An edge-computing-based deep foundation pit construction intelligent early warning monitoring method

Abstract

The application discloses an intelligent early warning and monitoring method for deep foundation pit construction based on edge computing and belongs to the technical field of edge computing. The method comprises the following steps: receiving monitoring data at an edge node arranged at a deep foundation pit construction site and generating a construction stage context; extracting data quality fingerprints, dividing the monitoring data into trusted data, repairable missing data, repairable abnormal data and data to be retested; generating a first candidate repair value, a second candidate repair value and a third candidate repair value according to the repairable missing data, the repairable abnormal data and the data to be retested; obtaining quality constraint risk values according to the first candidate repair value, the second candidate repair value and the third candidate repair value respectively; and outputting formal early warnings, review early warnings, local retests or observation events according to the quality constraint risk values. The application selects corresponding repair channels and generates candidate repair values for different types of monitoring data, thereby improving the pertinence of missing data and abnormal data repair.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to an intelligent early warning and monitoring method for deep foundation pit construction based on edge computing. Background Technology

[0002] Deep foundation pit construction typically refers to the deep excavation and support work carried out in building construction, municipal engineering, rail engineering, or underground space development to create underground structural construction space. This process is usually accompanied by changes in the stress on the retaining structure, changes in water levels inside and outside the pit, displacement of the surrounding soil, surface settlement, and changes in construction procedures. It is characterized by rapid changes in working conditions, numerous influencing factors, and a long risk transmission chain. Therefore, it is necessary to continuously collect, analyze, and provide early warnings for key monitoring data during the construction process to reduce the probability of instability, excessive deformation, or secondary safety accidents.

[0003] In existing technologies, monitoring points are typically set up at the deep foundation pit construction site to collect monitoring data such as displacement, settlement, water level, support status, and equipment status. The on-site nodes or back-end platform then generate construction stage information based on the monitoring data, and perform quality assessment, classification, and correction processing on the monitoring data. After the monitoring data processing is completed, the risk status, risk probability, and early warning results are determined based on the corrected monitoring data, confidence information, and preset risk model, thereby outputting formal early warnings, verification early warnings, partial retests, or observation events, and adjusting the subsequent monitoring process.

[0004] However, existing technologies still have the following shortcomings: First, existing solutions mostly process monitoring data uniformly after collection, lacking synchronous constraints on changes in construction stages, communication status, and regional differences. This results in a weak connection between data quality judgment and current construction conditions, easily leading to the problem of mixed judgment of missing and abnormal data. Second, existing solutions do not clearly distinguish the processing paths for repairable missing data, repairable abnormal data, and data to be retested, easily incorporating data problems of different causes into similar correction processes, reducing the specificity of correction results. Third, existing solutions often focus more on the correction results themselves when making early warning judgments, underutilizing the credibility of candidate repair values ​​obtained from different repair channels, resulting in insufficient reflection of data quality differences in risk output.

[0005] Therefore, there is an urgent need to provide an intelligent early warning and monitoring method for deep foundation pit construction based on edge computing to solve the above problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide an intelligent early warning and monitoring method for deep foundation pit construction based on edge computing.

[0007] To solve the above-mentioned technical problems, one technical solution adopted by the present invention is: to provide an intelligent early warning and monitoring method for deep foundation pit construction based on edge computing, comprising the following steps: The edge node deployed at the deep foundation pit construction site receives monitoring data and generates a construction phase context based on the monitoring data; Data quality fingerprints are extracted from the monitoring data and the construction phase context, and the monitoring data is divided into reliable data, repairable missing data, repairable abnormal data, and data to be retested based on the data quality fingerprints. Based on the repairable missing data, the repairable abnormal data, and the data to be retested, a preset repair channel is selected to generate a first candidate repair value, a second candidate repair value, and a third candidate repair value. A first confidence level, a second confidence level, and a third confidence level are determined based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. The first candidate repair value, the second candidate repair value, and the third candidate repair value are then input together with the first confidence level, the second confidence level, and the third confidence level into a preset early warning model to obtain a quality constraint risk value. Based on the quality constraint risk value and the first confidence level, the second confidence level, and the third confidence level, output formal warnings, review warnings, partial retests, or observation events, and update the construction phase context of subsequent monitoring data based on the formal warnings, review warnings, partial retests, or observation events.

