Construction scheme intelligent dynamic auditing method and system based on multi-model cooperation

Abstract

The invention discloses a construction scheme intelligent dynamic auditing method and system based on multi-model collaboration, and relates to the technical field of construction scheme auditing, and the method comprises the steps: obtaining a construction scheme, specifically comprising a construction scaffold scheme document and a matched drawing, extracting key parameters from the construction scheme document and the matched drawing, and carrying out the construction scheme auditing; the key parameters comprise the scaffold body type, the erection height, the number of machine positions, the material specification, the lifting step and the safety device configuration; constructing a scaffold construction specification knowledge graph, and forming a compliance verification rule base; collecting historical construction site monitoring data; providing an adjustment suggestion based on the real-time monitoring data; by automatically extracting key parameters in a construction scheme and establishing a standardized database, careless omission caused by manual extraction is effectively avoided, a scaffold construction specification knowledge graph is constructed to form a multi-dimensional compliance verification rule base, and comprehensive and accurate verification in multiple aspects is realized by combining and collecting historical construction site monitoring data.

Description

Technical Field

[0001] The present invention relates to the technical field of construction plan review, and more specifically, to a method and system for intelligent dynamic review of construction plans based on multi-model collaboration. Background Art

[0002] In the construction industry, scaffolding is a crucial support structure for aerial work. The compliance and safety of its construction plan are directly related to the safety of construction workers and the progress of the project. As construction projects become taller and more complex, scaffolding structures are becoming increasingly diverse, and the construction environment is becoming increasingly complex. Traditional construction plan review methods are no longer able to meet the dynamic and precise management needs. In the actual audit process, the data processing and association mechanisms are weak, and the extraction of key parameters such as the spacing between vertical poles and the position of wall supports is prone to omissions and lacks a standardized database. At the same time, there is a lack of effective linkage between the various parameters, monitoring data and audit models. For example, the impact of vertical pole spacing deviation and verticality deviation on load calculation is not included, resulting in the audit results being out of touch with reality; on the other hand, the audit system is rigid and lacks a closed loop. Compliance verification is only based on a simple comparison of fixed specification clauses, and there is a lack of real-time monitoring and dynamic audit mechanisms. The audit process is mostly one-way, and the results cannot be fed back to optimize the front-end links, making it difficult to adapt to the dynamic management needs of complex construction scenarios. Summary of the Invention

[0003] To solve the above problems, the present invention provides an intelligent dynamic review method for construction plans based on multi-model collaboration.

[0004] The present invention provides a method for intelligent dynamic review of construction plans based on multi-model collaboration, comprising: S1: Obtain a construction plan, specifically including a construction scaffolding plan document and supporting drawings, and extract key parameters from the construction plan document and supporting drawings, including scaffolding frame type, erection height, number of machine positions, material specifications, lifting steps, and safety device configuration; Identify the vertical pole spacing and wall support positions in the supporting drawings from key parameters and establish a parameter standardization database; S2: Build a knowledge graph of scaffolding construction specifications and form a compliance verification rule base; Based on the knowledge graph, the structural parameter compliance review is divided into structural parameter compliance, process logic rationality, safety device integrity and emergency measure effectiveness; S3: Collect historical construction site monitoring data, including frame verticality, bolt tightening torque, anti-fall device status, and load change data, and transmit it through IoT devices; S4: Obtain the construction site monitoring data collected by S3 and conduct multi-model collaborative audit based on the audit dimensions divided by S2; The rule engine model verifies the ratio of rack height to floor height and the spacing between machine bays based on the specification clauses in the knowledge graph; The machine learning model is trained on historical violation cases to identify potential risks in the improvement step; S5: Generate a dynamic audit report based on the audit results of S4 and make adjustment suggestions based on real-time monitoring data.

[0005] Preferably, in said S1, the vertical pole spacing and the wall support position in the supporting drawings are identified from the key parameters, and when the parameter standardization database is established, the vertical pole spacing and the wall support position are quantified: The deviation between the pole spacing and the plan requirements is quantified to provide basic data for the subsequent S2 structural parameter compliance review; The coordinate transformation of the wall supports is performed to convert the coordinate parameters in the drawing into physical coordinates that match the actual geographic information of the construction site.

