Structural perpendicularity self-adaptive control method and system based on autonomous advanced correction

Abstract

The invention relates to the technical field of data processing, discloses a structure perpendicularity self-adaptive control method and system based on autonomous advanced correction, and is applied to intelligent regulation and control of an operation platform, concrete pouring and sliding and lifting speed in slip form construction. A deviation vector is constructed by collecting structural attitude and environment data and carrying out dynamic confidence weighted fusion, and three-stage progressive training is carried out based on the vector to obtain an advanced correction strategy network; after the dominant offset mode is recognized, a strategy network is input, and a construction correction instruction sequence including an operation platform adjustment scheme, a concrete pouring position adjustment scheme and a slip-up speed adjustment scheme is output; and the lifting stroke difference of all lifting points of the sliding formwork operation platform is adjusted according to the instruction to generate platform inclination, the concrete pouring amount is increased on the opposite side of the deviation direction, the sliding and lifting speed is controlled to execute perpendicularity correction, and structural response is monitored in real time and dynamic adjustment is conducted. The precision of perpendicularity control in the slip-form construction stage and the adaptive capacity of the system to the complex environment are improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a structural verticality adaptive control method and system based on autonomous progressive correction. Background Technology

[0002] Verticality control during the construction of high-rise buildings is a core aspect of ensuring structural safety and construction quality. Current technologies primarily employ a combination of manual monitoring and preset control parameters for verticality adjustment. This involves installing tilt and displacement sensors at key structural locations to collect structural attitude data, and then using a controller with fixed parameters to drive a leveling system to compensate for detected verticality deviations. Some automated systems utilize pre-programmed control logic to synchronously or differentially adjust the adjustment mechanism according to preset control strategies, attempting to achieve real-time verticality correction during construction. Traditional slipform construction corrects verticality deviations by manually adjusting the stroke difference between lifting points, the concrete pouring position, and the slipform speed. Construction personnel rely on experience to judge the degree of deviation and formulate corrective measures.

[0003] However, existing technologies have significant shortcomings in verticality control during the construction phase: manual measurements are spaced long and affected by obstructed views and environmental interference at the construction site, making it difficult to capture dynamic offset changes caused by the gradual increase in building height and load during construction; traditional fixed-parameter control strategies cannot adapt to the complex and ever-changing working conditions during the construction phase, especially when facing nonlinear disturbances such as sudden changes in wind load, fluctuations in concrete strength, and abnormal equipment response, which can easily lead to insufficient or excessive correction; pre-programmed control systems design control logic based on ideal working conditions, failing to fully consider the reliability differences of multi-source sensor data under different environmental conditions, and the use of fixed-weight fusion methods leads to a decrease in data quality when environmental disturbances are severe, affecting the accuracy of deviation assessment; sudden environmental disturbances at the construction site, such as strong winds and mechanical vibrations, often cause sensor data jumps, and existing systems lack a dynamic evaluation mechanism for the reliability of multi-source sensor data, and do not have the ability to learn and evolve strategies autonomously based on actual compensation effects, thus failing to continuously optimize control performance throughout the entire construction process. Summary of the Invention

[0004] This application provides a structural verticality adaptive control method and system based on autonomous progressive correction, which solves the problems of traditional verticality control methods being unable to adapt to complex and variable construction conditions, lacking autonomous learning ability and closed-loop feedback optimization mechanism, and improves the accuracy of verticality control during the construction stage and the system's adaptability to complex environments.

[0005] Firstly, this application provides a structural perpendicularity adaptive control method based on autonomous progressive correction, the structural perpendicularity adaptive control method based on autonomous progressive correction comprising: Step S1: Collect structural attitude data and environmental data, dynamically weight the data based on the intensity of environmental excitation, fuse the multi-source data, and construct a fused verticality deviation vector; Step S2: Perform three-stage progressive training based on the fused verticality deviation vector to obtain the advanced correction policy network; Step S3: Identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme; Step S4: Adjust the construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor the structural response in real time and adjust dynamically; Step S5: Perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

[0006] Secondly, this application provides a structural perpendicularity adaptive control system based on autonomous progressive correction, the structural perpendicularity adaptive control system based on autonomous progressive correction includes: The acquisition module is used to collect structural attitude data and environmental data, dynamically weight the confidence based on the intensity of environmental excitation, fuse multi-source data, and construct a fused verticality deviation vector. The training module is used to perform three-stage progressive training based on the fused verticality deviation vector to obtain an advanced correction policy network; The input module is used to identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme. The correction module is used to adjust construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor the structural response in real time and adjust dynamically; The verification module is used to perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

[0007] Thirdly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to perform the above-described adaptive control method for structural verticality based on autonomous progressive correction.

[0008] The technical solution provided in this application intelligently weights multi-source sensor data from the construction site by establishing a dynamic confidence weighting mechanism. Fixed confidence weights are set for tilt angle data, medium confidence weights are set for settlement data, and dynamic confidence weights are set for vibration data according to the exponential decay law. This solves the problem of insufficient data reliability in traditional fixed-weight fusion methods under construction environment interference. The spatiotemporal correlation fusion technology filters out short-term high-frequency noise and long-term drift components through an extended Kalman filter to construct a fused verticality deviation vector containing offset and offset velocity, providing high-quality state information for real-time control decisions during the construction phase. The three-stage progressive training mechanism breaks through the static limitations of traditional fixed-parameter control strategies. By sequentially completing static working conditions, dynamic disturbance rejection, and robust compensation learning, the advanced correction strategy network acquires hierarchical decision-making capabilities from basic construction conditions to complex scenario responses. The basic control strategy network learns the standard correction mapping relationship under normal slip-up conditions, the dynamic disturbance rejection strategy network learns dynamic adjustment strategies when wind load changes suddenly based on the extended state space, and the robust compensation strategy network further learns fault-tolerant control methods when equipment partially fails. The working condition adaptive selection mechanism automatically selects the most suitable strategy sub-network based on the current equipment health and environmental excitation intensity, significantly enhancing the system's adaptability to changing working conditions during construction. The dominant offset mode identification technology determines the dominant offset direction of the structure by calculating the proportional relationship between the offsets in the x and y directions through modal decomposition analysis. The influence weight allocation technology calculates the influence weight based on the distance from each parameter control point to the structure center and its geometric relationship with the dominant offset direction, thus solving the blindness of the traditional equal weight allocation method. This allows construction adjustments to be made in a targeted and precise manner based on the actual offset characteristics of the structure. The parameter optimization processing technology decouples and corrects the interference between parameters by establishing a mutual influence matrix of construction parameters. Combined with the construction response delay model, the timing adjustment ensures that the actual effective time of each parameter is synchronized, avoiding the negative impact of parameter coupling effect and response delay difference on the correction effect.

[0009] Digital twin simulation verification mechanism provides an effective means to evaluate the execution effect of control commands at construction sites. By comparing the expected verticality response results with the actual verticality response data, simulation errors are calculated and the sources of deviation are analyzed. For three types of failure causes—environmental abrupt changes, abnormal equipment response, and structural characteristic deviations—the corresponding strategy sub-network parameters are updated, enabling the system to possess true autonomous learning capabilities. Successful cases accumulated in the training sample library are statistically analyzed to extract construction parameter adjustment experience data corresponding to different deviation levels. Experience data range constraints are applied to the strategy network output results to avoid generating unreasonable parameters. The generated correction construction guidance table clearly defines specific construction parameters for three modes—x-direction deviation, y-direction deviation, and torsional deviation—and three deviation levels—slight, moderate, and severe. Dynamic adjustment rules are integrated to form a system that includes initial correction parameters, monitoring... This comprehensive construction operation guidance, which assesses threshold values ​​and dynamically adjusts amplitudes, realizes a complete control process from data acquisition and fusion, intelligent decision-making and reasoning, precise instruction generation to closed-loop feedback optimization. It continuously optimizes strategy parameters based on the actual correction performance at the construction site, avoiding the drawbacks of traditional preset control methods where parameters are fixed and cannot adapt to dynamic changes in construction. This achieves continuous improvement in verticality control performance throughout the construction process. In vertical structure construction fields such as slipform construction, the three-stage progressive training algorithm, dominant modality recognition algorithm, and parameter decoupling optimization algorithm of this application are applied. This fully considers the contribution of algorithm characteristics to the scheme, enabling the control strategy to adaptively select based on the complexity of construction conditions, accurately allocate based on structural offset characteristics, and autonomously learn and improve based on execution results. This provides an intelligent and adaptive technical solution for verticality control during the construction phase. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an embodiment of the adaptive control method for structural verticality based on autonomous progressive correction in this application. Figure 2 This is a schematic diagram of the three-stage progressive training process in the embodiments of this application; Figure 3 This is a schematic diagram of an embodiment of the adaptive control system for structural verticality based on autonomous progressive correction in this application. Detailed Implementation