[0008] The present invention is further configured such that: the edge node is deployed at a preset distance from the monitoring data in the deep foundation pit construction site; The edge node is used to perform the following processing: receiving the monitoring data, generating the construction stage context based on the monitoring data, extracting the data quality fingerprint, classifying the trusted data, the repairable missing data, the repairable abnormal data, and the data to be retested based on the data quality fingerprint, generating the first candidate repair value, the second candidate repair value, and the third candidate repair value based on the repairable missing data, the repairable abnormal data, and the data to be retested, determining the first confidence level, the second confidence level, and the third confidence level, and outputting the formal warning, the review warning, the partial retest, or the observation event.

[0009] The present invention is further configured such that the step of generating the construction stage context is: Construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information are extracted from the monitoring data, and a current operation record is generated based on the construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information. The current area, current level, current support status, current drainage status, and current communication status are determined based on the current operation record, and a stage feature record is generated based on the current area, current level, current support status, current drainage status, and current communication status. The construction stage label is determined by matching the stage feature record with the preset stage rules, and a transition state record is generated based on the construction stage label, the current area, and the construction stage context of the previous moment. The construction stage label is maintained, switched, or delayed based on the transition state record, and the construction stage context at the current moment is generated based on the updated construction stage label, the current region, the current communication state, and the construction stage context of the previous moment.

[0010] The present invention is further configured such that the data quality fingerprint extraction steps are as follows: Within a preset short-term analysis window, the monitoring data is grouped according to the reception time. Within the same construction stage defined by the construction stage context, the completeness, numerical stability, and signal clarity of each group of monitoring data are statistically analyzed. The completeness, numerical stability, and signal clarity are then combined in a preset order to obtain an initial quality vector. Each quality value in the initial quality vector is compared with the corresponding benchmark value in the preset quality benchmark corresponding to the construction stage context to obtain the benchmark deviation of each quality value. Then, according to the area correction value corresponding to the current area and the communication correction value corresponding to the current communication state in the construction stage context, the benchmark deviation of each quality value is corrected. The corrected quality values ​​are written into the corresponding positions in the preset order to obtain the context correction quality vector. The quality values ​​in the context-corrected quality vector are arranged from high to low to obtain the current arrangement order. The current arrangement order is then compared with the arrangement order of the previous analysis window, and the order difference between two adjacent quality values ​​in the current arrangement order and the overall offset of the overall average value of the context-corrected quality vector relative to the preset quality benchmark are determined. When the order difference exceeds the preset order condition or the overall offset exceeds the preset offset condition, a vector fluctuation mark is generated. The current arrangement order, the vector fluctuation mark, and the context-corrected quality vector are written into the same vector structure according to a preset field order to form a quality feature vector, and the quality feature vector is then processed to unify it. The unified quality feature vector is compared with a plurality of preset quality pattern templates one by one to determine the template proximity value between each quality pattern template and the quality feature vector. The target template proximity value with the largest value that is greater than a preset proximity threshold is selected from the template proximity values. The unique identifier of the quality pattern template corresponding to the target template proximity value is combined with the quality feature vector to obtain the data quality fingerprint.

[0011] The present invention is further configured such that: the specific content of dividing the monitoring data into reliable data, repairable missing data, repairable abnormal data, and data to be retested based on the data quality fingerprint is as follows: Read the unique identifier, the template proximity value, and the quality feature vector from the data quality fingerprint to determine the target quality pattern corresponding to the monitoring data; Based on the complete state, continuous state, adjacent state and associated state in the target quality pattern, the corresponding fields in the quality feature vector are compared item by item to determine the state with the largest deviation value as the dominant mismatch item, and the number of mismatches in the remaining states whose deviation values ​​exceed the preset deviation threshold is counted. When the dominant mismatch item does not exist and the template proximity value is greater than the preset confidence threshold, the monitoring data is determined as the confidence data; When the dominant mismatch item is the complete state, and the deviation values ​​corresponding to the continuous state, the adjacent state, and the associated state are all not greater than the preset deviation threshold, the monitoring data is determined as the repairable missing data. When the dominant mismatch item is the continuous state, the adjacent state, or the associated state, and the deviation value corresponding to the complete state is not greater than the preset deviation threshold, the monitoring data is determined as the repairable abnormal data. When the template proximity value is not greater than the preset confidence threshold, or the number of mismatches is greater than the preset number threshold, the monitoring data is determined as the data to be retested.