[0006] Preferably, when constructing the compliance verification rule base in S2, quantitative processing is performed in combination with the key parameters extracted in S1 and the calculated vertical pole spacing deviation rate and wall support conversion coordinates, specifically including: First, calculate the individual compliance of the structural parameters, including the individual compliance of the pole spacing parameters. and support space distribution compliance , individual compliance of pole spacing parameters is calculated as follows: ; Calculation of support spatial distribution compliance based on transformed coordinates of wall-attached supports , calculated as follows: ; in is the distance between the jth support and the adjacent support, is the average spacing, p is the number of supports; Then calculate the structural parameter compliance score based on the individual compliance of the structural parameters , the obtained structural parameter compliance score Transmitted to S4 as the verification threshold of the rule engine model.

[0007] Preferably, the logical rationality review dimension of the process divided in S2 is combined with the converted wall support coordinates and structural parameter compliance scores Conduct an assessment, which includes the following steps: According to the formula Calculate the position deviation coefficient of the wall support ,in is the coordinate of the jth wall support after transformation, is the coordinate of the construction machine position corresponding to the j-th wall support; Then according to the formula , calculate the process logic rationality coefficient ; in is the theoretical interval time, is the position deviation coefficient of the wall support corresponding to the i-th process, is the preset allowable deviation value, is the actual total interval time, n is the number of process connections; Will Machine learning models transferred to S4.

[0008] Preferably, after the S3 collects the construction site monitoring data, it is also necessary to pre-process the on-site monitoring data. The specific steps are as follows: According to the formula , calculate the vth pre-processed construction site monitoring data ,in is the monitoring data of the vth construction site; Transmit all pre-processed construction site monitoring data to S4.

[0009] Preferably, when the rule engine model in S4 verifies the integrity of the safety device, analysis is performed in combination with the pole spacing deviation rate and the structural parameter compliance score, and the specific steps include the following: According to the formula , calculate and obtain the safety device demand coefficient ; Then according to the formula , calculate and obtain the safety device configuration integrity index ,in is the actual number of safety devices in S1, The required quantity specified for the construction site; Safety device configuration integrity index Transfer to S5.

[0010] Preferably, when the machine learning model in S4 identifies a potential risk in the lifting step, the process logic data flow is driven by the change of the coordinates of the wall-mounted support, specifically including: Extract the converted time series data of the wall-mounted support coordinates (xs, ys, zs) to generate a data stream of the motion trajectory of each support during the lifting process, and use this data stream for process logic analysis; After receiving the trajectory data stream, the overlap of adjacent support trajectories is calculated in real time; The machine learning model dynamically adjusts the extraction dimension of the lifting and lowering step feature data according to the deviation between the overlap data and the historical average overlap.

[0011] Preferably, the specific training and risk identification steps of the machine learning model in S4 include: S41: Construct a training sample set and divide the historical improvement step data into normal samples, low-risk samples, and high-risk samples according to risk levels. High-risk samples must include abnormal cases where the overlap of the wall support trajectory exceeds the threshold of 0.8, and low-risk samples correspond to edge cases with overlaps between 0.5 and 0.8. S42: A CNN-LSTM hybrid network structure is used, where the CNN layer is used to extract spatial features from the time series data of the coordinates (xs, ys, zs) of the wall supports, including the mutation characteristics of the 3D coordinate differences between adjacent supports, and the LSTM layer is used to capture the temporal sequence dependencies of the lifting steps. It should be noted that the trajectory trend changes of more than three consecutive steps are particularly important. S43: Process logic rationality coefficient As the attention weight, the logical rationality coefficient of the process greater than the first threshold The corresponding process data is given a weight attenuation of 0.3-0.5, and the logical rationality coefficient of the process below the first threshold is Strengthen the feature extraction strength; in this embodiment, the basic value of the first threshold is 0.6; During training, the sample imbalance is optimized, where the loss weight of high-risk samples is set to 5 times that of normal samples, and the loss weight of low-risk samples is set to 2 times that of normal samples; S44: During identification, when the risk probability value output by the model is greater than 0.7, the structural parameter compliance score is automatically associated. If the score is less than 80 points, a double risk warning is triggered, and a visual risk report containing trajectory abnormality segments and parameter deviation values ​​is generated.

