A method for constructing a digital twin of a high slope anchor support construction of a hydropower station

Abstract

This invention proposes a method for constructing a digital twin for anchor cable support construction on high slopes of hydropower stations, belonging to the field of hydropower station digital twin technology. The method includes: uniquely identifying construction parameters and establishing a correspondence between physical entities and digital models; constructing a directed acyclic graph (DAG) for parameter association; collecting multi-source heterogeneous data to obtain standardized data; using a hierarchical weighted fusion method to perform association and fusion of the standardized data, and writing the fusion result into the DAG; performing parameter mapping on the fusion result based on the DAG and performing accuracy calibration; using key process quality acceptance nodes as trigger conditions, employing a process state machine-driven incremental differential update mechanism to propagate the accuracy-calibrated mapped parameter values ​​along the DAG and write them into the digital twin, obtaining a digital twin synchronized with the on-site construction status in real time. This invention can achieve accurate mapping of parameters throughout the entire construction process and real-time dynamic synchronization of the digital twin.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology for hydropower stations, and in particular to a method for constructing a digital twin for the construction of anchor cable support for high slopes in hydropower stations. Background Technology

[0002] The construction of prestressed anchor cable support for high slopes in hydropower station projects is a complex and parameter-intensive systematic project involving multiple key processes such as drilling, cable fabrication, grouting, tensioning, and anchor sealing. The construction parameters generated by each process are diverse in type and scattered in origin, and the site is often located in complex environments with high altitudes and large temperature differences. Sensor measurements are easily affected by air pressure fluctuations and temperature changes, leading to systematic deviations. Simultaneously, there is a lack of effective dynamic correlation between the physical state of the construction site and the digital twin model. Parameter mapping relationships are opaque, multi-source heterogeneous data is difficult to integrate, and twin updates rely on manual triggering or periodic full overwrites, failing to reflect the timely progress of on-site processes. This results in significant deficiencies in parameter accuracy, data integrity, and response time between the digital twin and the physical construction process, hindering the practical application of digital twin technology in the refined management and control of high slope support construction.

[0003] Chinese invention patent application CN117540186A discloses a method for assessing the stability of high road cut slopes based on multi-source data fusion. This method, targeting slope stability monitoring scenarios, collects multi-source data such as rainfall, surface displacement, deep subsurface displacement, groundwater level, and anchor cable stress through pre-configured detection items. It constructs a multi-source risk matrix centered on the state transition equation and combines risk profiling with a multi-mode fusion simulation model to comprehensively assess slope stability. While this approach has some value in the fusion analysis of multi-source slope data and comprehensive stability assessment, its overall focus is on slope safety assessment rather than the construction of a digital twin of the construction process. It does not address the structured coding of construction parameters and the mapping management of physical entities, nor does it establish a dynamic parameter update and accuracy calibration mechanism for construction procedures. In particular, it does not consider the elimination of systematic biases in sensors at high altitudes. Summary of the Invention

[0004] In view of this, the present invention provides a method for constructing a digital twin for the construction of anchor cable support on high slopes of hydropower stations. Through structured parameter encoding, parameter association directed acyclic graph modeling, environmental compensation, hierarchical weighted data fusion, two-factor adaptive accuracy calibration, and incremental differential update driven by process state machine, the method achieves accurate mapping of parameters throughout the construction process and real-time dynamic synchronization of the digital twin.

[0005] The technical solution of this invention is implemented as follows: This invention provides a method for constructing a digital twin for anchor cable support construction on high slopes of hydropower stations, including: S1. Organize all kinds of construction parameters in the entire process of anchor cable support construction, use multi-level coding rules to uniquely identify each parameter, and bind the parameter code with the attribute fields of physical entities and digital twin components in sync to establish a one-to-one correspondence between physical entities and digital twins. S2. Construct a directed acyclic graph of parameter association with each parameter as a node and the driving relationship between parameters as a directed edge. The nodes are divided into source parameter nodes and derived parameter nodes. The source parameter nodes store the sensor readings after correction by the environmental compensation coefficient. The derived parameter nodes are calculated sequentially by the mapping function corresponding to each node, and finally form the mapping parameter values ​​covering all parameters. S3. Perform anomaly correction and filtering / denoising processing on the multi-source heterogeneous data collected during the construction process to obtain standardized data. S4. Using parameter encoding as the index key, a hierarchical weighted fusion method is used to perform cross-dimensional association fusion of standardized data, and the fusion result is written into the attribute field of the corresponding node in the parameter-associative directed acyclic graph. S5. Perform parameter mapping on the fusion result based on the parameter-correlated directed acyclic graph, use the proportional correction coefficient and the offset correction coefficient to calibrate the accuracy of the mapped output value, and adaptively update the proportional correction coefficient and the offset correction coefficient according to the deviation between the measured value and the mapped output value to obtain the accuracy-calibrated mapped parameter value. S6. Using the key process quality acceptance node as the trigger condition, the incremental differential update mechanism driven by the process state machine is adopted to propagate the accuracy-calibrated mapping parameter values ​​along the parameter-related directed acyclic graph and write them into the digital twin. When the acceptance fails, the state re-verification is triggered to obtain a digital twin that is synchronized with the on-site construction status in real time.

[0006] Preferably, the multi-level coding rule in step S1 is a four-level coding rule, with the coding structure consisting of project code, slope zone code, anchor cable number, and parameter type code in sequence; wherein the parameter type code distinguishes parameters into five categories: design parameters, construction parameters, quality parameters, cost parameters, and archived parameters, and the coding order of each level is spliced ​​together to form a unique identifier for all parameters of each anchor cable.

[0007] Preferably, the sensor readings stored in the source parameter node in step S2 are converted into the compensated real physical quantity through a two-factor environmental compensation transformation. The compensation formula is: ; in, The original sensor reading; The pressure correction factor is defined as: ; Standard atmospheric pressure The air pressure was measured on-site. The temperature difference correction factor is defined as: ; For the temperature coefficient of the corresponding sensor, The temperature difference between day and night at the site; After compensation Write the source parameter node attribute field.