[0012] The present invention is further configured such that the preset repair channel includes a first repair channel, a second repair channel, and a third repair channel; The first repair channel is used to perform historical record extraction, stage record filtering, and trend continuation processing on the repairable missing data to generate the first candidate repair value; The second repair channel is used to perform synchronous record extraction, adjacent record filtering and difference correction processing on the repairable missing data or the repairable abnormal data to generate the second candidate repair value; The third repair channel is used to perform corresponding record extraction, associated record filtering and offset conversion processing on the repairable abnormal data or the data to be retested, and generate the third candidate repair value.

[0013] The present invention is further configured such that the steps for generating the first candidate repair value, the second candidate repair value, and the third candidate repair value are as follows: The first repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data, retrieves the historical records of the same monitoring location in the same construction stage arranged in reverse order of reception time, selects the first historical record, the second historical record, and the third historical record that are closest to the current time, calculates the first change value between the first historical record and the second historical record, and calculates the second change value between the second historical record and the third historical record; When the first change value and the second change value change in the same direction, compare the first time interval between the first historical record and the current time and the second time interval between the second historical record and the current time to determine the main change value corresponding to the smaller time interval. Then compare the difference between the first change value and the second change value. When the difference is not greater than a preset proximity value, take the middle value between the first change value and the second change value as the continuing change value. When the difference is greater than the preset proximity value, take the main change value as the continuing change value. When the first change value and the second change value do not change in the same direction, the change value with the smaller absolute value is taken as the continued change value, and the continued change value is superimposed on the first historical record to obtain the first candidate repair value; The second repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data or the repairable abnormal data, retrieves the synchronous records of adjacent positions within the same construction stage and the synchronous records of the adjacent positions at the previous moment, calculates the position difference of each synchronous record relative to the current monitoring data, calculates the offset difference of each synchronous record relative to the current monitoring data, retains the synchronous records whose change direction is consistent with the current monitoring data and whose offset difference is less than the position difference, averages the offset differences corresponding to the retained synchronous records to obtain the adjacent position correction value, compares the adjacent position correction value with a preset adjacent position correction interval, when the adjacent position correction value falls within the preset adjacent position correction interval, the adjacent position correction value is retained, when the adjacent position correction value exceeds the preset adjacent position correction interval, the interval boundary value consistent with the direction of the adjacent position correction value is truncated as the target adjacent position correction value, if the target adjacent position correction value is positive, the target adjacent position correction value is added to the current monitoring data, if the target adjacent position correction value is negative, the absolute value of the target adjacent position correction value is subtracted from the current monitoring data to obtain the second candidate repair value; The third repair channel reads the current monitoring data corresponding to the repairable abnormal data or the data to be retested, the target quality mode, and the current construction stage context. It retrieves the current corresponding record and the previous corresponding record of the related project within the same construction stage, calculates the correlation change value between the current corresponding record and the previous corresponding record, calculates the historical correspondence difference between the current monitoring data and the previous corresponding record, and superimposes the correlation change value and the historical correspondence difference to form a correlation conversion value. The correlation conversion value is compared with a preset correlation correction interval. When the correlation conversion value falls within the preset correlation correction interval, the correlation conversion value is retained. When the correlation conversion value exceeds the preset correlation correction interval, the interval boundary value with the same direction as the correlation conversion value is truncated as the target correlation conversion value. If the target correlation conversion value is positive, the target correlation conversion value is added to the current monitoring data. If the target correlation conversion value is negative, the absolute value of the target correlation conversion value is subtracted from the current monitoring data to obtain the third candidate repair value.