[0012] Preferably, when generating the dynamic audit report in S5, a multi-data fusion loop feedback path is constructed, specifically including: Safety device configuration data, lifting step risk data, and load verification data are fed into the comprehensive analysis phase, and a three-dimensional data cube is created based on the data generation timestamp; The 3D data cube is sliced ​​and analyzed layer by layer. The data of each layer is used to generate adjustment suggestions. The preliminary adjustment parameters are output based on the distribution density of the slice data. The parameters are initially adjusted and input into the three-dimensional data cube in reverse to drive the cube reconstruction. The reconstructed three-dimensional data cube is sliced ​​and analyzed again, and the final audit report data is generated iteratively to form a data self-optimization cycle.

[0013] The present invention also proposes an intelligent dynamic review system for construction plans based on multi-model collaboration, including: Parameter association processing module: receives the quantization parameters of S1 and the rule base data of S2, performs the parameter association calculation, and outputs the associated verification threshold; Monitoring data feedback module: transmits the pre-processed data of S3 to each model of S4 in real time to achieve dynamic adjustment of model parameters; Model result fusion module: integrates the calculation results of each model in S4, performs cross-correction, and outputs the fused audit indicators; Reporting and optimization closed-loop module: Based on the S5 audit report and association rules, feedback instructions for front-end steps are generated to achieve closed-loop optimization of the entire process. Beneficial effects

[0014] By automatically extracting key parameters from the construction plan and establishing a standardized database, omissions in manual extraction are effectively avoided, laying an accurate data foundation for the audit work; a knowledge graph of scaffolding construction specifications is constructed to form a multi-dimensional compliance verification rule library, combined with historical construction site monitoring data, and using rule engines, machine learning, digital twins and other multiple models to collaborate for dynamic audits, achieving comprehensive and accurate verification of structural parameter compliance, process logic rationality and other aspects, and the data between each model is closely linked, which can timely identify potential risks and verify whether the construction load exceeds the limit; at the same time, by constructing a multi-data fusion loop feedback path, the audit results can flow back to the front-end link, forming a closed-loop optimization of the entire process, significantly improving the efficiency and accuracy of the construction plan review, effectively ensuring the safety and reliability of scaffolding construction, and better adapting to the dynamic management needs of high-rise and complex construction projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the generation method of the present invention. DETAILED DESCRIPTION

[0016] like Figure 1 Shown: A construction plan intelligent dynamic review method based on multi-model collaboration, including: S1: Obtain a construction plan, specifically including a construction scaffolding plan document and supporting drawings, and extract key parameters from the construction plan document and supporting drawings, including scaffolding frame type, erection height, number of machine positions, material specifications, lifting steps, and safety device configuration; It should be noted that the supporting drawings include but are not limited to: Scaffolding plan layout (marking the plane position and spacing of vertical poles, horizontal poles and sweeping poles); Scaffolding elevation drawing (marking the frame erection height, vertical spacing of wall supports, and extension positions of vertical poles); Detailed drawings of wall-mounted support nodes (marking the connection method between the support and the building structure, bolt type and installation dimensions); Layout of the frame lifting positions (marking the position number, location and corresponding relationship with the support); Identify the vertical pole spacing and wall support positions in the supporting drawings from key parameters and establish a parameter standardization database; S2: Build a knowledge graph of scaffolding construction specifications and form a compliance verification rule base; Based on the knowledge graph, the structural parameter compliance review is divided into structural parameter compliance, process logic rationality, safety device integrity and emergency measure effectiveness; S3: Collect historical construction site monitoring data, including frame verticality, bolt tightening torque, anti-fall device status, and load change data, and transmit it through IoT devices; S4: Obtain the construction site monitoring data collected by S3 and conduct multi-model collaborative audit based on the audit dimensions divided by S2; The rule engine model verifies the ratio of rack height to floor height and the spacing between machine bays based on the specification clauses in the knowledge graph; It should be noted that the rule engine first establishes a connection with the knowledge graph that stores the specification terms, and reads various specifications related to frame construction through a specific interface, including the limiting standards for the ratio of frame height to floor height, specific requirements for machine position spacing, etc., to ensure that the obtained specification terms are complete and up-to-date. Parse the regulatory clauses extracted from the knowledge graph and convert the rules described in natural language into machine-readable logical expressions. For example, convert "rack height must not exceed 3 times the floor height" into mathematical logic such as "rack height ≤ floor height × 3", while clarifying that the distance between camera positions must meet quantitative rules such as "minimum distance ≥ 2 meters"; Collect data such as the actual height of the frame, the height of the corresponding floor, and the actual spacing between each camera position in historical construction scenes. Data sources may include sensors at the construction site and manually entered measurement data to ensure data accuracy and timeliness; Substitute the actual data collected into the converted logical expression for verification. Check whether the ratio of the rack height to the floor height complies with the regulations, and check whether the distance between each camera position is within the specified range; A report is generated based on the verification results. If all data complies with the rules, a "verification passed" message is output. If there are problems such as the ratio of the rack height to the floor height exceeding the standard or the spacing between the machine positions being too small, the specific violations and the corresponding regulatory clauses are clearly pointed out so that construction personnel can make timely corrections. The machine learning model is trained on historical violation cases to identify potential risks in the lifting process. It should be noted that historical violation case data related to the frame lifting process is collected. This data should include various parameters of the lifting process, such as lifting speed, force conditions during lifting, environmental factors (wind speed, temperature, etc.), and corresponding violation results (such as frame tilt, component damage, etc.). At the same time, the data is cleaned to remove duplicate, erroneous, or incomplete records. Preprocess the cleaned historical data, including feature extraction and data standardization. Extract meaningful features from the raw data that can identify potential risks, such as the rate of change of lifting speed and force fluctuations in different time periods. Standardize feature data of different magnitudes to the same interval to avoid the impact of data magnitude differences on model training results. Select an appropriate machine learning algorithm (such as decision trees, random forests, neural networks, etc.) and train the model using pre-processed historical violation case data. During the training process, by continuously adjusting the model parameters, the model can learn the correlation between each parameter in the lifting step and the violation risk. For example, when the lifting speed exceeds a certain threshold and the wind speed is greater than level 5, the probability of tilt risk increases significantly; Use a portion of historical data that was not used in training as a validation set to verify the trained model and evaluate metrics such as the accuracy and recall of the model in identifying potential risks. If these metrics do not meet expectations, readjust the algorithm parameters or increase the feature dimension and train again until the model performance meets the requirements. The model was deployed in an actual construction scenario, collecting real-time parameters from historical frame lifting steps (such as historical lifting speeds, real-time forces, and environmental data) and feeding them into the trained machine learning model. The model analyzed these parameters based on learned patterns and identified potential risks during the lifting process, such as predicting that "at the historical lifting speed, there is a risk of frame sway within the next 10 seconds."