[0012] This application provides a structural verticality adaptive control method and system based on autonomous progressive correction. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the adaptive control method for structural verticality based on autonomous progressive correction in this application includes: Step S1: Collect structural attitude data and environmental data, dynamically weight the data based on the intensity of environmental excitation, fuse the multi-source data, and construct a fused verticality deviation vector; Step S2: Perform three-stage progressive training based on the fused verticality deviation vector to obtain the advanced correction policy network; Step S3: Identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme; Step S4: Adjust construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor structural response in real time and adjust dynamically; Step S5: Perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

[0014] Specifically, structural attitude data and environmental data are collected and fused to construct a fused verticality deviation vector. Inclination sensors monitor the cylinder wall to obtain x- and y-direction inclination data. Displacement sensors measure foundation settlement data. Vibration sensors collect triaxial vibration acceleration data and calculate the acceleration standard deviation as the environmental excitation intensity. Based on the environmental excitation intensity, the data from each sensor are weighted according to confidence levels. A fixed weight of 0.8 is assigned to inclination data, a moderate weight of 0.5 to settlement data, and a dynamic weight based on exponential decay for vibration data, resulting in a sensor dataset with confidence level labels. This dataset is then input into an extended Kalman filter for state estimation. The filter calculates the prior state estimate through a prediction step and calculates and corrects the Kalman gain through an update step to obtain the posterior state estimate. A coordinate transformation is performed on the posterior state estimate. The inclination angle is multiplied by the cylinder height to obtain the offset, and the rate of change of inclination angle is multiplied by the height to obtain the offset velocity, thus obtaining the fused verticality deviation vector.

[0015] An advanced correction strategy network is obtained through a three-stage progressive training based on the fused verticality deviation vector. The first stage uses historical data under normal slip zone conditions for static training, employing historical verticality deviation as input and successfully corrected construction parameters as labels to train the neural network and establish a basic control strategy network. The second stage expands the training conditions by incorporating historical data on wind load and offset velocity changes, training the network to dynamically adjust correction parameters under wind load disturbances, resulting in a dynamic disturbance rejection strategy network. The third stage further expands the conditions by adding data on abnormal equipment response and partial equipment failure, training the network to compensate through other equipment when equipment malfunctions, resulting in a robust compensation strategy network. Finally, the three strategy networks are integrated, and a selection mechanism is established based on the complexity of the current construction conditions: the basic control network is selected when there is no wind load and the equipment is normal; the dynamic disturbance rejection network is selected when the wind speed standard deviation exceeds a threshold; and the robust compensation network is selected when an equipment anomaly is detected, resulting in the advanced correction strategy network.

[0016] The dominant offset mode of the fused verticality deviation vector is identified and a construction correction command sequence is output. Modal decomposition is performed on the fused verticality deviation vector, calculating the ratio of the x-direction offset to the y-direction offset. When the x-direction offset is significantly greater than the y-direction offset, the dominant offset direction is determined to be x; when the y-direction offset is significantly greater than the x-direction offset, the dominant offset direction is determined to be y; when the two are roughly equal, it is determined to be the torsional mode. The dominant offset mode identification result is obtained. Then, a construction parameter position matrix is ​​established to record the position coordinates of each area of ​​the operating platform, concrete pouring points, and slip control points relative to the structural center. The distance from each parameter control point to the structural center is calculated and normalized to obtain the distance weight. The position vector of each parameter control point and the dominant offset direction vector are calculated. The included angle is cosine transformed to obtain the direction influence factor. The distance weight is multiplied by the direction influence factor and normalized to obtain the influence weight allocation of each construction parameter. The verticality deviation vector and influence weight allocation are input into the advanced correction strategy network. The corresponding strategy sub-network is selected for inference based on the complexity of the current working condition. The network outputs the theoretical adjustment amount of each construction parameter. The theoretical adjustment amount is allocated according to the influence weight to obtain the preliminary construction correction scheme. The parameters of the preliminary scheme are optimized. Decoupling correction is performed according to the mutual influence relationship between each construction parameter to eliminate interference between parameters. Timing adjustment is performed in combination with the construction response delay characteristics to obtain the construction correction instruction sequence including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slipform speed adjustment scheme.

[0017] The construction parameters are adjusted according to the construction correction instruction sequence to perform verticality correction and dynamic adjustment. The construction correction instruction sequence is converted into construction operation guidance. The lifting stroke of each lifting point is adjusted according to the operation platform adjustment plan, so that the lifting stroke of the lifting point on one side of the structural offset direction is less than that of the lifting point on the opposite side, causing the operation platform to tilt. According to the concrete pouring position adjustment plan, the amount of concrete poured is increased on the opposite side of the structural offset direction and decreased on the offset side. According to the slipform speed adjustment plan, the overall slipform speed is adjusted to control the concrete demolding strength to reach the set value before slipforming. During the execution of construction adjustment, structural tilt sensor data is continuously collected to calculate the amount and rate of reduction of the current verticality deviation relative to the reference deviation before correction, so as to obtain correction effect monitoring data. Based on the monitoring data, the correction effect is judged. When the deviation decreases in the expected direction and the rate of reduction is within a reasonable range, the current construction parameters are maintained. When the rate of reduction of the deviation exceeds the set upper limit, the construction adjustment amplitude is reduced to avoid over-correction. When the deviation does not decrease as expected or increases in the opposite direction, the current construction adjustment is paused and a new construction correction instruction sequence is generated. The instruction parameters, execution process data and correction effect data of this construction adjustment are linked and recorded to obtain construction correction case data.

[0018] The corrected verticality response results are verified through simulation, and the strategy network is updated to generate the target control strategy. The construction correction command sequence and the corrected actual verticality response data are input into the digital twin model. The digital twin model calculates the expected verticality response results based on structural geometric parameters, material properties, and construction parameters. The expected results are compared with the actual response data to calculate the deviation and obtain simulation error data. Based on the simulation error data, deviation source analysis is performed. When the simulation error is less than a preset threshold, construction correction case data are marked as successful cases and added to the training sample library. When the simulation error exceeds the threshold, environmental data and equipment response data during construction execution are analyzed. The deviation source type is determined based on sudden changes in environmental excitation intensity, abnormal equipment response, and deviations in structural response characteristics. The advanced correction strategy network is updated accordingly based on the deviation source type. When the deviation source is environmental... When the environment changes abruptly, the dynamic disturbance rejection strategy network parameters are updated. When the deviation source is abnormal equipment response, the robust compensation strategy network parameters are updated. When the deviation source is structural characteristic deviation, the basic control strategy network parameters are updated. Statistical analysis is performed on successful cases in the training sample library to extract the operating platform tilt angle range, concrete pouring eccentricity range, and slip-lift speed adjustment range corresponding to different degrees of verticality deviation, obtaining empirical data for construction parameter adjustment. The empirical data is compared and verified with the updated strategy network output. When the strategy network output is within the empirical range, it is confirmed to be reasonable. When it exceeds the range, correction constraints are applied. Based on the verification results, a correction construction guidance table is generated for different offset modes and deviation degrees, clarifying the specific construction parameters under various offset conditions. The construction guidance table is combined with the dynamic adjustment rules to form a complete construction operation guidance document, obtaining the target control strategy.