[0014] The present invention is further configured such that the methods for determining the first confidence level, the second confidence level, and the third confidence level are as follows: Read the first historical record, the second historical record, the third historical record, and the current construction stage context corresponding to the first candidate repair value; calculate the first repair difference between the first candidate repair value and the first historical record; calculate the first amplitude difference between the first change value and the second change value; calculate the first time difference between the first time interval and the second time interval; and assign the first repair difference, the first amplitude difference, and the first time difference to a first confidence order in ascending order; and determine the first confidence level based on the first confidence order. Read the synchronization record, the neighbor correction value, the target neighbor correction value, and the current construction stage context corresponding to the second candidate repair value; calculate the second offset difference between the second candidate repair value and each synchronization record; count the number of valid synchronization records that are consistent with the change direction of the second candidate repair value; calculate the second correction difference between the neighbor correction value and the target neighbor correction value; and assign the second offset difference, the number of valid synchronization records, and the second correction difference to a second confidence order from best to worst; and determine the second confidence level based on the second confidence order. Read the current corresponding record, the previous corresponding record, the associated change value, the historical corresponding difference, and the target quality pattern corresponding to the third candidate repair value; calculate the third corresponding difference between the third candidate repair value and the current corresponding record; compare the directional consistency between the associated change value and the historical corresponding difference; determine the matching length of the third candidate repair value falling into the state interval corresponding to the target quality pattern; and assign the third corresponding difference, the directional consistency, and the matching length to the third confidence order from best to worst; and determine the third confidence level based on the third confidence order.

[0015] The present invention is further configured such that: the preset early warning model is specifically a Bayesian network model. The Bayesian network model receives the first candidate repair value, the second candidate repair value, the third candidate repair value, and the first confidence level, the second confidence level, and the third confidence level. First, it determines the first risk state, the second risk state, and the third risk state based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. Then, it performs state correction on the first risk state, the second risk state, and the third risk state based on the first confidence level, the second confidence level, and the third confidence level to obtain the corresponding first risk probability, the second risk probability, and the third risk probability. The maximum value among the first risk probability, the second risk probability, and the third risk probability is determined as the quality constraint risk value.

[0016] The present invention is further configured such that the output steps for the formal early warning, the review early warning, the partial retest, or the observation event are as follows: Read the quality constraint risk value, the first risk probability, the second risk probability, the third risk probability, the first confidence level, the second confidence level, and the third confidence level; determine the target risk probability and target confidence level corresponding to the quality constraint risk value; and determine the target risk state corresponding to the target risk probability. The target risk probability is compared with a preset formal risk threshold, a preset review risk threshold, and a preset observation risk threshold, and the target confidence level is compared with a preset high confidence threshold and a preset medium confidence threshold to obtain a formal judgment result, a review judgment result, a retest judgment result, or an observation judgment result. When the formal judgment result is true, the formal warning is output and a formal record is generated. When the review judgment result is true, the review warning is output and a review record is generated. When the retest judgment result is true, the local retest is output and a retest record is generated. When the observation judgment result is true, the observation event is output and an observation record is generated. Based on the official record, the review record, the retest record, or the observation record, the corresponding output area, output duration, and subsequent acquisition frequency are determined respectively, and the output area, the output duration, and the subsequent acquisition frequency are written into the construction stage context corresponding to the subsequent monitoring data.

[0017] The beneficial effects of this invention are as follows: 1. This invention improves the accuracy of monitoring data identification in deep foundation pit construction by extracting data quality fingerprints and classifying monitoring data types by combining the context of the construction stage; 2. This invention improves the targeting of missing and abnormal data repair by selecting corresponding repair channels for different types of monitoring data and generating candidate repair values; 3. This invention improves the reliability of formal early warning, review early warning, and local retest outputs by inputting candidate repair values ​​and their confidence levels into the early warning model for determination. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0020] Please see Figure 1A method for intelligent early warning and monitoring of deep foundation pit construction based on edge computing includes the following steps: The edge nodes deployed at the deep foundation pit construction site receive monitoring data and generate construction phase context based on the monitoring data; Data quality fingerprints are extracted from monitoring data and construction phase context, and the monitoring data are divided into reliable data, repairable missing data, repairable abnormal data, and data to be retested based on the data quality fingerprints. Based on the repairable missing data, repairable abnormal data, and data to be retested, a preset repair channel is selected to generate the first candidate repair value, the second candidate repair value, and the third candidate repair value. The first confidence level, the second confidence level, and the third confidence level are determined based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. The first candidate repair value, the second candidate repair value, and the third candidate repair value, together with the first confidence level, the second confidence level, and the third confidence level, are input into the preset early warning model to obtain the quality constraint risk value. Based on the quality constraint risk value and the first, second, and third confidence levels, output formal warnings, review warnings, partial retests, or observation events, and update the construction phase context of subsequent monitoring data based on the formal warnings, review warnings, partial retests, or observation events.