[0017] When the model identifies a potential risk, it immediately issues an early warning signal containing the risk type, possible time of occurrence, and corresponding risk level, reminding on-site managers to take appropriate preventive measures. It should be noted that the frame at the construction site is fully scanned and measured to obtain the structural parameters of the frame (such as material, size, connection method, etc.). Based on this data, a model that is completely consistent with the physical frame is constructed to ensure that the model is highly consistent with the physical body in terms of geometric shape, physical properties, etc. Based on the actual construction conditions, various working condition parameters are set, including the size and distribution of construction loads (such as the weight of materials placed in different locations), environmental loads (such as wind loads and earthquake loads), construction operation steps (such as lifting and moving), etc., covering common normal working conditions and extreme working conditions; The digital twin model is used to simulate the stress state of the frame under different working conditions. Through methods such as finite element analysis, the stress, strain, displacement and other data of each component of the frame under different loads are calculated to restore the stress conditions of the frame during actual construction. Compare the simulated load data of the frame with the load-bearing limit of the frame material and the safety threshold of the structural design to determine whether the construction load exceeds the bearing capacity of the frame under the historical simulation conditions. For example, if the calculated stress of a member exceeds the yield strength of its material, it means that the load under this condition exceeds the safety range; Analyze the simulation results and summarize the weak points and safety margins of the frame under different working conditions. Feedback the analysis results to the construction design personnel to provide a basis for optimizing the frame structure and adjusting the construction plan, thus avoiding safety accidents caused by load exceeding the limit during actual construction. S5: Generate a dynamic audit report based on the audit results of S4 and make adjustment suggestions based on real-time monitoring data.

[0018] As an optional embodiment: in said S1, the vertical pole spacing and the wall support position in the supporting drawings are identified from the key parameters, and when the parameter standardization database is established, the vertical pole spacing and the wall support position are quantified: The deviation between the pole spacing and the plan requirements is quantified to provide basic data for the subsequent S2 structural parameter compliance review; It should be noted that, in this embodiment, the specific steps for calculating the deviation rate of the vertical pole spacing are as follows: ;in, is the deviation rate of the pole spacing, is the actual spacing of the poles extracted from the drawing, The standard spacing specified in the construction plan document; The coordinate transformation of the wall supports is carried out to convert the coordinate parameters in the drawings into physical coordinates that match the actual geographic information of the construction site. It should be noted that this lays the foundation for the subsequent topological relationship analysis.

