the physical model of which is based on the Arrhenius equation to simulate the hydration reaction, and the data model uses GBDT algorithm to fit the historical strength data, and the prediction accuracy is improved through model fusion.

[0013] Further, the intelligent linkage response based on the predicted risk includes the introduction of information entropy theory to dynamically adjust the risk level weight according to the uncertainty degree of real-time data, wherein when the sensor data fluctuation exceeds 3 times the standard deviation, the entropy weight coefficient of the corresponding risk factor is automatically increased by 20%, making the system more sensitive to sudden abnormalities; at the same time, a risk evolution timeline is established to record the full-cycle data changes of each risk event from germination to occurrence, and the hidden Markov model HMM is used to learn the risk evolution pattern to predict the development trend of potential risks.

[0014] Further, the intelligent linkage response based on the predicted risk also includes the establishment of a warning control linkage rule library, when the curing shed temperature and humidity exceed the set threshold, the steam valve opening degree and spray pump power are automatically adjusted, and the standby ventilation equipment is started to form a closed-loop control; when the tensioning equipment tensioning force is detected to be abnormal, the tensioning operation is automatically suspended and the equipment operation permission is locked, and the implementation effect of the risk response measures is simulated in the multi-source data fusion engine, and after simulation verification, it is executed to the physical equipment to reduce the risk of misoperation.

[0015] In a second aspect, an intelligent risk early warning system for prefabricated box girder production in bridge engineering comprises: A data acquisition module configured to build a multi-dimensional data acquisition network for data acquisition; A preprocessing module configured to preprocess the acquired multi-dimensional data; A data fusion module configured to build a multi-source data fusion engine and establish a data quality scoring mechanism to realize data fusion; A prediction module configured to use a multi-modal fusion analysis model to predict risks of the fused data; A response module configured to intelligently respond based on the predicted risks.

[0016] In a third aspect, the present application provides a computer-readable storage medium having a plurality of instructions stored therein, the instructions being adapted to be loaded and executed by a processor of a terminal device to implement the method.

[0017] In a fourth aspect, the present application provides a terminal device comprising a processor and a computer-readable storage medium, the processor being configured to implement the instructions, and the computer-readable storage medium being configured to store a plurality of instructions, the instructions being adapted to be loaded and executed by the processor to implement the method.

[0018] In summary, the present application has the following beneficial technical effects: Through the distributed sensor network and the dynamic preprocessing algorithm, high-frequency collection of 8 types of core parameters such as temperature, strain and stress is realized, the data acquisition error is reduced by more than 60%, and the integrity is improved to more than 99%. Based on BIM+ big data technology, a multi-dimensional data fusion platform is built, the data barriers in the production link are broken, a full-chain digital mapping from raw materials to finished products is formed, and cross-scale and full-factor data support is provided for risk early warning. Relying on intelligent algorithms such as tensor decomposition and fusion model, the traditional experience judgment is upgraded to data-driven quantitative analysis, the accuracy of key risk identification is improved to more than 90%, and the early warning time of major risks is shortened by more than 50% compared with manual inspection. Through the dynamic risk matrix and the entropy weight adjustment mechanism, the risk level is dynamically calibrated in real time, and a closed-loop prevention and control system of "prediction-identification-early warning-disposal" is built. Through reinforcement learning scheduling and equipment linkage control, intelligent collaboration of production factors such as personnel, equipment and process is realized, the efficiency of key processes is improved by 20%-30%, and resource waste is reduced by more than 15%. Based on the data-driven decision support system, the production plan and resource allocation can be dynamically optimized, the prefabricated box girder production is transformed from "passive response" to "active optimization", and the overall engineering comprehensive cost is reduced by 8%-12%. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1is a schematic diagram of the method of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below with reference to the accompanying drawings.

[0021] Embodiment 1 Reference Figure 1 The bridge engineering prefabricated box girder production intelligent risk early warning method of the present embodiment comprises: I. Deepening construction of data acquisition and collection system (1) Multi-dimensional data collection network expansion 1. New monitoring data types Raw material traceability data: When raw materials such as cement and steel bars enter the site, batch information, manufacturer, and quality inspection report data are recorded through RFID tags, and blockchain technology is used for storage. The blockchain uses a consortium chain architecture, and each block contains the hash value of the previous block, a timestamp, and raw material data to ensure that the data cannot be tampered with. For example, when the steel bar batch information is uploaded, a hash value is generated using the SHA256 algorithm, which is packaged with the block header information to achieve full-chain traceability from raw materials to finished products.