[0021] This method generates a construction phase context at the edge node and extracts data quality fingerprints from the monitoring data based on the construction phase context, making the classification of monitoring data more consistent with the construction status. Then, it generates a first candidate repair value, a second candidate repair value, and a third candidate repair value through a preset repair channel, and determines the quality constraint risk value by combining the first confidence level, the second confidence level, and the third confidence level, thereby improving the pertinence and reliability of formal early warning, review early warning, local retesting, or observation events.

[0022] Edge nodes are deployed at a predetermined distance from the monitoring data at the deep foundation pit construction site; Preset distance: refers to the deployment distance between the edge node and the monitoring data source measured along the deployment path on the construction site. The preset distance is used to ensure that the edge node can balance transmission latency, on-site coverage and maintenance safety when receiving monitoring data. The preset distance is set based on the current distribution density of monitoring points in the area, the stable transmission range corresponding to the current communication status, the edge node processing cycle, and the avoidance requirements of construction vehicles and support components. When the distribution density of monitoring points is high and the current communication status is weak, the preset distance is reduced; when the distribution density of monitoring points is low and the current communication status is stable, the preset distance is increased.

[0023] Edge nodes are used to perform the following processes: receive monitoring data, generate construction phase context based on monitoring data, extract data quality fingerprints, classify data into trusted data, repairable missing data, repairable abnormal data, and data to be retested based on data quality fingerprints, generate first candidate repair values, second candidate repair values, and third candidate repair values ​​based on repairable missing data, repairable abnormal data, and data to be retested, determine first confidence level, second confidence level, and third confidence level, and output formal warning, review warning, partial retest, or observation event.

[0024] The edge node is specifically a field processing terminal deployed near the deep foundation pit construction site. The field processing terminal includes a data access module, a stage context generation module, a data quality fingerprint extraction module, a data classification module, a repair channel execution module, a confidence level determination module, and an early warning output module. The data access module is used to receive monitoring data, the stage context generation module is used to generate construction stage context based on monitoring data, the repair channel execution module is used to execute the first repair channel, the second repair channel, and the third repair channel, and the early warning output module is used to output formal early warnings, verification early warnings, local retests, or observation events.

[0025] One embodiment of the present invention is as follows: the steps for generating the construction phase context are: Extract construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information from the monitoring data, and generate the current operation record based on the construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information; Monitoring data is collected through displacement acquisition devices, settlement acquisition devices, water level acquisition devices, support status acquisition devices, and equipment status acquisition devices set at various monitoring locations within the current area. Displacement acquisition devices are used to collect displacement data, settlement acquisition devices are used to collect settlement data, water level acquisition devices are used to collect water level data, support status acquisition devices are used to collect support change information, and equipment status acquisition devices are used to collect equipment change information. After collection, all data are written to the on-site processing terminal according to a unified time base to form monitoring data that can be called in the context of the construction stage.

[0026] The current area, current level, current support status, current drainage status, and current communication status are determined based on the current work record, and a stage feature record is generated based on the current area, current level, current support status, current drainage status, and current communication status. The construction stage label is determined by matching the stage feature record with the preset stage rules, and a transition state record is generated based on the construction stage label, the current area, and the construction stage context of the previous moment. Preset stage rules: This is a set of rules used to map the current work record to the construction stage label. The rule set is based on the information of excavation level change, support change, drainage change, construction location information and equipment change, and is set based on the construction organization sequence, the order of support installation and the rhythm of drainage adjustment.

[0027] The steps for generating a transitional state record are as follows: Read the construction stage label, the current area, and the construction stage context of the previous moment. Determine whether the construction stage label is consistent with the stage label of the previous moment. If they are consistent, write the current area status into the hold field. If they are inconsistent, write the support change information, excavation change information, or drainage change information that caused the change into the switch field. Then, write the current communication status and the change start time into the transitional state record.