[0019] In one embodiment, step S1 includes: The structure under construction is monitored by tilt sensors, displacement sensors and vibration sensors to obtain structural tilt data, foundation settlement data and vibration acceleration data, and the environmental excitation intensity is calculated based on the vibration acceleration data. The confidence weights of the sensor data are calculated based on the intensity of environmental excitation. Fixed confidence weights are set for tilt angle data, medium confidence weights are set for settlement data, and dynamic confidence weights are set for vibration data according to the exponential decay law, resulting in a sensor dataset with confidence labels. The sensor dataset with confidence level labels is input into the extended Kalman filter for state estimation, and the filtered state estimation result is obtained. The filtered state estimation results are processed by coordinate transformation, converting the tilt angle into an offset and the tilt angle change rate into an offset velocity, to obtain the fused verticality deviation vector.

[0020] Specifically, tilt sensors, displacement sensors, and vibration sensors are used to monitor the structure during construction, acquire raw data, and calculate the environmental excitation intensity. The tilt sensor is installed at a key height section on the cylinder wall to calculate the deflection angle of the structure relative to the vertical by detecting the components of gravitational acceleration in the x and y axes. The displacement sensor is set up between the operating platform and the bottom reference plane to measure the verticality deviation of the platform relative to the reference plane using the laser ranging principle. The vibration sensor is fixed on the surface of the cylinder structure to collect vibration acceleration data of the structure in the x, y, and z axes. The collected triaxial vibration acceleration data are grouped according to time windows, and the standard deviation of the acceleration values ​​within the most recent 60 seconds is calculated as the environmental excitation intensity parameter. The larger the standard deviation value, the more severe the wind load or construction vibration on the structure and the higher the environmental excitation intensity.

[0021] Based on the environmental excitation intensity, a confidence-weighted calculation is performed on the data from each sensor to obtain a sensor dataset with confidence labels. This step sets differentiated confidence weights for the measurement reliability differences of different sensors under different environmental conditions. A fixed confidence weight of 0.8 is assigned to the structural tilt angle data collected by the tilt sensor, as tilt angle measurement, based on the principle of gravitational acceleration, is relatively less affected by environmental disturbances and the measurement results are relatively stable and reliable. A medium confidence weight of 0.5 is assigned to the foundation settlement data collected by the displacement sensor, as settlement measurement is affected by the properties of the foundation soil and the distribution of construction loads, resulting in a moderate degree of uncertainty. The vibration acceleration data collected by the dynamic sensors are assigned dynamic confidence weights according to an exponential decay law. When the standard deviation of the environmental excitation intensity is small, it indicates that the construction environment is stable and the vibration data is less affected by noise, resulting in a higher confidence weight. When the standard deviation of the environmental excitation intensity increases, it indicates that there is strong wind or severe vibration and the vibration data is greatly affected by instantaneous disturbances, resulting in a rapid decay of the confidence weights according to an exponential function. The raw data collected by each sensor are associated with the corresponding confidence weight values ​​and stored to form a sensor dataset with confidence labels. Each record in this dataset contains four fields: collection timestamp, sensor number, measurement value, and confidence weight value.

[0022] The sensor dataset with confidence level labels is input into an extended Kalman filter (EPF) for state estimation, yielding filtered state estimates. The EPF is a recursive optimal state estimation algorithm that filters noisy observation data by establishing a state-space model of the system. The state-space model uses the structural tilt angle and the rate of change of tilt angle as state variables, establishing a state transition equation to describe the evolution of the state over time. Simultaneously, it establishes an observation equation to describe the mapping relationship between sensor measurements and the true state. The filter's prediction step calculates the prior state estimate and prior error covariance matrix based on the posterior state estimate from the previous time step and the state transition equation. The prior state estimate represents the prediction of the current state based on the system dynamics model, and the prior error covariance matrix quantifies the uncertainty of the prediction. The update step first calculates the Kalman gain, which is equal to the product of the prior error covariance matrix and the Kalman gain. The Kalman gain is calculated by dividing the transpose of the observation matrix by the sum of the observation matrix multiplied by the prior error covariance matrix multiplied by the transpose of the observation matrix and the observation noise covariance matrix. The magnitude of the Kalman gain determines the degree of confidence in the actual sensor measurement. When the prior error covariance is large, i.e., the prediction uncertainty is high, the Kalman gain is larger and the measurement value is more trusted. When the observation noise covariance is large, i.e., the measurement noise is large, the Kalman gain is smaller and the prediction value is more trusted. After calculating the Kalman gain, it is multiplied by the measurement residual to obtain the state correction. The measurement residual is equal to the actual sensor measurement value minus the observation matrix multiplied by the prior state estimate. The prior state estimate is added to the state correction to obtain the posterior state estimate. The posterior state estimate integrates the prediction information of the system dynamics model and the measurement information of the sensor. At the same time, the posterior error covariance matrix is ​​updated to quantify the uncertainty of the posterior estimate, resulting in a filtered state estimate that includes the filtered structure tilt angle and the rate of change of tilt angle.

[0023] The filtered state estimation results are processed by coordinate transformation to obtain a fused verticality deviation vector. The tilt angle data in the filtered state estimation results are in degrees or radians, representing the angular deviation of the structure relative to the vertical line. To intuitively reflect the actual horizontal offset distance and offset velocity of the top of the structure, coordinate transformation is required. The filtered x-direction tilt angle value is multiplied by the current vertical height of the cylinder from the foundation to the construction surface to obtain the horizontal offset of the top of the structure in the x-direction. The filtered y-direction tilt angle value is multiplied by the current vertical height of the cylinder to obtain the horizontal offset of the top of the structure in the y-direction. The filtered x-direction tilt angle change rate value is multiplied by the current vertical height of the cylinder to obtain the x-direction offset velocity. The filtered y-direction tilt angle change rate value is multiplied by the current vertical height of the cylinder to obtain the y-direction offset velocity. These are combined to form a fused verticality deviation vector containing four components: x-direction offset, y-direction offset, x-direction offset velocity, and y-direction offset velocity.

[0024] In one embodiment, step S2 includes: Static working condition training is performed based on the fused verticality deviation vector, and a foundation control strategy network is established using historical construction data under normal slip zone conditions. Based on the basic control strategy network, the training conditions are extended and historical data of wind load change and offset velocity change are added to carry out dynamic disturbance rejection training, so as to obtain the dynamic disturbance rejection strategy network. Based on the extended training conditions of the dynamic anti-disturbance strategy network, historical data of abnormal device response and partial device failure are added for robust compensation training to obtain the robust compensation strategy network. The basic control strategy network, dynamic disturbance rejection strategy network, and robust compensation strategy network are integrated, and a strategy selection mechanism is established based on the complexity of construction conditions to obtain an advanced correction strategy network.

[0025] Specifically, a basic control strategy network is established based on static condition training using a fused verticality deviation vector. Static condition cases with stable wind speed, normal equipment, and uniform concrete strength are selected from historical construction databases. The fused verticality deviation vector in each case is used as the input feature, and the corresponding adjustment amount of the operating platform, the concrete pouring eccentricity, and the slipform speed adjustment amount are used as the output labels. A supervised learning training sample set is constructed, and a deep neural network architecture is used to establish the basic control strategy network. The input layer receives the four components of the fused verticality deviation vector, the hidden layer extracts the mapping relationship between verticality deviation and construction parameters through a nonlinear activation function, and the output layer generates three construction adjustment parameters. During the training process, the backpropagation algorithm is used to update the weight parameters according to the error gradient between the network output and the real labels. After multiple iterations, the network learns to output reasonable construction adjustment parameters based on the verticality deviation under static conditions.

[0026] The dynamic disturbance rejection strategy network is obtained by expanding the training conditions of the basic control strategy network. Dynamic disturbance cases containing sudden changes in wind load or drastic changes in offset velocity are selected from the historical database. The input dimensions of the basic network are expanded by adding features such as current wind speed, wind speed change rate, and historical offset velocity change trend to the original four components. The weights of the basic network are kept as the initialization, and the weights are randomly initialized on the newly added connections corresponding to the expanded input. The network continues to train using historical cases containing dynamic disturbances. The network learns how to dynamically adjust construction parameters to resist the influence of disturbances when it detects a sudden increase in wind load or an abnormally fast offset velocity. For example, when the wind speed suddenly increases, the network outputs a larger platform tilt angle and pouring eccentricity to prevent further offset caused by wind load.