[0008] Preferably, in step S2, the mapping function of the derived parameter node takes the compensation value of the source parameter node as input, performs a breadth-first traversal on the directed acyclic graph of parameter association, and sequentially triggers the calculation of the mapping function of each derived parameter node and writes the derived parameter value. Each directed edge... The general form of the mapping function is: ; in, For derived parameter values, The current stored value, As the baseline value for derived parameters, Let k be the field calibration coefficient. Let K be the k-th basis function, where k = 1, 2, ..., K, K is the order of the basis function expansion, and k represents the index of the basis function expansion term; After the output value of the mapping function for each derived parameter node is written, the design constraint boundary check is performed immediately: ; in, For derived parameter nodes The current mapping value, and These are the lower and upper design limits bound to the node, respectively. When a node's status is abnormal, an early warning signal is generated and transmitted.

[0009] Preferably, the multi-source heterogeneous data in step S3 is divided into five categories of collection sources: Design data is imported all at once during the design phase. Real-time sensor data is automatically collected through the field sensor network. On-site construction and quality data, including construction logs, process acceptance records, and quality inspection reports, are entered by on-site personnel. Geological and environmental data are collected through geological survey and environmental monitoring equipment; Cost accounting data is uploaded regularly through the cost management system; The above five types of data are processed for format normalization and unit of measurement to form standardized data with consistent format. The anomaly correction is performed to fill in and remove missing values ​​and outliers that exceed the reasonable range. The filtering and denoising uses filtering algorithms to suppress noise and drift in sensor signals in complex high-altitude environments.

[0010] Preferably, the filtering and noise reduction in step S3 targets the pressure jump data caused by wind interference and the sensor drift data caused by day-night temperature difference in the real-time sensor data. The filtering algorithm is used to perform suppression processing so that the integrity of the cleaned data is not less than 99%.

[0011] Preferably, the layered weighted fusion method described in step S4 is executed in three layers: bottom-layer fusion, middle-layer fusion, and top-layer fusion. The underlying fusion process applies the same parameter to multi-source data according to the fusion weights of each data source. Perform weighted fusion, fusion value for: ; in, The measured value of this parameter is given by data source s. To correspond to the fusion weights, the initial fusion weights for each data source are comprehensively calibrated based on sensor accuracy levels, historical data integrity, and on-site calibration records, and are determined through a fusion quality feedback coefficient. Dynamic adjustment; The middle layer fusion uses the four-level parameter code as a unified index key to perform cross-dimensional association matching, and associates and mounts the construction parameters, quality parameters, cost parameters and geological environment data after the bottom layer fusion to the same anchor cable data record. The top-level fusion writes the cross-dimensional complete data records after the middle-level association into the attribute fields of the corresponding component nodes of the digital twin model according to the parameter encoding index.

[0012] Preferably, the accuracy calibration in step S5 uses a proportional correction factor. With offset correction factor Constructing a two-factor correction pair, for the parameter nodes Mapping output value Perform calibration, accuracy calibration value for: ; The adaptive update is based on the deviation between the measured value and the mapped output value. To drive, adjust according to preset step size and Update them separately to ensure that the mapping accuracy is no less than 98%; Used to correct proportionality deviation. Used to correct constant deviations, both gradually converge as calibration data accumulates during process advancement.

[0013] Preferably, the process state machine in step S6 includes a state to be activated. Execution status Pending acceptance status Accepted status Status pending re-inspection Five states; The incremental differential update mechanism calculates the mapped value after accuracy calibration. With the twin's current stored value The difference : ; Only when Write is executed at the time. To update the threshold, which is used to filter out minute fluctuations caused by sensor noise; Process-level updates are by Triggered, the update scope is limited to the current process source parameter node set and propagates breadth-first along the directed acyclic graph to downstream derived parameter nodes; Overall update by Triggered, the update scope is expanded to all nodes in the current process and their associated strength weights. Not lower than the threshold Strongly interconnected nodes across processes; Save a full parameter state snapshot after each overall update. Those that fail the inspection will be allowed to enter. Restore to the previous process snapshot .

[0014] Preferably, it also includes a dual-path reverse optimization closed loop: The first approach is based on the deviations observed during each acceptance test. As input, weighted attenuation coefficient Association strength weights for corresponding directed edges in a directed acyclic graph with parameter associations. Perform a decay update: ; The second approach introduces a fusion quality feedback coefficient. Record the average deviation of the data source s over the last K acceptance tests. Adjust the step size according to the fusion weight. Perform the update: ; Weighting of data sources during underlying fusion Proportional to the present Data sources with consistently large historical biases will have their weights automatically reduced. Two reverse optimization signals drive the mapping structure of the parameter-correlated directed acyclic graph and the underlying fusion weights to converge bidirectionally, so that the overall mapping accuracy is maintained at an error of no more than 2% and the data integrity is no less than 99%. The superscripts t and t+1 represent the acceptance round number.

[0015] The present invention has the following advantages over the prior art: (1) This invention uses a multi-level coding rule to uniquely identify various parameters in the entire process of anchor cable support construction, and synchronously binds the parameter code with the attribute fields of physical entities and digital twin components, establishing a one-to-one correspondence between physical entities and digital twins. This structured coding mechanism effectively solves the problem of scattered storage and difficulty in unified management of various types of construction parameters, enabling all construction parameters of any single anchor cable to be accurately located and completely traced through the coding index.

[0016] (2) This invention constructs a directed acyclic graph (DAG) covering all construction parameters, using parameters as nodes and the driving relationships between parameters as directed edges. It manages source parameter nodes and derived parameter nodes hierarchically, and calculates derived parameter values ​​sequentially through mapping functions. Compared to traditional black-box mapping methods that rely on direct assignment or condition triggering, the DAG structure explicitly expresses the dependencies between parameters. Parameter changes propagate automatically along the graph structure, achieving a linked mapping of "physical state change—parameter change—model attribute update," thus improving the interpretability and maintainability of the parameter mapping process.

[0017] (3) Before storing sensor readings at the source parameter node, this invention introduces air pressure correction coefficients and temperature difference correction coefficients to perform environmental compensation on the original readings, specifically eliminating the systematic biases generated by the sensors under high altitude, low air pressure, and large diurnal temperature differences. This environmental compensation mechanism acts at the very beginning of the data entering the mapping process, ensuring from the source that the physical quantities entering the twin can truly reflect the actual on-site conditions, thereby avoiding the accumulation and amplification of systematic errors in subsequent mapping and fusion stages.