[0019] It should also be noted that the specific steps for coordinate conversion of wall supports are as follows: ;in is the three-dimensional coordinate of the support after transformation, The support coordinates extracted from the drawing, , and They are the scaling factors of the drawing in the x, y, and z directions respectively (calculated by identifying the coordinate axis scales marked on the drawing).

[0020] As an optional embodiment: when constructing the compliance verification rule base in S2, quantitative processing is performed in combination with the key parameters extracted in S1 and the calculated vertical pole spacing deviation rate and wall support conversion coordinates, specifically including: First, calculate the individual compliance of the structural parameters, including the individual compliance of the pole spacing parameters. and support space distribution compliance , individual compliance of pole spacing parameters is calculated as follows: ; Calculation of support spatial distribution compliance based on transformed coordinates of wall-attached supports , calculated as follows: ; in is the distance between the jth support and the adjacent support, is the average spacing, p is the number of supports; Then calculate the structural parameter compliance score based on the individual compliance of the structural parameters , the obtained structural parameter compliance score Transmitted to S4 as the verification threshold of the rule engine model.

[0021] It should be noted that the structural parameter compliance score The specific calculation steps are as follows: By calculating the compliance scores of the vertical pole spacing and the spatial distribution of supports, we focus on the core parameters and solve the problem of isolated parameter evaluation.

[0022] As an optional embodiment: the logical rationality review dimension of the process divided in S2 is combined with the converted wall support coordinates and structural parameter compliance scores Conduct an assessment, which includes the following steps: According to the formula Calculate the position deviation coefficient of the wall support ,in is the coordinate of the jth wall support after transformation, is the coordinate of the construction machine position corresponding to the j-th wall support. It should be noted that in actual climbing frame construction scenarios, each wall support must be matched with a specific machine position to ensure that the climbing frame can be stably attached to the building structure. Then according to the formula , calculate the process logic rationality coefficient ; in is the theoretical interval time, is the position deviation coefficient of the wall support corresponding to the i-th process, is the preset allowable deviation value, is the actual total interval time, and n is the number of process connections. It should be noted that the safety of scaffolding not only depends on the quality of a single process (such as whether the poles are vertical), but also on the rationality of the process connection (for example, if the poles are erected before the foundation is compacted, it will cause the frame to sink; if the scaffolding planks are laid before the sweeping poles are installed, the risk of the poles becoming unstable will increase). Existing technologies often evaluate the compliance of a single process separately (such as only checking the distance between the poles), but ignore the impact of "time intervals between process connections" and "deviations in structural parameters during connection" on overall safety.

[0023] This coefficient quantifies the "time rationality", "deviation controllability" and "structural compliance" of the connection, and then compares it with the "actual total interval time". The final value obtained can directly reflect the risk level of the process connection ( The closer it is to 1, the more logical it is and the lower the risk); Will Machine learning models transferred to S4.

[0024] As an optional embodiment: after the construction site monitoring data is collected in step S3, the on-site monitoring data needs to be pre-processed. The specific steps are as follows: According to the formula , calculate the vth pre-processed construction site monitoring data ,in is the monitoring data of the vth construction site; All pre-processed construction site monitoring data are transmitted to S4. By associating process logic with construction site monitoring data, all monitoring data can retain their own physical meaning while being deeply associated with process logic, ultimately providing more reliable input for S4 and risk assessment.

[0025] As an optional embodiment, when the rule engine model in S4 verifies the integrity of the safety device, analysis is performed in combination with the pole spacing deviation rate and the structural parameter compliance score. The specific steps include the following: According to the formula , calculate and obtain the safety device demand coefficient ; Then according to the formula , calculate and obtain the safety device configuration integrity index ,in is the actual number of safety devices in S1, The required quantity specified for the construction site; Safety device configuration integrity index Transfer to S5.