[0027] Robust compensation strategy network is obtained by expanding the training conditions of the dynamic disturbance rejection strategy network. Since there are few cases of complete failure of actual equipment, training data of abnormal equipment conditions are generated by a digital twin simulation system. In the simulation, the response delay or failure scenario of the lift point is set to simulate the change of structural verticality and the required compensation measures under abnormal conditions. The input dimension of the dynamic disturbance rejection network is further expanded to include equipment status characteristics such as the health score of each lift point, response delay time and historical failure frequency. The network continues to be trained to learn how to compensate for the effect of the failed lift point by increasing the adjustment of other normal lift points when the health of a certain lift point is detected to decline. At the same time, a safety constraint penalty term is introduced into the reward function. When the network output causes a certain lift point to exceed the safety range, a negative reward is given.

[0028] The three policy networks are integrated and a policy selection mechanism is established to obtain an advanced correction policy network. The three networks of basic control, dynamic disturbance rejection, and robust compensation are saved as independent sub-network models. A working condition complexity assessment module is established to calculate the construction working condition complexity in real time based on the current sensor data. The assessment logic first checks the equipment health index. If any improvement point exists where the health is below the threshold or the response delay exceeds twice the normal value, the robust compensation network is selected. If the equipment is normal, the environmental excitation intensity is checked. If the wind speed standard deviation or vibration standard deviation exceeds the threshold, the dynamic disturbance rejection network is selected. If the equipment is normal and the environment is stable, the basic control network is selected. Each time the working condition assessment module receives a new fused verticality deviation vector, it performs a working condition judgment. Based on the judgment result, the deviation vector is routed to the corresponding sub-network for inference and output of the construction correction instruction sequence.

[0029] Figure 2 This is a schematic diagram of the three-stage progressive training process in an embodiment of this application. Figure 2 As shown, the loss curves of the three-stage progressive training clearly demonstrate the progressive construction process of the policy network's capabilities. In the 0-40 training rounds, the basic control network converges rapidly under normal slip zone conditions, with the training loss quickly decreasing from 2.5 to 0.15, establishing a basic mapping relationship between verticality deviation and construction parameters. In the 40-100 training rounds, the dynamic disturbance rejection network learns coping strategies for sudden wind load changes and offset velocity variations based on the expanded state space, with the training loss gradually decreasing from 2.5 to 0.22. In the 100-200 training rounds, the robust compensation network learns fault-tolerant control methods through abnormal equipment operating condition samples, with the training loss decreasing from 2.3 to 0.45. The step-like descent characteristics of the three curves verify the effectiveness of the progressive training mechanism, indicating that the system gradually acquires complete capabilities from basic control to handling complex scenarios through hierarchical learning.

[0030] In one embodiment, step S3 includes: Modal decomposition analysis is performed on the fused verticality deviation vector to calculate the ratio between the offset in the x-direction and the offset in the y-direction. The dominant offset direction is determined based on the offset ratio to obtain the dominant offset mode recognition result. The influence weight of each construction parameter on the correction is calculated based on the dominant offset mode recognition results. The influence weight allocation of each construction parameter is obtained according to the geometric relationship between the operating platform position, the concrete pouring area, the slip control point and the dominant offset direction. The verticality deviation vector and the influence weight allocation are input into the advanced correction strategy network for reasoning and decision-making. The corresponding strategy sub-network is selected according to the complexity of the current construction conditions to obtain the preliminary construction correction scheme. The preliminary construction correction plan is optimized by parameters, decoupled and corrected according to the mutual influence between various construction parameters, and the timing is adjusted in combination with the construction response delay characteristics to obtain the construction correction instruction sequence.

[0031] Specifically, modal decomposition analysis is performed on the fused verticality deviation vector to determine the dominant offset direction and obtain the dominant offset mode identification result. Two components, x-direction offset and y-direction offset, are extracted from the fused verticality deviation vector. The absolute values ​​of the x-direction offset and y-direction offset are calculated, and their ratio is calculated. When the absolute value of the x-direction offset is significantly greater than the absolute value of the y-direction offset, the ratio is greater than 3, indicating that the dominant offset direction is x-direction, and the structure mainly tilts unidirectionally along the x-direction. When the absolute value of the y-direction offset is significantly greater than the absolute value of the x-direction offset, the ratio is less than 0.33, indicating that the dominant offset direction is y-direction, and the structure mainly tilts unidirectionally along the y-direction. When the absolute values ​​of the x-direction offset and y-direction offset are comparable, the ratio is between 0.5 and 2, indicating that the structure is offset in both directions simultaneously.

[0032] Based on the dominant offset mode recognition results, the influence weights of each construction parameter on the correction are calculated to obtain the influence weight allocation. A construction parameter position matrix is ​​established to record the position coordinates of each lifting point area, concrete pouring point, and slip control point of the operation platform relative to the geometric center of the structure. A plane coordinate system is established with the geometric center of the structure as the origin. The distance from each parameter control point to the structural center is calculated. The greater the distance, the greater the adjustment moment of the parameter control point on the structural attitude. The dominant offset direction vector is determined according to the dominant offset mode recognition results. When the dominant offset direction is x-axis, the dominant offset direction vector is a unit vector along the positive x-axis. The angle between the position vector of each parameter control point and the dominant offset direction vector is calculated. The direction influence factor is obtained by performing a cosine transformation on the angle. When the included angle is 0 degrees, the cosine value is +1, indicating that the control point of this parameter is located directly in front of the offset direction and contributes the most to the correction. When the included angle is 90 degrees, the cosine value is 0, indicating that the control point of this parameter is located on the perpendicular side of the offset direction and contributes less to the correction. When the included angle is 180 degrees, the cosine value is -1, indicating that the control point of this parameter is located on the opposite side of the offset direction. The distance parameter is normalized so that the weight corresponding to the maximum distance is 1 to obtain the distance weight. The direction influence factor and the distance weight are multiplied to obtain the preliminary influence weight of each construction parameter. The preliminary influence weight is normalized so that the sum of the weights of all construction parameters is one to obtain the influence weight allocation of each construction parameter.

[0033] The fused verticality deviation vector and influence weight allocation are input into the advanced correction strategy network for inference and decision-making to obtain a preliminary construction correction scheme. After receiving the fused verticality deviation vector, the advanced correction strategy network first judges the current construction condition through the working condition complexity assessment module, reads the equipment health index and environmental excitation intensity parameters, and selects the basic control, dynamic disturbance rejection or robust compensation strategy sub-network according to the judgment rules. The selected strategy sub-network performs forward propagation calculation on the input fused verticality deviation vector. After nonlinear transformation through the input layer and multiple hidden layers, the theoretical adjustment amount of each lifting point of the operating platform, the theoretical eccentricity of each area of ​​concrete pouring, and the theoretical velocity adjustment amount of each slip control point are generated in the output layer. These theoretical adjustment amounts are general correction parameters learned by the network based on historical training samples. The theoretical adjustment amounts are combined with the influence weight allocation. For parameter control points with larger influence weights, larger adjustment tasks are assigned; for parameter control points with smaller influence weights, smaller adjustment tasks are assigned. The preliminary construction correction scheme is obtained by multiplying the theoretical adjustment amount by the corresponding influence weight.

[0034] The preliminary construction correction plan is optimized to obtain a sequence of construction correction instructions. A construction parameter mutual influence matrix is ​​established to describe the coupling relationship between various construction parameters. Each element of this matrix represents the coupling influence of an action at a certain parameter control point on other parameter control points. For example, increasing the stroke at a certain lifting point will cause a change in the bearing capacity of adjacent lifting points; increasing the pouring volume in a certain area will change the overall center of gravity distribution, affecting the effective function of other areas. Based on the mutual influence matrix, the preliminary construction correction plan is decoupled and corrected. For each parameter control point, its initial adjustment is subtracted from the coupling influence of all other parameter control points. The coupling influence is equal to the sum of the sum of the initial adjustment amounts of other parameter control points multiplied by the corresponding elements of the mutual influence matrix. This decoupling correction is then performed. The subsequent adjustments eliminated the mutual interference between parameters. A construction response delay model was established to describe the time delay from receiving the instruction to the actual effect of each parameter control point. The response delay of the lifting point includes the hydraulic system pressure establishment time and mechanical transmission response time. The response delay of the pouring point includes the hopper movement time and concrete flow and diffusion time. The response delay of the slipform point includes the speed control instruction transmission time and the actual speed change time. The timing was adjusted according to the response delay time of each parameter control point. The instruction was issued later for parameter control points with fast response and earlier for parameter control points with slow response, so that the actual effect time of each parameter control point was synchronized as much as possible, forming a construction correction instruction sequence that includes parameter control point number, target adjustment amount, action time and action speed.