[0018] (4) This invention uses parameter encoding as the index key and employs a hierarchical weighted fusion method to perform bottom-level weighted fusion, mid-level cross-dimensional correlation matching, and top-level twin integration on standardized multi-source heterogeneous data. It integrates design data, real-time sensor data, manually entered data, third-party testing data, and cost archive data into a single data record, and uses a fusion quality feedback coefficient to drive the adaptive adjustment of the weights of each data source. This mechanism effectively overcomes the difficulties of fusion due to inconsistent multi-source data formats and significant differences in source credibility, enabling the twin to carry multi-dimensional construction information and supporting subsequent multi-dimensional collaborative analysis.

[0019] (5) This invention employs a two-factor accuracy calibration mechanism consisting of a proportional correction coefficient and an offset correction coefficient to calibrate the mapped output value. The deviation between the measured value and the mapped value drives the two correction coefficients to continuously and adaptively update as the process progresses. This two-factor design can address both proportional and constant deviations. As construction data accumulates, the correction coefficients gradually converge, and the overall mapping accuracy continuously improves. Compared with a single error correction method, this design has stronger adaptive capability and more stable calibration effect.

[0020] (6) This invention uses the quality acceptance node of key processes as the trigger condition and adopts a process state machine-driven incremental differential update mechanism. Each time, only the parameter increments that have changed are calculated and written, replacing the traditional full overwrite method, which effectively reduces the writing overhead and improves the update efficiency. When the acceptance fails, the system automatically triggers a status re-verification and restores the system to the snapshot of the previous process, ensuring that the state of the digital twin strictly corresponds to the actual construction state on site. This mechanism enables the digital twin to have the ability to continuously and dynamically update with the construction process and supports accurate backtracking of the state of any process node. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram of the layered weighted fusion architecture of the present invention; Figure 3 This is a flowchart of the process state machine and incremental differential update of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, this invention provides a method for constructing a digital twin for anchor cable support construction on high slopes of hydropower stations, comprising: S1. Organize all kinds of construction parameters in the entire process of anchor cable support construction, use multi-level coding rules to uniquely identify each parameter, and bind the parameter code with the attribute fields of physical entities and digital twin components in sync to establish a one-to-one correspondence between physical entities and digital twins. S2. Construct a directed acyclic graph of parameter association with each parameter as a node and the driving relationship between parameters as a directed edge. The nodes are divided into source parameter nodes and derived parameter nodes. The source parameter nodes store the sensor readings after correction by the environmental compensation coefficient. The derived parameter nodes are calculated sequentially by the mapping function corresponding to each node, and finally form the mapping parameter values ​​covering all parameters. S3. Perform anomaly correction and filtering / denoising processing on the multi-source heterogeneous data collected during the construction process to obtain standardized data. S4. Using parameter encoding as the index key, a hierarchical weighted fusion method is used to perform cross-dimensional association fusion of standardized data, and the fusion result is written into the attribute field of the corresponding node in the parameter-associative directed acyclic graph. S5. Perform parameter mapping on the fusion result based on the parameter-correlated directed acyclic graph, use the proportional correction coefficient and the offset correction coefficient to calibrate the accuracy of the mapped output value, and adaptively update the proportional correction coefficient and the offset correction coefficient according to the deviation between the measured value and the mapped output value to obtain the accuracy-calibrated mapped parameter value. S6. Using the key process quality acceptance node as the trigger condition, the incremental differential update mechanism driven by the process state machine is adopted to propagate the accuracy-calibrated mapping parameter values ​​along the parameter-related directed acyclic graph and write them into the digital twin. When the acceptance fails, the state re-verification is triggered to obtain a digital twin that is synchronized with the on-site construction status in real time.

[0025] This invention addresses the core issues of prestressed anchor cable support construction on high slopes of hydropower stations (typical altitude 3100m, on-site air pressure 63.6kPa, and diurnal temperature difference 15℃), and proposes a systematic improvement scheme covering the entire process of parameter mapping, data fusion, and dynamic updating. These issues include the disconnect between the physical construction process and the digital model, the difficulty in integrating multi-source heterogeneous data, and the lack of dynamic response capability of the twin. At the parameter mapping level, pressure correction coefficients and temperature difference correction coefficients are introduced to compensate for the environmental conditions of the original sensor readings, constructing a general correlation mapping function framework. Simultaneously, design parameters are embedded as constraint boundary layers in the graph structure, allowing design intent to directly participate in real-time constraints and anomaly warnings during the mapping process. This achieves structured management of the linked mapping of "physical state change—parameter change—model attribute update." At the accuracy calibration level, a two-factor adaptive accuracy calibration mechanism consisting of proportional correction coefficients and offset correction coefficients is proposed. This mechanism continuously and adaptively updates as the process progresses, keeping the mapping accuracy error stably controlled within 2%. At the dynamic update level, a hierarchical incremental update algorithm driven by a process state machine is proposed. A five-state process state machine clarifies the update timing, and an incremental differential mechanism replaces full overwrite. This is complemented by a three-level over-threshold correction logic, structured update log management, and a dual-path reverse optimization closed loop. This ensures that the digital twin strictly corresponds to the on-site construction state, can be accurately traced back, and can continuously self-optimize. Ultimately, a high-precision, dynamically adaptive, and complete digital twin construction technology system is built, achieving a mapping accuracy of no less than 98%, data integrity of no less than 99%, and update latency of no more than 10 seconds.

[0026] In one embodiment of the present invention, step S1 includes: The project involves the construction of prestressed anchor cable support for a high slope in a hydropower station in Southwest China. The site altitude is approximately 3100m, the measured air pressure is approximately 63.6kPa, and the diurnal temperature range is approximately 15℃. Before construction, all anchor cable construction parameters were systematically analyzed according to their physical relationships, the transmission paths between parameters were identified, and the corresponding design constraint boundaries were marked for each construction parameter, including three attributes: lower design limit, upper design limit, and design target value. This provided the input basis for mapping the constraint layer in step S2.