[0026] As an optional embodiment: when the machine learning model in S4 identifies potential risks in the lifting step, the process logic data flow is driven by the change of the wall support coordinates, specifically including: Extract the converted time series data of the wall-mounted support coordinates (xs, ys, zs) to generate a data stream of the motion trajectory of each support during the lifting process, and use this data stream for process logic analysis; After receiving the trajectory data stream, the overlap of adjacent support trajectories is calculated in real time. It should be noted that in this embodiment, the specific calculation steps are as follows: In 5 consecutive sampling periods, the three-dimensional coordinate sequences of adjacent supports are (x1, y1, z1) to (x5, y5, z5) and (x1', y1', z1') to (x5', y5', z5'); Calculate the Euclidean distance of the corresponding time points respectively ; The value of i ranges from 1 to 5, and the number n of statistical distance values ​​less than 1 / 10 of the theoretical spacing between adjacent supports is the overlap ratio R = (n / 5) × 100%; The machine learning model dynamically adjusts the extraction dimension of the lifting and lowering step feature data based on the deviation between the overlap data and the historical average overlap. It should be noted that the specific steps are as follows: Calculate the standard deviation σ of the historical overlap data. When historical overlap is less than the historical average overlap - σ, add feature dimensions. The number of added features is the integer of the product of the total number of historical feature dimensions and (historical average overlap - historical overlap) / historical average overlap. When historical overlap is greater than the historical average overlap + σ, reduce feature dimensions. The number of reduced features is the integer of the product of the total number of historical feature dimensions and (historical overlap - historical average overlap) / historical average overlap, thus forming a closed loop of risk identification data. Feature dimensions added include the rate of change of support trajectory curvature, the number of intersections between adjacent support trajectories, and the standard deviation of the deviation angle between the support trajectory and the theoretical trajectory. When reducing feature dimensions, prioritize removing the dimension with the lowest correlation with overlap. The correlation is determined by calculating the Pearson correlation coefficient between each dimension and the overlap data.

[0027] As an optional embodiment: the specific training and risk identification steps of the machine learning model in S4 include: S41: Construct a training sample set and divide the historical improvement step data into normal samples, low-risk samples, and high-risk samples according to risk levels. High-risk samples must include abnormal cases where the overlap of the wall support trajectory exceeds the threshold of 0.8, and low-risk samples correspond to edge cases with overlaps between 0.5 and 0.8. S42: A CNN-LSTM hybrid network structure is used, where the CNN layer is used to extract spatial features from the time series data of the coordinates (xs, ys, zs) of the wall supports, including the mutation characteristics of the 3D coordinate differences between adjacent supports, and the LSTM layer is used to capture the temporal sequence dependencies of the lifting steps. It should be noted that the trajectory trend changes of more than three consecutive steps are particularly important. S43: Process logic rationality coefficient As the attention weight, the logical rationality coefficient of the process greater than the first threshold The corresponding process data is given a weight attenuation of 0.3-0.5, and the logical rationality coefficient of the process below the first threshold is Strengthen the feature extraction strength; in this embodiment, the basic value of the first threshold is 0.6; During training, the sample imbalance is optimized, where the loss weight of high-risk samples is set to 5 times that of normal samples, and the loss weight of low-risk samples is set to 2 times that of normal samples; S44: During identification, when the risk probability value output by the model is greater than 0.7, the structural parameter compliance score is automatically associated. If the score is less than 80 points, a double risk warning is triggered, and a visual risk report containing trajectory abnormality segments and parameter deviation values ​​is generated.

[0028] As an optional embodiment: when generating the dynamic audit report in S5, a multi-data fusion loop feedback path is constructed, specifically including: Safety device configuration data, lifting step risk data, and load verification data are fed into the comprehensive analysis phase, and a three-dimensional data cube is created based on the data-generated timestamps. The timestamps are accurate to the second, and each dimension corresponds to a data type. The data in each dimension is stored in layers by time, with each layer containing all three types of data at the corresponding time point. The 3D data cube is sliced ​​and analyzed layer by layer. The data of each layer is used to generate adjustment suggestions. The initial adjustment parameters are output based on the distribution density of the slice data. The distribution density is calculated as the ratio ρ of the number of specific data types to the total number of slice data. The number of specific data types is the number of target data in a certain time slice; The total number of slice data is all types of data in a certain time slice; Calculate the mean μ and standard deviation σ of the historical slice data ρ of the same type. When the historical ρ>μ+σ, the adjustment parameter is (ρ-μ) / μ×benchmark adjustment value. When ρ<μ-σ, the adjustment parameter is (μ-ρ) / μ×the inverse value of the benchmark adjustment value. Initial adjustment parameters are fed back into the 3D data cube to drive reconstruction. The reconstructed 3D data cube is then sliced ​​and analyzed again, and the final audit report data is generated iteratively, forming a data self-optimization cycle. During reconstruction, the corresponding data layer is corrected according to the adjustment parameters. ρ is calculated again and the adjustment parameters are updated until the adjustment parameter change rate is less than 5% for three consecutive iterations, generating the final data. The final audit report data is written into the dynamic report, and the core parameters flow back to the parameter extraction link of S1 to optimize the initial parameter extraction strategy.