[0035] In one embodiment, the influence weight of each construction parameter on the correction is calculated based on the dominant offset mode identification results. The influence weight allocation of each construction parameter is obtained according to the geometric relationship between the operating platform position, the concrete pouring area, the slip control point, and the dominant offset direction, including: Establish a construction parameter location matrix, record the position coordinates of each area of ​​the operating platform, concrete pouring point and slip control point relative to the center of the structure, calculate the distance from each parameter control point to the center of the structure, and obtain the distance parameter set. Based on the dominant migration mode identification results, the dominant migration direction vector is determined, the angle between the position vector of each parameter control point and the dominant migration direction vector is calculated, and the angle is subjected to cosine transformation to obtain the direction influence factor. The distance parameter set is normalized to obtain the distance weight; the direction influence factor is multiplied with the distance weight to obtain the preliminary influence weight of each construction parameter. The initial influence weights are normalized so that the sum of the weights of all construction parameters is 1, thus obtaining the influence weight allocation of each construction parameter.

[0036] Specifically, a construction parameter position matrix is ​​established to record the position coordinates of each parameter control point and calculate the distance to obtain a distance parameter set. A plane rectangular coordinate system is established with the geometric center of the structure as the origin. The lifting points corresponding to each area of ​​the operating platform are located in different directions around the structure. The x and y coordinates of each lifting point in the coordinate system are recorded. The concrete pouring points are distributed above the operating platform and the coordinates of each pouring point are recorded at the corresponding hopper placement positions. The speed control positions of the slip control points corresponding to each lifting point are the same as the coordinates of the lifting points. The coordinates of all parameter control points are arranged in rows to form a construction parameter position matrix. The distance from each parameter control point to the center of the structure is calculated. For parameter control points with coordinates x and y, the distance to the origin is calculated as the square root of x squared plus y squared. The distance values ​​of all parameter control points are combined to form a distance parameter set.

[0037] Based on the dominant migration mode identification results, the dominant migration direction vector is determined and the direction influence factor is calculated. When the dominant migration mode is in the x-direction, the dominant migration direction vector is set as a unit vector along the positive x-axis with components of x-direction 1 and y-direction 0. When the dominant migration mode is in the y-direction, the dominant migration direction vector is set as a unit vector along the positive y-axis with components of x-direction 0 and y-direction 1. For each parameter control point, a position vector is constructed based on its coordinate position. The position vector points from the origin to the parameter control point, and its components are the x-coordinate and y-coordinate of that point. The position direction is then calculated. The angle between the parameter and the dominant offset direction vector is calculated by dividing the dot product of the two vectors by the inverse cosine function of the product of the magnitudes of the two vectors. The cosine value of the angle is calculated by performing a cosine transformation on the calculated angle. When the angle is 0 degrees, the cosine value is 1, indicating that the parameter control point is located directly in front of the offset direction. When the angle is 90 degrees, the cosine value is 0, indicating that the parameter control point is located on the perpendicular side of the offset direction. When the angle is 180 degrees, the cosine value is -1, indicating that the parameter control point is located on the opposite side of the offset direction. The cosine values ​​of all parameter control points are used to form the direction influence factor set.

[0038] The distance parameter set is normalized to obtain distance weights, and the initial influence weights are calculated. The normalization process sets the maximum distance value in the distance parameter set to 1, and other distance values ​​are scaled proportionally. For a parameter control point with a distance of d, its normalized distance weight is equal to d divided by the maximum distance value in the distance parameter set. The distance weights of all parameter control points are obtained. The directional influence factor of each parameter control point is multiplied by the corresponding distance weight. For a certain parameter control point, its initial influence weight is equal to the directional influence factor of that point multiplied by the distance weight of that point. The initial influence weights of all parameter control points are calculated.

[0039] The initial influence weights are normalized to obtain the influence weight allocation of each construction parameter. The sum of the initial influence weights of all parameter control points is calculated. For each parameter control point, its initial influence weight is divided by the sum of the initial influence weights to obtain the final influence weight of that point. After normalization, the sum of the weights of all construction parameters is strictly equal to 1, thus obtaining the influence weight allocation of each construction parameter.

[0040] In one embodiment, step S4 includes: The construction correction command sequence is converted into construction operation guidance. The lifting stroke of each lifting point is adjusted according to the operation platform adjustment plan, so that the lifting stroke of the lifting point on one side of the structural offset direction is less than that of the lifting point on the opposite side, resulting in the tilt of the operation platform. The concrete pouring position is adjusted according to the concrete pouring position adjustment plan, increasing the amount of concrete poured on the opposite side of the structural offset direction and decreasing the amount of concrete poured on one side of the structural offset direction. The overall slipforming speed is adjusted according to the slipforming speed adjustment plan, and slipforming is carried out only after the concrete demolding strength reaches the set value. During the construction adjustment process, structural tilt sensor data is continuously collected to calculate the amount and rate of reduction of the current verticality deviation relative to the baseline deviation before correction, thereby obtaining monitoring data on the correction effect. The correction effect is judged based on the monitoring data of the correction effect. When the deviation decreases in the expected direction and the rate of decrease is within a reasonable range, the current construction parameters are maintained. When the rate of decrease of the deviation exceeds the set upper limit, the construction adjustment range is reduced. When the deviation does not decrease as expected or increases in the opposite direction, the current construction adjustment is paused and a new construction correction instruction sequence is generated. The command parameters, execution process data, and correction effect data of this construction adjustment are linked and recorded to obtain construction correction case data.

[0041] Specifically, the construction correction instruction sequence is transformed into construction operation guidance and three types of construction adjustments are executed. The construction correction instruction sequence includes the target stroke difference of each lifting point, the target eccentricity of each pouring point, and the target slip velocity of each control point. The instruction sequence is transformed into an operation guidance document that can be executed by on-site construction personnel. The lifting stroke of each lifting point is adjusted according to the operation platform adjustment plan. When the structure shifts to the east, the west side is the opposite side of the shift, and the east side is the side of the shift direction. The stroke of each lifting point is adjusted so that the slip stroke of the east lifting point is less than that of the west lifting point each time. For example, the stroke of the east lifting point is set to 25 cm, and the stroke of the west lifting point is set to 30 cm, which produces the operation platform tilted to the east. The tilt of the platform causes the subsequently poured concrete to flow to the west under the action of gravity. Adjustments are made according to the concrete pouring position. The plan adjusts the concrete pouring position. On the opposite side of the structural offset (west side), the concrete pouring volume is increased by moving the hopper position or increasing the number of pours, so that the thickness on the west side reaches 35 cm. On the side of the structural offset (east side), the concrete pouring volume is reduced, so that the thickness on the east side is only 25 cm. The increased weight of the concrete on the west side creates a counterweight effect on the west side, causing the structure to gradually return to the westward side. The overall slipforming speed is adjusted according to the slipforming speed adjustment plan. After the concrete pouring is completed, the demolding strength of the concrete at the bottom of the formwork is tested using a concrete strength tester. When the demolding strength reaches the set value of 0.3 MPa to 0.5 MPa, the hydraulic system is activated to start slipforming. If the demolding strength is too low, the slipforming time is delayed to allow the concrete to continue to solidify. If the demolding strength has met the standard, slipforming is carried out at the adjusted speed.