[0027] For the construction of prestressed anchor cable support on high slopes of hydropower stations, there are numerous parameters involved on the construction site, including design parameters, construction process parameters, quality acceptance parameters, cost parameters, and archiving parameters. Furthermore, dozens or even hundreds of anchor cables may be distributed within the same slope area. Without systematic identification of each anchor cable's parameters, problems such as data confusion and difficulty in tracing the source can easily arise during the construction of the digital twin. Therefore, this step adopts a four-level coding rule to uniquely identify all construction parameters. The four-level coding structure is as follows: project code, slope zone code, anchor cable number, and parameter type code. The project code identifies the hydropower station project to which it belongs, for example, "RM"; the slope zoning code corresponds to the sub-item code of the slope project, for example, "030601"; the anchor cable number follows the construction sequence number, for example, "0001"; the parameter type code distinguishes five categories: design parameters (SJ), construction parameters (SG), quality parameters (ZL), cost parameters (ZJ), and archived parameters (GD). The codes at each level are concatenated in sequence to form a unique identifier for all parameters of each anchor cable. A complete code example is "RM-030601-0001-SG". After the code is generated, it is synchronously bound to the attribute fields of the corresponding components in the anchor cable physical entity label and digital twin model, establishing a one-to-one correspondence between "physical entity - code - digital model". This ensures that any parameter can be uniquely located to the corresponding node in the physical entity and digital model through the code, achieving comprehensive traceability. This coding system also serves as a unified index key for cross-dimensional correlation matching in the multi-source data mid-layer fusion in step S4, running through the entire data fusion process.

[0028] In one embodiment of the present invention, step S2 includes: A three-dimensional mapping model of "physical parameters - digital parameters - model attributes" is constructed, and the design parameters are transformed into digital parameters that can be recognized and stored by the digital twin. These parameters are then associated with the corresponding component attributes of the model to form the initial static skeleton of the twin.

[0029] Organize all the construction parameters identified in step S1 into a directed acyclic graph. This allows for unified management of the linkage and mapping relationships between parameters.

[0030] The node set V contains two types of nodes: source nodes ( ), which can be parameters directly collected by sensors or on-site, such as grouting pressure P, grouting volume Q, tension force F, drilling depth D, etc.; derived nodes ( (i) parameters that need to be calculated using a mapping function, such as grouting density. Anchor cable stress verification value Process completion rate (C), etc.

[0031] Each node It contains four attributes: parameter encoding (Corresponding to step S1 level 4 encoding), current twin storage value Parameter type Node update status (Values ​​are "Pending Update / Updated / Abnormal"). Each directed edge in the directed edge set E. Indicates parameters For parameters There exists a mapping relationship, where the direction of an edge represents the transmission path of a change in physical state, and each edge is accompanied by a mapping function. Association strength weight .

[0032] The mapping is performed in two stages. The first stage is the environmental compensation assignment for the source parameter nodes: the raw sensor readings are transformed by a two-factor environmental compensation to obtain the compensated true physical quantities. The compensation formula is: ; in, The original sensor reading; The pressure correction factor is defined as follows: , Standard atmospheric pressure The air pressure was measured on-site. The temperature difference correction factor is defined as follows: , For the temperature coefficient of the corresponding sensor, This refers to the temperature difference between day and night on site. In this project... , ,correspond After compensation Write the source parameter node attribute field.

[0033] The second level involves the associative mapping function propagating along the directed edges: After the source node is assigned a value, the system performs a breadth-first traversal of the parameter-associated directed acyclic graph, triggering the mapping function of each derived node level by level to calculate and write the derived parameter values. Each directed edge... The general form of the mapping function is: ; in, For derived parameter values, The current stored value, As the baseline value for derived parameters, Let k be the field calibration coefficient. The basis functions are selected based on the physical relationships between the parameters, including linear and logarithmic terms. It should be noted that in the above formulas, the summation index k, the superscript k of the field calibration coefficients, and the subscript k of the basis functions all represent the ordinal numbers of the same basis function expansion term; the three correspond one-to-one.

[0034] Take two pairs of typical highly correlated parameters as examples: Regarding the relationship between grouting pressure and grouting density (edge) The triggering condition is ( The minimum effective grouting pressure threshold is set to 0.5 MPa. When the condition is met, the mapping function is taken as follows: ; in, The baseline grouting density is given by [reference value], and Q is the measured grouting volume after environmental compensation. For reference grouting volume, These are the field calibration coefficients.

[0035] Verification of tension force and anchor cable stress (side) The mapping function is taken as: ; in, The measured tension force after first-level compensation, where A is the cross-sectional area of ​​the anchor cable. This is the temperature difference correction factor. Calculation results and design strength The comparison determines whether the strength requirements are met. All other parameters are managed uniformly within the graph using the same node-edge-function structure.

[0036] In the directed graph G, each parameter node is... Bind the corresponding design constraint nodes It contains three attributes: design lower limit (e.g., lower limit of grouting pressure 0.1MPa), upper limit of design pressure (e.g., grouting pressure upper limit 0.5MPa), design target value (e.g., tensile stress design value of 1500kN). The derived parameter values ​​are output in the second-level mapping function. Then, immediately perform design constraint boundary verification: ; in, and These are the lower and upper design limits bound to the node, respectively, both derived from the design constraint boundaries marked in step S1.

[0037] When a node is in an abnormal state, the system sends a warning signal to the state machine of step S6 to prevent this step from entering the overall update process until the abnormality is resolved.

[0038] The above steps propose an explicit two-level mapping based on a directed acyclic graph, which maintains the physical validity of the mapping results even in complex high-altitude environments. Temperature coefficients can be calibrated separately for different sensor types. Alternatively, when the acquisition frequency is increased, a sliding window mean can be introduced to replace single-point compensation in order to reduce the impact of short-term fluctuations on the source node assignment and further compress the first-level compensation error.

[0039] In one embodiment of the present invention, step S3 includes: The multi-source heterogeneous data generated during construction is clearly divided into five collection sources: Design data, including BIM model, anchor cable design parameters (drilling depth 60m, drilling angle 15°, grouting pressure range 0.1~0.5MPa, tensile stress design value 1500kN, etc.), and slope support scheme, are imported all at once during the design phase; Real-time sensor data, including borehole depth, grouting pressure, and tensile stress, are automatically collected through a sensor network deployed on-site, with a sampling frequency of once per minute. On-site construction and quality data, including construction logs, process acceptance records, and quality inspection reports, are entered by on-site personnel and recorded on the same day they occur. Geological and environmental data, including high-altitude air pressure (63.6 kPa), diurnal temperature range (15°C), geological structure and groundwater information, were collected through geological exploration and environmental monitoring equipment. Cost accounting data, including design quantities, material purchase unit prices, labor costs, and settlement information, are uploaded regularly through the cost management system.