[0029] The present invention also proposes an intelligent dynamic review system for construction plans based on multi-model collaboration, including: Parameter association processing module: receives the quantization parameters of S1 and the rule base data of S2, performs the parameter association calculation, and outputs the associated verification threshold; Monitoring data feedback module: transmits the pre-processed data of S3 to each model of S4 in real time to achieve dynamic adjustment of model parameters; Model result fusion module: integrates the calculation results of each model in S4, performs cross-correction, and outputs the fused audit indicators; Reporting and optimization closed-loop module: Based on the S5 audit report and association rules, feedback instructions for front-end steps are generated to achieve closed-loop optimization of the entire process.

[0030] The above are only preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions that fall within the scope of protection of the present invention are within the scope of protection of the present invention. It should be pointed out that for ordinary technical personnel in this technical field, certain improvements and modifications that do not depart from the principles of the present invention should also be considered as the scope of protection of this template.

Claims

1. A construction plan intelligent dynamic review method based on multi-model collaboration, characterized by: include: S1: Obtain a construction plan, specifically including a construction scaffolding plan document and supporting drawings, and extract key parameters from the construction plan document and supporting drawings, including scaffolding frame type, erection height, number of machine positions, material specifications, lifting steps, and safety device configuration; Identify the vertical pole spacing and wall support positions in the supporting drawings from key parameters and establish a parameter standardization database; S2: Build a knowledge graph of scaffolding construction specifications and form a compliance verification rule base; Based on the knowledge graph, the structural parameter compliance review is divided into structural parameter compliance, process logic rationality, safety device integrity and emergency measure effectiveness; S3: Collect historical construction site monitoring data, including frame verticality, bolt tightening torque, anti-fall device status, and load change data, and transmit it through IoT devices; S4: Obtain the construction site monitoring data collected by S3 and conduct multi-model collaborative audit based on the audit dimensions divided by S2; The rule engine model verifies the ratio of rack height to floor height and the spacing between machine bays based on the specification clauses in the knowledge graph; The machine learning model is trained on historical violation cases to identify potential risks in the improvement step; S5: Generate a dynamic audit report based on the audit results of S4 and make adjustment suggestions based on real-time monitoring data.

2. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 1 is characterized in that: In S1, the vertical pole spacing and the wall support positions in the supporting drawings are identified from the key parameters. When the parameter standardization database is established, the vertical pole spacing and the wall support positions are quantified: The deviation between the pole spacing and the plan requirements is quantified to provide basic data for the subsequent S2 structural parameter compliance review; The coordinate transformation of the wall supports is performed to convert the coordinate parameters in the drawing into physical coordinates that match the actual geographic information of the construction site.

3. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 1 is characterized in that: When constructing the compliance verification rule base in S2, quantitative processing is performed in combination with the key parameters extracted in S1 and the vertical pole spacing deviation rate and the wall support conversion coordinates calculated in claim 2, specifically including: First, calculate the individual compliance of the structural parameters, including the individual compliance of the pole spacing parameters. and support space distribution compliance , individual compliance of pole spacing parameters is calculated as follows: ； Calculation of support spatial distribution compliance based on transformed coordinates of wall-attached supports , calculated as follows: ； in is the distance between the jth support and the adjacent support, is the average spacing, p is the number of supports; Then calculate the structural parameter compliance score based on the individual compliance of the structural parameters , the obtained structural parameter compliance score Transmitted to S4 as the verification threshold of the rule engine model.

4. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 1 is characterized in that: The logical rationality review dimension of the process divided in S2, combined with the converted wall support coordinates and structural parameter compliance scores Conduct an assessment, which includes the following steps: According to the formula Calculate the position deviation coefficient of the wall support ,in is the coordinate of the jth wall support after transformation, is the coordinate of the construction machine position corresponding to the j-th wall support; Then according to the formula , calculate the process logic rationality coefficient ; in is the theoretical interval time, is the position deviation coefficient of the wall support corresponding to the i-th process, is the preset allowable deviation value, is the actual total interval time, n is the number of process connections; Will Machine learning models transferred to S4.

5. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 3 is characterized in that: After S3 collects the construction site monitoring data, it is necessary to pre-process the on-site monitoring data. The specific steps are as follows: According to the formula , calculate the vth pre-processed construction site monitoring data ,in is the monitoring data of the vth construction site; Transmit all pre-processed construction site monitoring data to S4.

6. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 5 is characterized in that: When the rule engine model in S4 verifies the integrity of the safety device, it analyzes the pole spacing deviation rate and the structural parameter compliance score. The specific steps include the following: According to the formula , calculate and obtain the safety device demand coefficient ; Then according to the formula , calculate and obtain the safety device configuration integrity index ,in is the actual number of safety devices in S1, The required quantity specified for the construction site; Safety device configuration integrity index Transfer to S5.

7. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 4 is characterized by: When the machine learning model in S4 identifies potential risks in the lifting step, it drives the process logic data flow through the coordinate changes of the wall support, specifically including: Extract the converted time series data of the wall-mounted support coordinates (xs, ys, zs) to generate a data stream of the motion trajectory of each support during the lifting process, and use this data stream for process logic analysis; After receiving the trajectory data stream, the overlap of adjacent support trajectories is calculated in real time; The machine learning model dynamically adjusts the extraction dimension of the lifting and lowering step feature data according to the deviation between the overlap data and the historical average overlap.

8. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 7 is characterized in that: The specific training and risk identification steps of the machine learning model in S4 include: S41: Construct a training sample set and divide the historical improvement step data into normal samples, low-risk samples, and high-risk samples according to risk levels. High-risk samples must include abnormal cases where the overlap of the wall support trajectory exceeds the threshold of 0.8, and low-risk samples correspond to edge cases with overlaps between 0.5 and 0.

8. S42: A CNN-LSTM hybrid network structure is used, where the CNN layer is used to extract spatial features from the time series data of the wall support coordinates (xs, ys, zs), including the mutation characteristics of the 3D coordinate differences between adjacent supports, and the LSTM layer is used to capture the time series dependencies of the lifting steps; S43: Process logic rationality coefficient As the attention weight, the logical rationality coefficient of the process greater than the first threshold The corresponding process data is given a weight attenuation of 0.3-0.5, and the logical rationality coefficient of the combined process below the first threshold is Enhance feature extraction strength; S44: During identification, when the risk probability value output by the model is greater than 0.7, the structural parameter compliance score is automatically associated. If the score is less than 80 points, a double risk warning is triggered, and a visual risk report containing trajectory abnormality segments and parameter deviation values ​​is generated.

9. The method for intelligent dynamic review of construction plans based on multi-model collaboration according to claim 8 is characterized in that: When generating the dynamic audit report in S5, a multi-data fusion loop feedback path is constructed, specifically including: Safety device configuration data, lifting step risk data, and load verification data are fed into the comprehensive analysis phase, and a three-dimensional data cube is created based on the data generation timestamp; The 3D data cube is sliced ​​and analyzed layer by layer. The data of each layer is used to generate adjustment suggestions. The preliminary adjustment parameters are output based on the distribution density of the slice data. The parameters are initially adjusted and input into the three-dimensional data cube in reverse to drive cube reconstruction. The reconstructed three-dimensional data cube is sliced ​​and analyzed again to iteratively generate the final audit report data.

10. An intelligent dynamic review system for construction plans based on multi-model collaboration, characterized by: The method is implemented based on a method for intelligent dynamic review of construction plans based on multi-model collaboration as described in any one of claims 1 to 9, and is characterized by comprising: Parameter association processing module: receives the quantization parameters of S1 and the rule base data of S2, performs the parameter association calculation, and outputs the associated verification threshold; Monitoring data feedback module: transmits the pre-processed data of S3 to each model of S4 in real time to achieve dynamic adjustment of model parameters; Model result fusion module: integrates the calculation results of each model in S4, performs cross-correction, and outputs the fused audit indicators; Reporting and optimization closed-loop module: Based on the S5 audit report and association rules, feedback instructions for front-end steps are generated to achieve closed-loop optimization of the entire process.

Images