[0022] Micro-strain monitoring data: Fiber Bragg grating strain sensors are implanted in key parts of the prefabricated box girder formwork and pedestal, with a sampling frequency of 10 Hz to monitor the micro-strain changes of the structure during concrete pouring and tensioning. The sensor uses wavelength encoding technology with a measurement accuracy of 1 με, which can capture 0.001% of the structural deformation.

[0023] Noise and dust data: Noise monitors (range 30130 dB, accuracy ±1.5 dB) and dust concentration sensors (measurement range 010 mg / m³, resolution 1 μg / m³) are installed to collect data every 5 minutes, and the environmental protection early warning system is linked to trigger an early warning when the dust concentration exceeds 800 μg / m³.

[0024] 2. Three-dimensional layout scheme of monitoring points Spatial layering: Unmanned aerial vehicle inspection platforms are deployed in the upper space, equipped with multispectral cameras (resolution 5 cm) and laser radars (point cloud density 10 points / m²), which scan the overall working conditions of the beam factory every week to obtain spatial data such as site settlement (accuracy ±2 mm) and equipment placement compliance; The ground layer uses a combination of fixed sensors and mobile inspection robots, with the robot equipped with a 16-line laser radar and a 4K camera, which patrols according to the preset path (interval 30 minutes) to identify abnormal equipment states in real time.

[0025] Time gradient collection: The concrete curing temperature data adopts a dynamic sampling strategy. In the initial setting period (06h), it is collected every 10 minutes; in the final setting period (624h), it is collected every 30 minutes; in the stable strength growth period (after 24h), it is collected every hour. Through time series change analysis, the curing effect is analyzed.

[0026] 3. Micro-strain data processing algorithm Wavelet transform denoising: db4 wavelet is used to decompose the original strain signal for 5 layers. The threshold value λ is calculated as follows: (σ is the noise standard deviation, and n is the data length). After soft threshold function processing, the signal-to-noise ratio of the reconstructed signal is improved by more than 15dB, effectively filtering out environmental vibration interference. The formula is as follows: wherein, is the wavelet coefficient after threshold processing, is the wavelet basis function, is the approximation coefficient.

[0027] 4. Temperature gradient collection algorithm Dynamic sampling frequency control: According to the temperature change rate, the sampling frequency is automatically switched. When the change rate is ≥5℃ / h, the sampling frequency is 6 times / hour; when 1℃ / h<change rate<5℃ / h, the sampling frequency is 2 times / hour; when the change rate is ≤1℃ / h, the sampling frequency is 1 time / hour.

[0028] II. Data transmission and integrated platform upgrade architecture (I) Hybrid network transmission system 1. Core layer network design Dual-redundant fiber ring network is built with industrial-grade ring network switches, with a bandwidth of 10Gbps, supporting ERPS (Ethernet Ring Protection Switch) protocol, and fault switching time <50ms. Each ring network node is equipped with an optical module (transmission distance 20km) to ensure high-speed and stable connection between the central server and the edge nodes.

[0029] Data encryption uses the national SM4 algorithm, with a block length of 128 bits and a key length of 128 bits, with an encryption speed of more than 1Gbps, meeting the security requirements of high-frequency data transmission.

[0030] 2. Access layer transmission strategy Mobile devices (gantry cranes, inspection robots) use 5G+Beidou dual-mode transmission: In 5G mode, NRD2D technology is used, with a peak rate of 1.2Gbps and a latency of <20ms; Beidou mode supports short message communication (120 Chinese characters each time) and automatically switches in 5G blind areas. The signal quality evaluation function is: Q=0.5RS+0.3RC+0.2RB Where RS is signal strength (-120 dBm ~ 0 dBm, normalized), RC is channel quality indicator (0 ~ 15, normalized), and RB is bit error rate (1e-6 ~ 1, normalized). When Q < 0.3, the Beidou switch is triggered.