[0040] The above five types of raw data are processed for format standardization and unit of measurement standardization: real-time sensor data is uniformly converted to JSON format, with fields including timestamp, anchor cable code and parameter value, numerical precision is retained to 2 decimal places, and sampling frequency is 1 time / minute; text record data (construction log, quality inspection report) is uniformly converted to PDF / OFD format, with fields named according to the standard of "anchor cable number - process name - inspection time - inspection result"; numerical data is standardized in unit of measurement, with grouting pressure in MPa, drilling depth in m, and tension stress in kN; geological environment data and cost data are formatted according to the corresponding industry standards to form standardized data with consistent format.

[0041] Three cleaning operations are performed sequentially on the standardized data. The first is redundancy removal: duplicate construction record entries are deleted, data records with missing values ​​exceeding a set threshold are removed, and invalid or low-quality data are excluded from subsequent fusion processes. The second is anomaly correction: for missing data collected by sensors, the mean of valid readings within the nearest time window of the same sensor is used to fill in the gaps; for other types of missing or outliers, a trend fitting method is used, employing a Kalman filter to establish a state-space model of the sensor time-series data, replacing outlier readings with predicted values. The correction operation and the values ​​before and after correction are recorded for traceability and verification. The third is filtering and denoising: for short-term fluctuations in grouting pressure caused by wind interference at high-altitude construction sites and sensor drift caused by temperature differences, a low-pass filtering algorithm is used to denoise the sensor time-series data. Specifically, a moving average filter or a Butterworth low-pass filter can be used, with the cutoff frequency determined during the on-site calibration stage based on the signal frequency characteristics of various sensors. After these three cleaning operations, cleaned data that meets the fusion quality requirements is output, with data integrity of no less than 99%.

[0042] like Figure 2 As shown, in one embodiment of the present invention, step S4 includes: A multi-source data fusion model is constructed, which is executed according to three layers of logic: bottom-level fusion, middle-level fusion, and top-level fusion, and is equipped with an adaptive adjustment mechanism for fusion weights.

[0043] Initial fusion weights of each data source Based on a comprehensive calibration process considering sensor accuracy level, historical data integrity, and on-site calibration records, data sources with high accuracy levels and good historical integrity (such as real-time data from calibrated sensors) are assigned higher initial weights, while manually entered data (such as on-site construction logs) are assigned relatively lower initial weights. Subsequently, the weights of each data source are further optimized using the fusion quality feedback coefficients output from the dual-path reverse optimization closed-loop in step S6. Dynamic adjustments are made, and the fusion weight of data sources with larger historical deviations is automatically reduced.

[0044] The underlying fusion compares and weights multi-source data with the same parameter, and the fused value is... for: ; in, The measured value of this parameter is given by data source s. For the corresponding fusion weights.

[0045] Taking this embodiment as an example: sensor data of anchor cable tension stress (weight) ) and on-site measured data (weight) The weighted fusion yielded 1502kN, which was used to replace a single data source to improve data accuracy; similarly, the grouting parameters were fused at the bottom layer to obtain fused data with a grouting pressure of 0.54MPa (after compensation) and a grouting volume of 931L.

[0046] The mid-layer fusion uses the four-level parameter codes established in step S1 as a unified index key to perform cross-dimensional association matching: using the anchor cable number (e.g., "0001") as the primary key, it retrieves all parameter type codes corresponding to the anchor cable (SG construction parameters, ZL quality parameters, ZJ cost parameters, GD archiving parameters). The construction parameters (grouting pressure 0.54MPa, grouting volume 931L), quality parameters (grouting density 96%, borehole verticality deviation no greater than 1.2%), and cost parameters (grouting material consumption 1000L, anchor cable material 60m) after the bottom layer fusion are associated and mounted with the geological environment data (air pressure 63.6kPa, temperature difference 15℃) in the same data record through the code index, establishing a cross-dimensional complete data record centered on a single anchor cable.

[0047] The top-level fusion records the cross-dimensional complete data after the middle-level association, and maps it to the corresponding component nodes in the digital twin model constructed in step S2 according to the parameter encoding index. Various parameter values ​​are written into the construction status attribute, quality status attribute and cost attribute fields of the corresponding component, realizing the deep integration of fused data and twin model components, forming a twin working status carrying actual construction measurement data, and providing high-quality data input for parameter mapping and accuracy calibration in step S5.

[0048] The three-layer fusion architecture described above decouples data processing from model updates, and the underlying weight calibration mechanism enables the fusion results to adaptively respond to changes in the reliability of the data source. Dynamic weight adjustment based on Bayesian estimation can be introduced in the underlying fusion stage, using the sensor's factory accuracy as a priori and the degree of conformity between the current batch data and the historical average as the likelihood. This allows for the capture of weight change signals in the early stages of sensor performance degradation, reducing the cumulative effect of deviations across processes.

[0049] In a preferred embodiment, the underlying fusion stage assigns fusion weights to each data source. The dynamic adjustment adopts a Bayesian accuracy estimation mechanism based on conjugate priors, which is implemented as follows: The measurement accuracy of data source s is modeled as a random variable, with the accuracy parameter... (i.e., the reciprocal of the measurement variance) characterizes its reliability and... Assign a Gamma conjugate prior distribution: ; in, For shape parameters, For the scale parameter, the initial values ​​of both are... and The accuracy level of the sensor is determined by a combination of factors: the higher the factory accuracy level of the sensor, the better. The larger the ratio, the higher the prior expectation accuracy.

[0050] After each underlying fusion execution, the squared residual between the current measurement value of data source s and the mean of this fusion is used as the likelihood information to perform a single-step update on the posterior distribution parameters: ; ; in, Let be the measurement value of data source s in the t-th fusion. This is the average value of the underlying fusion. The squared residual of the data source in this measurement is the deviation from the fusion center, which directly quantifies the reliability of this measurement; the superscripts t and t+1 indicate the execution sequence number of the underlying fusion.

[0051] Precision parameters The posterior expected value is: ; Fusion weights Proportional to the posterior expected accuracy of each data source, and obtained after normalization: ; The aforementioned mechanism captures the deviation of each data source in real time through squared residuals during each underlying fusion process. This allows the fusion weights to be continuously iterated and updated within the process, resulting in a higher response frequency to sensor performance degradation than the process-level updates of the dual-path reverse optimization closed loop. This allows for the reduction of corresponding data source weights in the early stages of sensor performance degradation, effectively minimizing the cross-process accumulation effect of deviations. This mechanism is related to the fusion quality feedback coefficient. To form complementarity: The system is responsible for adjusting the weights at the macro level between processes, while Bayesian precision estimation is responsible for updating the weights at the micro level within processes. Together, they ensure the accuracy and robustness of the underlying fusion results.