[0028] The construction phase label is maintained, switched, or delayed based on the transition state record, and the construction phase context at the current moment is generated based on the updated construction phase label, the current region, the current communication state, and the construction phase context of the previous moment.

[0029] One embodiment of the present invention is as follows: the steps for extracting data quality fingerprints are as follows: Within a preset short-term analysis window, the monitoring data is grouped according to the reception time. Within the same construction phase defined by the construction phase context, the completeness, numerical stability, and signal clarity of each group of monitoring data are statistically analyzed. The completeness, numerical stability, and signal clarity are then combined in a preset order to obtain the initial quality vector. The preset short-term analysis window refers to the continuous analysis period that is captured backward from the current moment. The analysis period is used to ensure that the monitoring data within the same construction stage are comparable. Its setting is based on the edge node processing cycle, the monitoring frequency in the current area, and the rate of change of construction status.

[0030] The statistical methods for monitoring the completeness, numerical stability, and signal clarity of the data are as follows: within a short analysis window, the completeness is obtained by the ratio of the number of records that should have been recorded to the number of records that were actually recorded; the numerical stability is obtained by the consistency of the amplitude and direction of the numerical fluctuations of adjacent records; and the signal clarity is obtained by the reverse of the number of data jumps, the number of repetitions, and the number of invalid records.

[0031] Preset order: This is the fixed order in which the integrity, numerical stability, and signal clarity are written into the initial quality vector. The fixed order is set based on the influence on subsequent quality judgment from strong to weak.

[0032] Each quality value in the initial quality vector is compared with the corresponding benchmark value in the preset quality benchmark corresponding to the construction stage context to obtain the benchmark deviation of each quality value. Then, based on the area correction value corresponding to the current area and the communication correction value corresponding to the current communication state in the construction stage context, the benchmark deviations are corrected and the corrected quality values ​​are written to the corresponding positions in a preset order to obtain the context correction quality vector. The corresponding benchmark values ​​in the preset quality benchmarks refer to the benchmark values ​​for completeness, numerical stability, and signal clarity that are preset for the construction stage label and the current area. These benchmark values ​​are set based on historical construction records, experience records of similar areas, and stable acquisition requirements at the current level.

[0033] The area correction value corresponding to the current area and the communication correction value corresponding to the current communication status: The area correction value is used to characterize the inherent differences in the current area caused by spatial location, support density and acquisition obstruction, and the communication correction value is used to characterize the impact of the current communication status on the quality of the monitoring data arrival.

[0034] The specific steps for correcting the deviation of each reference are as follows: First, subtract the corresponding reference value from each quality value to form the reference deviation. Then, determine the quality item to which the reference deviation belongs. If the quality item is strongly affected by the region, add the region correction value. If the quality item is strongly affected by communication, add the communication correction value. If both affect the quality item, add them in the order of region first and then communication to obtain the corrected quality values.

[0035] The quality values ​​in the context-corrected quality vector are arranged from highest to lowest to obtain the current arrangement order. The current arrangement order is then compared with the arrangement order of the previous analysis window, and the order difference between two adjacent quality values ​​in the current arrangement order and the overall offset of the overall average value of the context-corrected quality vector relative to the preset quality benchmark are determined. When the order difference exceeds the preset order condition or the overall offset exceeds the preset offset condition, a vector fluctuation mark is generated. Preset order condition: refers to the maximum range of order changes allowed when the current order is compared with the order in the previous analysis window. It is set based on the relatively stable change pattern of each quality item within the same construction stage.

[0036] Preset offset condition: refers to the maximum allowable overall offset range of the context-corrected quality vector relative to the preset quality benchmark. It is set based on the normal drift level of the current region and the current communication state.

[0037] The steps for generating vector fluctuation markers are as follows: first, compare the order difference between two adjacent quality values, then calculate the offset between the overall average value and the corresponding benchmark average value. When the order difference exceeds the preset order condition or the offset exceeds the preset offset condition, write the fluctuation field to indicate a changing state; otherwise, write the fluctuation field to indicate a stable state.