[0031] 3. Edge node data compression Principal component analysis (PCA) is used to reduce the dimensionality of high-frequency vibration data, retaining 95% of the variance corresponding to the principal components, with a compression ratio of 1:10. The specific steps are: calculate the data covariance matrix → eigenvalue decomposition → select the eigenvectors with cumulative contribution rate ≥ 95% → projection dimensionality reduction. Formula: Z = W T (X - μ) Where X is the original data matrix (n x m), μ is the mean vector, and W is the eigenvector matrix (m x k, k is the dimensionality after dimensionality reduction).

[0032] (II) Data hub capability expansion 1. BIM-data fusion engine The prefabricated box girder BIM model (created by Revit, precision LOD400) is bound with real-time monitoring data, and the model and data interaction is realized through the IFC standard interface. The temperature field data is displayed in the form of a cloud chart on the BIM model, and the 30-day historical data curve of the corresponding sensor can be queried by clicking the model node, with an error of <2%.

[0033] Access to meteorological API data (precision to township level), get 72-hour temperature and humidity, wind power prediction, combined with beam production plan, through linear programming algorithm to optimize the use of health shed, energy consumption reduced by 15%.

[0034] 2. Data governance system Establish a data quality scoring mechanism, score from four dimensions of integrity (missing rate <5%), accuracy (error rate <3%), consistency (time stamp deviation <10 seconds), and timeliness (transmission delay <30 seconds). Scoring formula: Q = 0.3(1 - missing rate) + 0.4(1 - error rate) + 0.2(1 - time deviation rate) + 0.1(1 - delay rate) Low-score data (Q < 0.7) triggers automatic resampling process, repaired by adjacent sensor data interpolation (interpolation error <5%).

[0035] Three, data analysis and risk identification model (I) Multi-modal fusion analysis model 1. Spatio-temporal correlation tensor decomposition A three-dimensional analysis model of the process in time and space was constructed, and the concrete pouring temperature, formwork strain, and tension stress data of the same box girder were spatiotemporally aligned. The multi-parameter coupling anomaly pattern was identified by the tensor decomposition algorithm, and the strain-stress mismatch phenomenon in a specific platform area during tensioning was detected at a certain time period. At the same time, a knowledge graph method was introduced to construct a knowledge graph in the field of precast box girder production, containing 100+ types of entities and 500+ types of relationships, for semantic reasoning for risk tracing.

[0036] Among them, a three-dimensional tensor of "time-space-process" is constructed. (100 sampling points in the time dimension, 50 sensors in the spatial dimension, and 20 parameters in the process dimension), using Tucker decomposition to extract anomaly patterns: Among them, the core tensor The pattern matrices U, V, and W correspond to time, space, and process characteristics, respectively. Anomaly detection is achieved through reconstruction error: When e>0.15, it is identified as an anomaly, such as strain-stress mismatch in a certain area during tensioning, with an early warning accuracy rate of 92%.

[0037] 2. Concrete Strength Growth Fusion Model The physical model is based on the Arrhenius equation: The data model employs the GBDT algorithm, taking temperature data (past 72 hours) and time t as input, and outputting intensity predictions. Fusion model: The weights w are dynamically adjusted through Bayesian optimization, with the objective function being MSE. After optimization, the prediction error is less than 3%, improving accuracy by 40% compared to a single model.

[0038] (II) Dynamic Risk Assessment System 1. Dynamic adjustment of risk entropy value Calculate the information entropy of a certain risk factor: Entropy weight: When data fluctuation exceeds 3 standard deviations, the weighting adjustment factor is: When the standard deviation Sj of the gantry crane vibration data suddenly increases, its risk weight is automatically increased by 20% to 50%, making the system more sensitive to sudden anomalies.

[0039] 2. Three-dimensional risk matrix model Risk index calculation: Where P is the likelihood (0-1, based on historical data statistics), C is the severity of consequences (0-10, expert scoring), and T is the pre-warning time coefficient (T = 1 - t 预警 / t 临界 ).

[0040] Grade division: Red (RI ≥ 7): such as abnormal brake system of gantry crane (pre-warning time < 2h) Yellow (4 ≤ RI < 7): such as over-standard temperature and humidity in health pavilion (pre-warning time 224h) Blue (RI < 4): such as personnel not wearing safety helmet (pre-warning time > 24h).