[0052] In one embodiment of the present invention, step S5 includes: Based on the parameter-correlation directed acyclic graph and two-level mapping function established in step S2, parameter mapping and accuracy calibration are performed on the fused data output in step S4.

[0053] First, the first-level environmental compensation transformation is performed on the source parameter nodes to convert the fused data into the compensated real physical quantities and write them into each source node (the compensation formula is the same as in step S2). Then, the system performs a breadth-first traversal along the directed graph, triggering the second-level mapping function of each derived parameter node in turn, calculating and writing all derived parameters, and simultaneously performing design constraint boundary verification. Nodes that exceed the constraint boundary are marked as abnormal states and early warning signals are generated.

[0054] Regarding accuracy calibration, the mapping output values ​​for each parameter node are... Use proportional correction factor With offset correction factor The constructed two-factor correction performs accuracy calibration, and the accuracy calibration value is... for: ; Adaptive update based on measured values Mapping output value deviation To drive, update according to preset step sizes. and : ; ; in, To adjust the learning rate for offset, The learning rate is a proportional adjustment, and all of them are positive hyperparameters that control the update step size of the adjustment factor.

[0055] Used to correct proportionality deviation. Used to correct constant deviations, both gradually converge as calibration data accumulates during process advancement, ensuring that the mapping accuracy error does not exceed 2% and the overall accuracy is not less than 98%.

[0056] Taking this embodiment as an example: the on-site measured grouting density was 96%, the twin mapping value was 95.8%, and the deviation was... The mapping was deemed successful, and the value was written after calibration. ; To address the proportional deviation of tensile stress caused by diurnal temperature variations at high altitudes Correction, constant deviation from The correction and the synergistic effect of the two factors ensure that the overall accuracy continues to meet the indicator requirements.

[0057] The aforementioned two-factor accuracy calibration mechanism upgrades mapping accuracy control from single-parameter adjustment to a two-way collaborative correction of proportional and offset parameters, adapting to the characteristics of multiple error sources coexisting in high-altitude engineering. It can be used for... and An adaptive decay strategy is introduced, which gradually reduces the step size as the calibration rounds increase, similar to the learning rate decay in stochastic gradient descent. This prevents the correction factor from oscillating excessively due to occasional perturbations in the later stages of convergence, further improving the stability of accuracy calibration.

[0058] like Figure 3 As shown, in one embodiment of the present invention, step S6 includes: Taking the quality acceptance nodes of the seven key processes in anchor cable construction (drilling, anchor cable fabrication and installation, grouting, concrete anchor pier pouring, tensioning, anchor sealing, and unit evaluation) as the core triggering conditions, the incremental differential update mechanism driven by the process state machine is used to achieve real-time synchronization between the twin and the on-site construction status.

[0059] The process state machine includes states to be activated. Execution status Pending acceptance status Accepted status Status pending re-inspection Five states. Each process Initially in After the previous process is completed, proceed to After the construction operation is completed, proceed to After passing quality inspection, it will be transferred to If the inspection fails, it will be transferred to And trigger a status verification.

[0060] Specifically, This indicates the start of a process. The system initializes the status of the associated parameter nodes for this process to "pending update", triggering the module to start real-time monitoring of the quality acceptance status of this process. This indicates that the construction parameters have been collected. The triggering module captures the update trigger signal, and the data retrieval module retrieves the multi-source data fused by S4 according to the trigger signal, triggering the process-level incremental update. This indicates that the acceptance is qualified (e.g., for the grouting process: the quality inspection personnel enter that the grouting density is 96% ≥ 95%, which is qualified), triggering an overall incremental update + two-factor correction update + status snapshot saving; If the acceptance test fails or a constraint boundary warning is triggered (e.g., grouting density is below 95%), an early warning will be triggered and updates will be suspended. The status will be re-inspected and the warning will be retried after rectification. External signals (design changes or major changes in geological conditions) will trigger an overall emergency update, which will fully update the digital twin.

[0061] Incremental differential update mechanism calculates the mapped value after accuracy calibration. With the twin's current stored value The difference : ; Only when Write is executed at the time. To update the threshold, which is used to filter out small fluctuations caused by sensor noise.

[0062] Process-level incremental updates are provided by The conversion triggers an update that is limited to the set of source parameter nodes for the current process. After the incremental writing of the source node is completed, the breadth-first propagation is automatically performed along the S2 directed graph to update all downstream derived nodes level by level, and the process status is displayed synchronously. Overall incremental update by The conversion triggers an update that extends to all nodes in the current process and the set of strongly related nodes across processes. ; in, For the correlation strength threshold, only weights The cross-process strongly correlated parameter nodes are included in this update to achieve cross-process parameter linkage and synchronization, with an update delay of no more than 10 seconds. Taking one example: After the grouting process is completed, the quality acceptance personnel enter the acceptance result (grouting density 96%, qualified), triggering an overall incremental update. The data call module calls the fused grouting parameters (grouting pressure 0.54MPa, grouting volume 931L, grouting density 96%). The parameter update module updates the above parameters synchronously to the grouting process attributes of the corresponding anchor cable in the twin model, and synchronously displays the grouting completion status. The verification module compares the on-site measured data with the updated data. The deviation is 0.2%, and the update is judged to be qualified. The feedback module feeds back the update result (grouting process completed, parameters updated) to the control platform. After the subsequent tensioning process is completed and accepted, the update of the corresponding parameters in the twin is triggered, realizing process-level hierarchical updates.

[0063] Save a full parameter state snapshot after each overall update. Those that fail the inspection will be allowed to enter. Restore to the previous process snapshot To ensure the traceability of twin data, the update delay is no more than 10 seconds.

[0064] Perform accuracy verification on the updated parameters (allowable deviation ≤ 2%), and if the deviation exceeds the threshold, calculate the deviation by the order of magnitude. Implement a three-level correction response: Level 1 (minor, ) Perform automatic correction, directly calling the updated two-factor correction pair in S5. Recalculate calibration values ​​for nodes exceeding the threshold. After the correction is complete, re-enter the verification module. If approved, the response is upgraded to Level 2; otherwise, it is upgraded to Level 2 (moderate). The system performs auxiliary corrections and manual confirmations. It traces the deviation source backward along the S2 directed graph, locating the source parameter node from the deviation node, generating a deviation tracing report, and pushing it to the control platform. The correction plan is then executed after manual confirmation, ensuring the correction process is traceable. Level 3 (Severe, Triggering the process state machine to enter The twin is restored to its previous state snapshot. At the same time, lock all parameter nodes of this process ( (The update process will be unlocked and retried after manual intervention and investigation.)