[0042] During the construction adjustment process, structural tilt sensor data is continuously collected, and correction effect monitoring data is calculated. The tilt sensor continuously monitors the tilt angle changes of the structure in the x and y directions at a sampling frequency of 10 Hz. Before the construction adjustment begins, the tilt angle value at the current moment is recorded as the reference deviation before correction. During the construction adjustment, the current value of the tilt sensor is read every 10 seconds. The current tilt angle is multiplied by the structural height to convert it into the current verticality deviation. The difference between the current verticality deviation and the reference deviation before correction is calculated to obtain the verticality deviation reduction. A positive value of the deviation reduction indicates that the deviation is decreasing, while a negative value indicates that the deviation is increasing. The deviation reduction is subtracted from the deviation reduction of two adjacent 10-second intervals and then divided by the time interval of 10 seconds to obtain the deviation reduction rate. The deviation reduction rate reflects the speed of structural attitude change, thus obtaining correction effect monitoring data that includes the deviation reduction and the deviation reduction rate.

[0043] The system judges the correction effect based on the monitoring data and executes a dynamic adjustment strategy. The judgment logic includes three types of conditions and corresponding actions. When the deviation reduction is continuously positive, it means that the deviation is decreasing in the expected direction. At the same time, when the deviation reduction rate is within a reasonable range of 1 cm to 3 cm per hour, the correction effect is judged to be good. The current construction parameters are maintained and no adjustment is made. When the deviation reduction rate exceeds the set upper limit of 3 cm per hour, it is judged that the correction speed is too fast and there is a risk of overshoot. The construction adjustment range is reduced, specifically the operating platform stroke difference is reduced to 70% of the original, the concrete pouring eccentricity is reduced to 70% of the original, and the slip speed difference is reduced to 70% of the original. When the deviation reduction is negative or the deviation reduction rate is negative, it means that the deviation has not decreased as expected or has increased in the opposite direction. The current correction strategy is judged to be ineffective. The current construction adjustment is paused, the stroke of each lifting point is restored to be consistent, the concrete pouring position is restored to symmetry, and the slip speed is restored to be unified. The system returns to step S3 and re-enters the current fused verticality deviation vector to generate a new construction correction instruction sequence.

[0044] The relevant data from this construction adjustment are linked and recorded to obtain construction correction case data. The recorded content includes the start time of this correction, the baseline deviation value before correction, the parameter values ​​in the construction correction instruction sequence, the tilt angle data sequence collected every 10 seconds during the construction adjustment execution, the end time of construction adjustment, the final deviation value after correction, and the evaluation result of correction effect. These data are linked and stored in chronological order to form complete construction correction case data. The case data is marked with whether the correction was successful, the type of strategy sub-network used, the characteristics of the working conditions encountered, and the final reduction in deviation. The case data is stored in the historical database for use in the subsequent strategy network update in step S5.

[0045] In one embodiment, step S5 includes: The construction correction command sequence and the corrected actual verticality response data are input into the digital twin model for simulation calculation to obtain the expected verticality response result; the expected verticality response result is compared with the corrected actual verticality response data, and the deviation between the two is calculated to obtain the simulation error data. Based on simulation error data, deviation source analysis is performed. When the simulation error is less than a preset threshold, the construction correction case data is marked as a successful case and added to the training sample library. When the simulation error exceeds the preset threshold, the environmental data and equipment response data during the construction adjustment execution are analyzed, and the deviation source type is determined according to the sudden change in environmental excitation intensity, abnormal equipment response, and deviation of structural response characteristics. The advanced correction strategy network is updated based on the type of deviation source. When the deviation source is a sudden change in the environment, the parameters of the dynamic disturbance rejection strategy network are updated. When the deviation source is an abnormal device response, the parameters of the robust compensation strategy network are updated. When the deviation source is a structural characteristic deviation, the parameters of the basic control strategy network are updated. The updated advanced correction policy network is integrated with the training sample library to obtain the target control policy.

[0046] Specifically, the construction correction command sequence and the corrected actual verticality response data are input into a digital twin model for simulation calculation to obtain the expected verticality response result. The digital twin model is based on the finite element analysis method and includes the geometric parameters of the structure, such as the height, diameter, and wall thickness of the cylinder, material properties such as the elastic modulus and Poisson's ratio of concrete, and boundary conditions such as the foundation constraint stiffness. The model input includes the stroke adjustment amount of each lifting point in the construction correction command sequence, the concrete increase or decrease amount at each pouring point, and the slip velocity parameters. The stroke difference of the lifting points is converted into the non-uniform support force acting on the structure, and the concrete increase or decrease amount is converted into the eccentric load at the top of the structure. The displacement field and stress field of the structure under these loads and boundary conditions are calculated by the finite element solver. The horizontal displacement value and tilt angle value of the top node of the structure are extracted as the expected verticality response result. The expected verticality response result is compared with the corrected actual verticality response data. The actual verticality response data comes from the final deviation value recorded after correction in the construction correction case data. The difference between the expected verticality response result and the actual verticality response data is calculated to obtain the simulation error data. The simulation error data reflects the prediction accuracy of the digital twin model.

[0047] Based on simulation error data, deviation sources are analyzed and case data are categorized. A preset threshold of 10% of the actual deviation value is set. When the simulation error data is less than this threshold, it indicates that the digital twin model prediction matches the actual response well, and the current strategy network decision is accurate. Construction correction case data are marked as successful cases, a success label is added to the case data, and the case is added to the training sample library for subsequent network training. When the simulation error data exceeds the preset threshold, it indicates that there is a significant deviation between the actual response and the expectation, and the source of deviation needs to be analyzed. Environmental data and equipment response data during the construction adjustment execution period are extracted from the construction correction case data. Environmental data includes the standard deviation of wind speed and vibration acceleration at each moment, and equipment response data includes the actual travel and response time of each lifting point. The analysis includes health score assessment, analysis of sudden changes in environmental excitation intensity by comparing the change in wind speed standard deviation before and after correction execution. If the wind speed standard deviation suddenly increases from the initial value of 0.1 m / s² to more than 0.4 m / s², it indicates that strong wind disturbance was encountered during the correction period. The analysis also includes abnormal equipment response by checking whether the actual travel of each lifting point reaches the target travel required by the instruction. If the actual travel of a lifting point is less than 80% of the target travel or the response time exceeds twice the normal value, it indicates that the lifting point has a fault. Finally, the analysis includes deviations in structural response characteristics by comparing the structural stiffness set in the digital twin model with the stiffness exhibited by the actual structure. If the actual reduction in deviation is systematically greater than the expected reduction, it indicates that the actual structural stiffness is lower than the model setting value. Based on the above three types of analysis results, the type of deviation source is determined.

[0048] Based on the type of deviation source, the advanced correction strategy network is updated accordingly. When the deviation source is a sudden environmental change, the construction correction case data is marked as a dynamic disturbance failure sample. The fused verticality deviation vector at the moment of wind speed change, the output construction correction command sequence at that time, and the actual correction effect data are extracted from the case. These data are input into the dynamic disturbance rejection strategy network for additional training. During the training process, the network weights are adjusted so that it can output a larger range of construction adjustment parameters to offset the disturbance when facing similar sudden wind load conditions. When the deviation source is an abnormal equipment response, the case is marked as an abnormal equipment failure sample. The abnormal lift point number, health score, working status of other normal lift points, and actual correction effect are extracted from the case. These are input into the robust compensation strategy network for training so that it learns how to reallocate the adjustment amount of other lift points for compensation when a lift point failure is detected. When the deviation source is a structural characteristic deviation, the case is marked as a model parameter deviation sample. The structural stiffness parameters in the digital twin model are corrected according to the systematic difference between the actual response and the expected response. At the same time, the data obtained by resimulating after parameter correction is input into the basic control strategy network for training so that it adapts to the actual structural characteristics.