[0041] Four, depth optimization of risk early warning and response mechanism (I) Intelligent linkage response system 1. Equipment linkage control PID control is adopted for temperature and humidity in health pavilion: Where e(t) is the temperature deviation (set value actual value), and the control accuracy is ±1℃. When the temperature exceeds the set value by 5℃, the standby ventilation equipment is automatically started, and the response time is <10 seconds.

[0042] When the tensioning equipment is abnormal, the system sends a stop command through the Modbus protocol and locks the operation permission at the same time, and the command transmission delay is <50ms.

[0043] 2. Personnel intelligent scheduling Scheduling model based on PPO algorithm: State space: S = {R, P, Sk, Ta} (risk level, personnel position coordinates, skill vector, remaining response time) Action space: A = {a1, a2, …, an} Reward function: R = 0.5rt + 0.3rs + 0.2rc Where rt is the time efficiency (10 for early completion, -10 for overtime), rs is the skill matching degree (5 for complete match), and rc is the cost (5 for low consumption).

[0044] Training parameters: learning rate 3e-4, discount factor 0.99, batch size 2048, scheduling efficiency improved by 30% after 500 rounds of training.

[0045] 3. AR remote assistance On-site personnel obtain expert guidance through AR glasses (resolution 1280x720), and experts can view real-time on-site images and equipment data (delay <200ms). The YOLOv8 algorithm is used to identify fault locations, with an accuracy of 90% and a labeling delay of <1 second.

[0046] (II) Closed-loop improvement mechanism 1. Early warning effect evaluation Establish three-dimensional evaluation indicators: Early warning accuracy = Correct early warning number / Total early warning number (target ≥ 90%) Early warning lead rate = Actual early warning time / Accident occurrence time (target ≤ 0.5) Response completion rate = On-time response number / Total response task number (target ≥ 95%) If the evaluation fails for three consecutive times (e.g., accuracy <80%), trigger the model to automatically reconfigure, optimize parameters using new data through transfer learning, and reconfigure in <4 hours.

[0047] 2. Iterative risk knowledge base Use the BERT model to automatically extract knowledge from accident reports, with an accuracy of 85%. The knowledge graph contains 1000+ entities, 5000+ relationships, and supports semantic reasoning (e.g., "insufficient tensile stress → may cause beam cracking"). Updated every quarter, with an annual knowledge coverage rate of 15%.

[0048] Embodiment 2 The embodiment provides a precast box girder production intelligent risk early warning system in bridge engineering, comprising: A data acquisition module configured to build a multi-dimensional data acquisition network for data acquisition; A preprocessing module configured to preprocess the collected multi-dimensional data; A data fusion module configured to build a multi-source data fusion engine and establish a data quality scoring mechanism to realize data fusion; A prediction module configured to use a multi-modal fusion analysis model to predict risks based on the fused data; A response module configured to intelligently respond based on the predicted risks.

[0049] A computer-readable storage medium having a plurality of instructions stored therein, the instructions being adapted to be loaded and executed by a processor of a terminal device to perform a precast box girder production intelligent risk early warning method in bridge engineering.

[0050] A terminal device comprises a processor and a computer readable storage medium, the processor is used for realizing instructions; the computer readable storage medium is used for storing a plurality of instructions, the instructions are suitable for being loaded by the processor and executing the intelligent risk early warning method for precast box girder production in bridge engineering.