[0065] For each incremental update operation, a structured update log is recorded. The log fields include: process number. Trigger timestamp Trigger type (process level / overall level / emergency), update node list and incremental values ​​of each node Verification deviation Correction level (no correction / Level 1 / Level 2 / Level 3), final write value Operator ID. Logs are persistently stored on the management platform, supporting multi-dimensional retrieval by process number, time range, and parameter code, providing complete digital records for construction quality traceability, responsibility identification, and subsequent project experience accumulation. The feedback module pushes each update and verification result (including deviation information and correction status) to the management platform in real time, achieving closed-loop management.

[0066] After the overall update is completed, a dual-path reverse optimization loop is executed simultaneously. The first path uses the deviation generated during each acceptance test. As input, weighted attenuation coefficient Association strength weights for corresponding directed edges in a directed acyclic graph with parameter associations. Perform a decay update: ; , This is the absolute value of the deviation in this verification. The larger the deviation, the weaker the association of that edge will be. Cross-process strong association filtering during overall-level updates ( Based on this, the correlation effectiveness of each parameter pair will be reassessed, and the mapping structure will continue to self-optimize as the construction progresses.

[0067] The second approach introduces a fusion quality feedback coefficient. Record the average deviation of the data source s over the last K acceptance tests. Adjust the step size according to the fusion weight. Perform the update: ; Weighting of data sources during underlying fusion Proportional to the present Data sources with consistently large historical deviations have their weights automatically reduced, while data sources with stable historical performance have their weights relatively increased. This enables dynamic assessment of data source reliability and adaptive redistribution of fusion weights, where the superscripts t and t+1 represent the acceptance round number.

[0068] Two reverse optimization signals drive the bidirectional convergence of the parameter-correlated directed acyclic graph mapping structure and the underlying fusion weights, ensuring that the overall mapping accuracy remains at an error level of no more than 2% and data integrity of no less than 99%.

[0069] The coupling mechanism between the aforementioned process state machine and incremental differential update enables the twin to automatically synchronize to the latest construction state at each process acceptance node, while the state snapshot mechanism retains complete backtracking capabilities. It also allows for the adjustment of correlation strength weights. Setting lower bound constraints prevents physically related parameters from approaching zero due to continuous deviations, maintains the physical structural integrity of the mapping graph during the backpropagation process, and avoids excessive decay that could cause the overall update path to break.

[0070] In a preferred embodiment, the lower limit constraint of the correlation strength weight is not a fixed constant, but is dynamically determined based on the stability of each parameter against the historical deviation sequence, as specifically implemented as follows.

[0071] For parameter-associated directed edges in a directed acyclic graph Record its corresponding parameter node The acceptance deviation sequence generated in the last M global level updates is as follows , where M is the sliding window length, which is a positive integer. The mean deviation of this sequence is defined as: ; Define the deviation stability coefficient The normalized standard deviation of this sequence is: ; in, To prevent the introduction of small positive numbers with a denominator of zero. The smaller the value, the more stable the deviation sequence of the last M times, and the higher the consistency of the mapping relationship between this parameter pair; The larger the value, the more drastic the deviation fluctuations and the lower the stability of the mapping relationship.

[0072] Based on the deviation stability coefficient, for directed edges Lower bound of association strength weight Defined as: ; in, For directed edges Based on the pre-set structural lower limit benchmark value according to the physical correlation between parameters, the physical transmission path marked in step S1 is used to determine the parameter pairs that are physically strongly correlated (such as grouting pressure and grouting density), and the parameter pairs that are weakly correlated across processes, take the larger value. This is a stability sensitivity coefficient, controlling the degree to which deviation fluctuations affect the lower limit of contraction. When the deviation sequence is extremely stable... The lower limit approaches the physical structural benchmark value. This fully protects strongly correlated edges from decaying to ineffectiveness; when the deviation fluctuates significantly, The lower limit is increased while the lower limit is moderately contracted, leaving sufficient room for adjustment in the attenuation mechanism.

[0073] Embedding the above adaptive lower bound into the first-path association strength weight decay update rule, we obtain the update formula with lower bound protection: ; in, For the directed edge during the t-th round of acceptance The correlation strength weight, This is the weight decay coefficient. This represents the absolute value of the deviation during this acceptance test. This is the current adaptive lower limit calculated using the above formula. Lower limit protection applies to cases with continuous deviation. The dynamic safety threshold, determined jointly by the physical structure and deviation stability, must be maintained at least as high as possible. During the backpropagation process, the physical structure integrity of the directed acyclic graph of parameter associations must be preserved to prevent the overall update path from breaking due to excessive decay. Deviation stability coefficient. The required historical deviation sequence is derived from the deviation field already recorded in the structured update log.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a digital twin of a hydropower high slope anchor cable support construction, characterized in that, include: S1. Organize all kinds of construction parameters in the entire process of anchor cable support construction, use multi-level coding rules to uniquely identify each parameter, and bind the parameter code with the attribute fields of physical entities and digital twin components in sync to establish a one-to-one correspondence between physical entities and digital twins. S2. Construct a directed acyclic graph of parameter association with each parameter as a node and the driving relationship between parameters as a directed edge. The nodes are divided into source parameter nodes and derived parameter nodes. The source parameter nodes store the sensor readings after correction by the environmental compensation coefficient. The derived parameter nodes are calculated sequentially by the mapping function corresponding to each node, and finally form the mapping parameter values ​​covering all parameters. S3. Perform anomaly correction and filtering / denoising processing on the multi-source heterogeneous data collected during the construction process to obtain standardized data. S4. Using parameter encoding as the index key, a hierarchical weighted fusion method is used to perform cross-dimensional association fusion of standardized data, and the fusion result is written into the attribute field of the corresponding node in the parameter-associative directed acyclic graph. S5. Perform parameter mapping on the fusion result based on the parameter-correlated directed acyclic graph, use the proportional correction coefficient and the offset correction coefficient to calibrate the accuracy of the mapped output value, and adaptively update the proportional correction coefficient and the offset correction coefficient according to the deviation between the measured value and the mapped output value to obtain the accuracy-calibrated mapped parameter value. S6. Using the key process quality acceptance node as the trigger condition, the incremental differential update mechanism driven by the process state machine is adopted to propagate the accuracy-calibrated mapping parameter values ​​along the parameter-related directed acyclic graph and write them into the digital twin. When the acceptance fails, the state re-verification is triggered to obtain a digital twin that is synchronized with the on-site construction status in real time.