[0049] The updated advanced correction strategy network is integrated with the training sample library to obtain the target control strategy. Statistical analysis is performed on successful cases in the training sample library, categorizing them into three levels based on the degree of vertical deviation: slight deviation (less than 0.1% of the structural height), moderate deviation (between 0.1% and 0.3%), and severe deviation (exceeding 0.3%). For each deviation level, the range of operating platform travel difference, concrete pouring eccentricity, and slipform speed adjustment used in successful cases are extracted to obtain empirical data for construction parameter adjustment. This empirical data is then compared and verified with the output results of the updated advanced correction strategy network. When the strategy network outputs construction parameters for a certain deviation at the corresponding level... When the output is within the empirical range, it is confirmed to be reasonable. When the output exceeds the empirical range, constraint correction is performed to limit the output value to the empirical range boundary. Based on the verified strategy network output results, a correction construction guidance table is generated for different offset modes and deviation degrees. The guidance table is classified into three modes: x-axis offset, y-axis offset, and xy-torsion. Under each mode, the corresponding operating platform stroke difference value, concrete pouring eccentricity value, and slip speed adjustment ratio are listed according to three deviation levels: slight, moderate, and severe. At the same time, the dynamic adjustment rules are recorded, including the reasonable range of deviation reduction rate, the adjustment amplitude reduction coefficient when the rate exceeds the limit, and the pause and recalculation conditions when the deviation increases in the opposite direction. The correction construction guidance table and the dynamic adjustment rules are integrated to form a construction operation guidance document to obtain the target control strategy.

[0050] In one embodiment, the updated advanced correction policy network is integrated with the training sample library to obtain a target control policy, including: Statistical analysis was performed on successful cases in the training sample library to extract the range of operating platform tilt angle, concrete pouring eccentricity, and slip speed adjustment range corresponding to different degrees of verticality deviation, and to obtain empirical data on construction parameter adjustment. The empirical data for adjusting construction parameters is compared and verified with the output results of the updated advanced correction strategy network. When the sequence of construction correction instructions output by the strategy network is within the range of empirical data, it is confirmed as a reasonable solution. When it exceeds the range of empirical data, the output results of the strategy network are corrected and constrained. Based on the validated strategy network output, a correction construction guide table is generated for different offset modes and deviation degrees, which clarifies the platform adjustment amount, pouring eccentricity amount and slip velocity difference when offset in the x direction, the corresponding parameters when offset in the y direction, and the comprehensive adjustment measures when offset in the torsional direction. By combining the correction construction guidelines with dynamic adjustment rules, a complete construction operation guideline is established, which includes initial correction parameters, monitoring and judgment thresholds, and dynamic adjustment range, thus obtaining the target control strategy.

[0051] Specifically, statistical analysis of successful cases in the training sample library yields empirical data on construction parameter adjustments. The library iterates through all construction correction cases marked as successful, reading the initial verticality deviation, operating platform travel difference, concrete pouring eccentricity, and slipform speed adjustment ratio for each case. Cases are categorized into three levels based on the degree of verticality deviation: slight deviation (top offset less than 0.001 / 1000 of the structure height), moderate deviation (offset between 0.001 and 0.003), and severe deviation (offset exceeding 0.003). For each deviation level, the operating platform travel difference values ​​of all successful cases within that level are statistically analyzed, calculating the minimum, maximum, and average travel difference. Similarly, the minimum, maximum, and average concrete pouring eccentricity and slipform speed adjustment ratios are statistically analyzed to obtain empirical data on construction parameter adjustments.

[0052] The empirical data for adjusting construction parameters is compared and verified with the output results of the updated advanced correction strategy network. When a new fused verticality deviation vector is collected and a construction correction instruction sequence needs to be generated, the deviation vector is input into the updated advanced correction strategy network. The strategy network selects the corresponding sub-network for inference based on the complexity of the working condition and outputs the construction correction instruction sequence. It determines whether the current verticality deviation is at a slight, moderate, or severe level, queries the empirical data for adjusting construction parameters at the corresponding level, and compares the operating platform travel difference output by the strategy network with the minimum and maximum values ​​of the parameter in the empirical data. If the output value is between the minimum and maximum values, the parameter is confirmed to be reasonable. If the output value is less than the minimum value, it is corrected to the minimum value. If the output value is greater than the maximum value, it is corrected to the maximum value. The same interval constraint judgment and correction operation is performed on the concrete pouring eccentricity and slip-up speed adjustment ratio.

[0053] Based on the output results of the validated strategy network, a correction construction guidance table is generated. The guidance table is organized according to two dimensions: dominant offset mode and deviation degree. The dominant offset mode includes three types: unidirectional offset in the x-direction, unidirectional offset in the y-direction, and xy composite torsional offset. The deviation degree includes three levels: slight, moderate, and severe. These are combined to form nine correction scenarios. For each scenario, the adjustment measures for the operating platform are clearly listed. For x-direction offset, the target travel difference value of the lifting points on both sides of the x-direction is recorded. For y-direction offset, the target travel difference value of the lifting points on both sides of the y-direction is recorded. For torsional offset, the differentiated travel values ​​of the lifting points in the four directions are recorded. The concrete pouring position adjustment measures are clearly defined. For x-direction offset, the pouring thickness that should be increased on the opposite side of the x-direction offset and the pouring thickness that should be decreased on the offset side are recorded. For y-direction offset, the pouring thickness adjustment corresponding to the y-direction is recorded. For torsional offset, the differentiated thickness distribution of multiple pouring points is recorded. The slip speed adjustment measures are clearly defined, and the percentage increase or decrease of the control points in each direction relative to the reference speed is recorded.

[0054] The target control strategy is obtained by combining the deviation correction construction guidance table with the dynamic adjustment rules. The dynamic adjustment rules define the judgment conditions and adjustment actions for adjusting construction parameters based on real-time monitoring data during the deviation correction execution process. This includes setting the reasonable range of deviation reduction rate to 1 cm to 3 cm per hour. When the measured deviation reduction rate is within this range, the current construction parameters remain unchanged. When the deviation reduction rate exceeds the upper limit of 3 cm per hour, the construction adjustment amplitude is reduced. The adjustment ratios of the operating platform stroke difference, concrete pouring eccentricity, and slipform speed are simultaneously reduced to 70% of their original values. When the deviation reduction rate is negative or the deviation increases in the opposite direction, the current construction adjustment is paused and the process returns to step S3 to regenerate the instruction sequence. The initial deviation correction parameters in the deviation correction construction guidance table, the monitoring judgment thresholds and adjustment amplitude coefficients in the dynamic adjustment rules are integrated into a complete construction operation guidance document. The document includes the offset mode recognition method, the specific construction parameter values ​​corresponding to each scenario, real-time monitoring requirements, and dynamic adjustment procedures, thus obtaining the target control strategy.

[0055] The above describes the adaptive control method for structural perpendicularity based on autonomous progressive correction in the embodiments of this application. The following describes the adaptive control system for structural perpendicularity based on autonomous progressive correction in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 3 One embodiment of the structural verticality adaptive control system based on autonomous progressive correction in this application includes: The acquisition module is used to collect structural attitude data and environmental data, dynamically weight the confidence based on the intensity of environmental excitation, fuse multi-source data, and construct a fused verticality deviation vector. The training module is used to perform three-stage progressive training based on the fused verticality deviation vector to obtain an advanced correction policy network; The input module is used to identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme. The correction module is used to adjust construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor the structural response in real time and adjust dynamically; The verification module is used to perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

[0056] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the structure verticality adaptive control method based on autonomous progressive correction.

[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a structural verticality adaptive control device (which may be a personal computer, server, or network device, etc.) based on autonomous progressive correction to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A structural verticality adaptive control method based on autonomous progressive correction, applied to real-time verticality control during the construction phase of high-rise buildings, characterized in that... The method includes: Step S1: Collect structural attitude data and environmental data, dynamically weight the data based on the intensity of environmental excitation, fuse the multi-source data, and construct a fused verticality deviation vector; Step S2: Perform three-stage progressive training based on the fused verticality deviation vector to obtain the advanced correction policy network; Step S3: Identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme; Step S4: Adjust the construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor the structural response in real time and adjust dynamically; Step S5: Perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

2. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 1, characterized in that, Step S1 includes: The structure under construction is monitored by tilt sensors, displacement sensors and vibration sensors to obtain structural tilt data, foundation settlement data and vibration acceleration data, and the environmental excitation intensity is calculated based on the vibration acceleration data. Based on the environmental excitation intensity, the confidence weight of each sensor data is calculated. A fixed confidence weight is set for the tilt angle data, a medium confidence weight is set for the settlement data, and a dynamic confidence weight is set for the vibration data according to the exponential decay law, so as to obtain a sensor dataset with confidence labels. The sensor dataset with confidence level labels is input into an extended Kalman filter for state estimation to obtain the filtered state estimation result. The filtered state estimation result is subjected to coordinate transformation processing to convert the tilt angle into an offset and the tilt angle change rate into an offset velocity, thereby obtaining the fused verticality deviation vector.

3. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 1, characterized in that, Step S2 includes: Static working condition training is performed based on the fused verticality deviation vector, and a foundation control strategy network is established using historical construction data under normal slip zone conditions. Based on the basic control strategy network, the training conditions are extended by adding historical data on wind load changes and offset speed changes to perform dynamic disturbance rejection training, thus obtaining the dynamic disturbance rejection strategy network. Based on the extended training conditions of the dynamic anti-disturbance strategy network, historical data of abnormal device response and partial device failure are added for robust compensation training to obtain the robust compensation strategy network. The basic control strategy network, the dynamic disturbance rejection strategy network, and the robust compensation strategy network are integrated, and a strategy selection mechanism is established based on the complexity of the construction conditions to obtain the advanced correction strategy network.

4. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 1, characterized in that, Step S3 includes: Modal decomposition analysis is performed on the fused verticality deviation vector to calculate the ratio between the offset in the x-direction and the offset in the y-direction. The dominant offset direction is determined based on the offset ratio to obtain the dominant offset mode recognition result. Based on the dominant offset mode recognition results, the influence weight of each construction parameter on the correction is calculated. According to the geometric relationship between the operating platform position, the concrete pouring area, the slip control point and the dominant offset direction, the influence weight allocation of each construction parameter is obtained. The fused verticality deviation vector and the influence weight allocation are input into the advanced correction strategy network for reasoning and decision-making. The corresponding strategy sub-network is selected according to the complexity of the current construction condition to obtain a preliminary construction correction scheme. The preliminary construction correction scheme is optimized by parameters, decoupled and corrected according to the mutual influence relationship between various construction parameters, and the timing is adjusted in combination with the construction response delay characteristics to obtain the construction correction instruction sequence.

5. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 4, characterized in that, The influence weights of each construction parameter on the correction are calculated based on the dominant offset mode recognition results. The influence weight allocation of each construction parameter is obtained according to the geometric relationship between the operating platform position, the concrete pouring area, the slip control point, and the dominant offset direction, including: Establish a construction parameter location matrix, record the position coordinates of each area of ​​the operating platform, concrete pouring point and slip control point relative to the center of the structure, calculate the distance from each parameter control point to the center of the structure, and obtain the distance parameter set. Based on the dominant offset mode identification results, the dominant offset direction vector is determined, the angle between the position vector of each parameter control point and the dominant offset direction vector is calculated, and the angle is subjected to cosine transformation to obtain the direction influence factor. The distance parameter set is normalized to obtain the distance weight; the direction influence factor is multiplied by the distance weight to obtain the preliminary influence weight of each construction parameter. The initial influence weights are normalized so that the sum of the weights of all construction parameters is 1, thus obtaining the influence weight allocation of each construction parameter.

6. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 1, characterized in that, Step S4 includes: The construction correction instruction sequence is converted into construction operation guidance. The lifting stroke of each lifting point is adjusted according to the operation platform adjustment plan, so that the lifting stroke of the lifting point on one side of the structural offset direction is less than that of the lifting point on the opposite side, resulting in the tilt of the operation platform. The concrete pouring position is adjusted according to the concrete pouring position adjustment plan, increasing the amount of concrete poured on the opposite side of the structural offset direction and decreasing the amount of concrete poured on one side of the structural offset direction. The overall slipforming speed is adjusted according to the slipforming speed adjustment plan, and slipforming is carried out only after the concrete demolding strength reaches the set value. During the construction adjustment process, structural tilt sensor data is continuously collected to calculate the amount and rate of reduction of the current verticality deviation relative to the baseline deviation before correction, thereby obtaining monitoring data on the correction effect. Based on the aforementioned correction effect monitoring data, the correction effect is judged. When the deviation decreases in the expected direction and the rate of decrease is within a reasonable range, the current construction parameters are maintained. When the rate of deviation decrease exceeds the set upper limit, the construction adjustment range is reduced. When the deviation does not decrease as expected or increases in the opposite direction, the current construction adjustment is paused and a new construction correction instruction sequence is generated. The command parameters, execution process data, and correction effect data of this construction adjustment are linked and recorded to obtain construction correction case data.

7. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 1, characterized in that, Step S5 includes: The construction correction instruction sequence and the corrected actual verticality response data are input into the digital twin model for simulation calculation to obtain the expected verticality response result; the expected verticality response result is compared with the corrected actual verticality response data, and the deviation between the two is calculated to obtain the simulation error data; Based on the simulation error data, deviation source analysis is performed. When the simulation error is less than a preset threshold, the construction correction case data is marked as a successful case and added to the training sample library. When the simulation error exceeds the preset threshold, the environmental data and equipment response data during the construction adjustment execution are analyzed, and the deviation source type is determined according to the sudden change in environmental excitation intensity, abnormal equipment response, and deviation of structural response characteristics. The advanced correction strategy network is updated in a targeted manner based on the type of deviation source. When the deviation source is a sudden change in the environment, the parameters of the dynamic anti-disturbance strategy network are updated. When the deviation source is an abnormal device response, the parameters of the robust compensation strategy network are updated. When the deviation source is a structural characteristic deviation, the parameters of the basic control strategy network are updated. The updated advanced correction policy network is integrated with the training sample library to obtain the target control policy.

8. The adaptive control method for structural perpendicularity based on autonomous progressive correction according to claim 7, characterized in that, The step of integrating the updated advanced correction policy network with the training sample library to obtain the target control policy includes: Statistical analysis was performed on the successful cases in the training sample library to extract the operating platform tilt angle range, concrete pouring eccentricity range, and slipform speed adjustment range corresponding to different degrees of verticality deviation, so as to obtain empirical data on construction parameter adjustment. The construction parameter adjustment experience data is compared and verified with the updated advanced correction strategy network output results. When the construction correction instruction sequence output by the strategy network is within the range of experience data, it is confirmed as a reasonable solution. When it exceeds the range of experience data, the strategy network output results are corrected and constrained. Based on the validated strategy network output, a correction construction guide table is generated for different offset modes and deviation degrees, which clarifies the platform adjustment amount, pouring eccentricity amount and slip velocity difference when offset in the x direction, the corresponding parameters when offset in the y direction, and the comprehensive adjustment measures when offset in the torsional direction. By combining the aforementioned correction construction guidance table with dynamic adjustment rules, a complete construction operation guide including initial correction parameters, monitoring and judgment thresholds, and dynamic adjustment range is established, thus obtaining the target control strategy.

9. A structural verticality adaptive control system based on autonomous progressive correction, applied in the construction phase of high-rise buildings, characterized in that, For implementing the adaptive control method for structural verticality based on autonomous progressive correction as described in any one of claims 1-8, the adaptive control system for structural verticality based on autonomous progressive correction comprises: The acquisition module is used to collect structural attitude data and environmental data, dynamically weight the confidence based on the intensity of environmental excitation, fuse multi-source data, and construct a fused verticality deviation vector. The training module is used to perform three-stage progressive training based on the fused verticality deviation vector to obtain an advanced correction policy network; The input module is used to identify the dominant offset mode of the fused verticality deviation vector, input the fused verticality deviation vector into the advanced correction strategy network, and output a sequence of construction correction instructions including the operation platform adjustment scheme, the concrete pouring position adjustment scheme, and the slip speed adjustment scheme. The correction module is used to adjust construction parameters according to the construction correction instruction sequence to perform verticality correction, monitor the structural response in real time and adjust dynamically; The verification module is used to perform simulation verification on the corrected verticality response results, update the advanced correction strategy network based on error feedback, and generate the target control strategy.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the adaptive control method for structural verticality based on autonomous progressive correction as described in any one of claims 1 to 8.

Images