[0051] The above are preferred embodiments of the present application, not limited by the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for intelligent risk early warning of precast box girder production in bridge engineering, characterized in that, include: Construct a multi-dimensional data acquisition network for data collection; Data preprocessing is performed on the collected multidimensional data; Build a multi-source data fusion engine and establish a data quality scoring mechanism to achieve data fusion; Risk prediction is performed on the fused data using a multimodal fusion analysis model; Intelligent and coordinated response based on predicted risks. 2.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 1, characterized in that, The construction of a multi-dimensional data acquisition network for data acquisition includes collecting raw material traceability data through blockchain based on a consortium blockchain architecture; and monitoring micro-strain changes of the structure in real time during concrete pouring and tensioning using grating strain sensors. In addition, noise monitoring instruments and dust concentration sensors are deployed to collect noise and dust data; multispectral cameras and lidar are deployed in spatial layers, and fixed sensors and mobile inspection robots are used on the ground layer to inspect equipment for abnormal conditions; and a dynamic sampling strategy is adopted to collect concrete curing temperature data regularly. 3.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 2, characterized in that, The data preprocessing of the collected multidimensional data includes a three-level preprocessing mechanism of sliding window filtering, outlier detection, and linear interpolation before temperature data sampling. Specifically, the sliding window mean filtering method is used to filter out high-frequency noise collected by the temperature sensor; outlier detection is achieved by identifying and marking temperature jump values ​​that exceed the normal range based on the 3σ principle; and the marked outliers are smoothed and repaired by linear interpolation to avoid affecting the calculation of the rate of change. 4.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 3, characterized in that, The construction of a multi-source data fusion engine and the establishment of a data quality scoring mechanism include integrating a lightweight BIM model engine to bind the precast box girder BIM model with real-time monitoring data in three dimensions, realizing BIM visualization of temperature field and stress field data, and supporting querying historical curves of corresponding sensor data through model nodes; establishing a data quality scoring mechanism to score data in real time from four dimensions: completeness, accuracy, consistency and timeliness, and triggering a data repair process for low-scoring data.

5. The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 4, characterized in that, The method of using a multimodal fusion analysis model to predict risks from fused data includes constructing a three-dimensional analysis model of the process in time and space, aligning the concrete pouring temperature, formwork strain, and tension stress data of the same box girder in time and space, identifying multi-parameter coupling anomaly patterns through tensor decomposition algorithm, and detecting strain-stress mismatch in a specific platform area during tensioning at a certain time period; at the same time, a knowledge graph method is introduced to construct a knowledge graph in the field of precast box girder production, containing 100+ types of entities and 500+ types of relationships, for semantic reasoning for risk tracing. 6.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 5, characterized in that, The method of using a multimodal fusion analysis model to predict risks from fused data also includes using an LSTMAttention neural network to predict the probability of concrete pouring defects in subsequent processes based on the rebar binding quality and formwork assembly accuracy data from previous processes. At the same time, a dual model for concrete strength growth is constructed and developed. The physical model is based on the Arrhenius equation to simulate hydration reaction, and the data model uses the GBDT algorithm to fit historical strength data. The prediction accuracy is improved through model fusion. 7.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 6, characterized in that, The intelligent linkage response based on the predicted risk includes introducing information entropy theory, dynamically adjusting risk level weight according to the uncertainty degree of real-time data, wherein when the sensor data fluctuation exceeds 3 times of the standard deviation, the entropy weight coefficient of the corresponding risk factor is automatically increased by 20%, so that the system is more sensitive to sudden abnormalities; meanwhile, a risk evolution time axis is established to record the whole cycle data change of each risk event from the bud to the occurrence, the risk evolution mode is learned through a hidden Markov model HMM, and the development trend of potential risks is predicted. 8.The intelligent risk early warning method for precast box girder production in bridge engineering according to claim 7, characterized in that, The intelligent linkage response based on the predicted risk also includes establishing a warning control linkage rule library, when the health care shed temperature and humidity exceed the set threshold, the steam valve opening degree and the spray pump power are automatically linked and adjusted, and the standby ventilation equipment is started to form a closed loop control; when the tensioning equipment tensioning force is abnormal, the tensioning operation is automatically suspended and the equipment operation permission is locked, and the implementation effect of the risk response measures is simulated in the multi-source data fusion engine, and after the simulation verification, the measures are issued to the physical equipment for execution to reduce the risk of misoperation.

9. An intelligent risk early warning system for precast box girder production in bridge engineering, characterized in that, It includes: A data acquisition module configured to build a multi-dimensional data acquisition network for data acquisition; A preprocessing module configured to preprocess the collected multi-dimensional data; A data fusion module configured to build a multi-source data fusion engine and establish a data quality scoring mechanism to realize data fusion; A prediction module configured to use a multi-modal fusion analysis model to predict risks based on the fused data; A response module configured to intelligently link and respond based on the predicted risks. 10.A terminal device, comprising a processor and a computer readable storage medium, the processor is configured to implement instructions; the computer readable storage medium is configured to store a plurality of instructions, characterized in that, The instructions are suitable for being loaded and executed by the processor to perform the method of claim 1.

Images