2. The method of claim 1, wherein, The multi-level coding rule mentioned in step S1 is a four-level coding rule. The coding structure is as follows: project code, slope zone code, anchor cable number, and parameter type code. The parameter type code distinguishes parameters into five categories: design parameters, construction parameters, quality parameters, cost parameters, and archived parameters. The codes at each level are concatenated to form a unique identifier for all parameters of each anchor cable.

3. The method according to claim 1, characterized in that, The sensor readings stored by the source parameter node in step S2 are converted into compensated real physical quantities by a two-factor environmental compensation The compensation formula is: ； wherein, is the sensor raw reading; Pc = pressure correction factor, defined as: ； for standard atmospheric pressure, for field measured air pressure; T is the temperature difference correction coefficient, defined as: ； to correspond to a temperature coefficient of the sensor, to correspond to an on-site diurnal temperature difference; compensated Write Source Parameter Node Property field.

4. The method according to claim 3, characterized in that, In step S2, the mapping function of the derived parameter node takes the compensation value of the source parameter node as input, performs a breadth-first traversal on the directed acyclic graph of parameter association, and sequentially triggers the calculation of the mapping function of each derived parameter node and writes the derived parameter value. Each directed edge... The general form of the mapping function is: ； in, For derived parameter values, The current stored value, As the baseline value for derived parameters, Let k be the field calibration coefficient. Let K be the k-th basis function, where k = 1, 2, ..., K, K is the order of the basis function expansion, and k represents the index of the basis function expansion term; After the output value of the mapping function for each derived parameter node is written, the design constraint boundary check is performed immediately: ； in, For derived parameter nodes The current mapping value, and These are the lower and upper design limits bound to the node, respectively. When a node's status is abnormal, an early warning signal is generated and transmitted.

5. The method according to claim 1, characterized in that, The multi-source heterogeneous data mentioned in step S3 is divided into five categories of collection sources: Design data is imported all at once during the design phase. Real-time sensor data is automatically collected through the field sensor network. On-site construction and quality data, including construction logs, process acceptance records, and quality inspection reports, are entered by on-site personnel. Geological and environmental data are collected through geological survey and environmental monitoring equipment; Cost accounting data is uploaded regularly through the cost management system; The above five types of data are processed for format normalization and unit of measurement to form standardized data with consistent format. The anomaly correction is performed to fill in and remove missing values ​​and outliers that exceed the reasonable range. The filtering and denoising uses filtering algorithms to suppress noise and drift in sensor signals in complex high-altitude environments.

6. The method according to claim 5, characterized in that, The filtering and noise reduction described in step S3 targets the pressure jump data caused by wind interference and the sensor drift data caused by day-night temperature difference in the real-time sensor data. The filtering algorithm is used to perform suppression processing to ensure that the integrity of the cleaned data is not less than 99%.

7. The method according to claim 2, characterized in that, The layered weighted fusion method described in step S4 is executed in three layers: bottom-layer fusion, middle-layer fusion, and top-layer fusion. The underlying fusion process applies the same parameter to multi-source data according to the fusion weights of each data source. Perform weighted fusion, fusion value for: ； in, The measured value of this parameter is given by data source s. To correspond to the fusion weights, the initial fusion weights for each data source are comprehensively calibrated based on sensor accuracy levels, historical data integrity, and on-site calibration records, and are determined through a fusion quality feedback coefficient. Dynamic adjustment; The middle layer fusion uses the four-level parameter code as a unified index key to perform cross-dimensional association matching, and associates and mounts the construction parameters, quality parameters, cost parameters and geological environment data after the bottom layer fusion to the same anchor cable data record. The top-level fusion writes the cross-dimensional complete data records after the middle-level association into the attribute fields of the corresponding component nodes of the digital twin model according to the parameter encoding index.

8. The method according to claim 7, characterized in that, The accuracy calibration described in step S5 uses a proportional correction factor. With offset correction factor Constructing a two-factor correction pair, for the parameter nodes Mapping output value Perform calibration, accuracy calibration value for: ； The adaptive update is based on the deviation between the measured value and the mapped output value. To drive, adjust according to preset step size and Update them separately to ensure that the mapping accuracy is no less than 98%; Used to correct proportionality deviation. Used to correct constant deviations, both gradually converge as calibration data accumulates during process advancement.

9. The method according to claim 1, characterized in that, The process state machine described in step S6 includes states to be activated. Execution status Pending acceptance status Accepted status Status pending re-inspection Five states; The incremental differential update mechanism calculates the mapped value after accuracy calibration. With the twin's current stored value The difference : ； Only when Write is executed at the time. To update the threshold, which is used to filter out minute fluctuations caused by sensor noise; Process-level updates are by Triggered, the update scope is limited to the current process source parameter node set and propagates breadth-first along the directed acyclic graph to downstream derived parameter nodes; Overall update by Triggered, the update scope is expanded to all nodes in the current process and their associated strength weights. Not lower than the threshold Strongly interconnected nodes across processes; Save a full parameter state snapshot after each overall update. Those that fail the inspection will be allowed to enter. Restore to the previous process snapshot .

10. The method according to claim 8, characterized in that, It also includes dual-path reverse optimization closed loop: The first approach is based on the deviations observed during each acceptance test. As input, weighted attenuation coefficient Association strength weights for corresponding directed edges in a directed acyclic graph with parameter associations. Perform decay update: ； The second approach introduces a fusion quality feedback coefficient. Record the average deviation of the data source s over the last K acceptance tests. Adjust the step size according to the fusion weight. Perform the update: ； Weighting of data sources during underlying fusion Proportional to the present Data sources with consistently large historical biases will have their weights automatically reduced. Two reverse optimization signals drive the mapping structure of the parameter-correlated directed acyclic graph and the underlying fusion weights to converge bidirectionally, so that the overall mapping accuracy is maintained at an error of no more than 2% and the data integrity is no less than 99%. The superscripts t and t+1 represent the acceptance round number.

Images