[0038] The current arrangement order, vector fluctuation marker, and context-corrected quality vector are written into the same vector structure according to the preset field order to form a quality feature vector, and the quality feature vector is then uniformly processed. Preset field order: refers to the order in which the current arrangement, vector fluctuation marker, and context-corrected quality vector are written into the quality feature vector. The field order is set in the manner of "sequential field - fluctuation field - numerical field", which is based on the principle of preserving the structural relationship first, then the fluctuation state, and finally the specific value.

[0039] The standardization process involves scaling each numerical field in the quality feature vector to the same dimensional range, transcribing the sequential fields using a fixed encoding method, and transcribing the fluctuation fields using a binary state, so that different fields can participate in subsequent template approximation calculations on the same comparison scale.

[0040] The unified quality feature vector is compared with multiple preset quality pattern templates one by one to determine the template proximity value between each quality pattern template and the quality feature vector. The target template proximity value with the largest value that is greater than the preset proximity threshold is selected from the template proximity values. The unique identifier of the quality pattern template corresponding to the target template proximity value is combined with the quality feature vector to obtain the data quality fingerprint.

[0041] Multiple preset quality model templates: These refer to a set of standard templates corresponding to the trustworthy state, the missing-dominant state, the anomaly-dominant state, and the retest-dominant state, respectively. The template set is set based on the typical arrangement order of quality feature vectors under different data states, as well as the distribution of fluctuation fields and numerical fields.

[0042] Method for determining template proximity value: The unified quality feature vector is compared with each quality pattern template in order of field. For the parts with the same order field, the order matching value is accumulated; for the parts with the same fluctuation field, the fluctuation matching value is accumulated; and for the parts with small differences in the numerical field, the numerical matching value is accumulated. The three types of matching values ​​are added together to obtain the corresponding template proximity value.

[0043] The specific method for combining the unique identifier of the quality pattern template corresponding to the target template proximity value with the quality feature vector is as follows: extract the unique identifier of the template corresponding to the target template proximity value, write the unique identifier to the front of the identifier field of the quality feature vector, and retain the target template proximity value as an additional field to form a data quality fingerprint.

[0044] Based on data quality fingerprints, the monitoring data is divided into trusted data, repairable missing data, repairable abnormal data, and data awaiting retesting. The specific details are as follows: Read the unique identifier, template proximity value and quality feature vector from the data quality fingerprint to determine the target quality pattern corresponding to the monitoring data; The steps for determining the target quality pattern corresponding to the monitoring data are as follows: Read the unique identifier, target template proximity value, and quality feature vector from the data quality fingerprint. First, determine the candidate quality pattern based on the unique identifier. Then, compare the target template proximity value with the proximity lower limit corresponding to the candidate quality pattern. If the target template proximity value is not lower than the proximity lower limit, then determine the candidate quality pattern as the target quality pattern. If the target template proximity value is lower than the proximity lower limit, then re-compare the quality feature vector with the other quality pattern templates and select the quality pattern with the second highest proximity value that exceeds the proximity lower limit as the target quality pattern.

[0045] Based on the complete state, continuous state, adjacent state and associated state in the target quality pattern, the corresponding fields in the quality feature vector are compared item by item to determine the state with the largest deviation value as the dominant mismatch item, and the number of mismatches in the remaining states whose deviation values ​​exceed the preset deviation threshold is counted. Based on the complete state, continuous state, adjacent state, and associated state in the target quality model, the specific content of comparing the corresponding fields in the quality feature vector item by item is as follows: First, read the complete field, continuous field, adjacent field, and associated field in the quality feature vector, and then compare them with the complete state, continuous state, adjacent state, and associated state in the target quality model respectively to obtain the complete deviation value, continuous deviation value, adjacent deviation value, and associated deviation value, and determine the dominant mismatch item from them.

[0046] Preset deviation threshold: The threshold used to determine whether each deviation value belongs to the allowable fluctuation range. It is set based on the normal fluctuation limit of monitoring data within the same construction stage, the upper limit of the difference between adjacent locations in the current area, and the upper limit of the allowable offset between related projects.

[0047] When the dominant mismatch term does not exist and the template proximity value is greater than the preset confidence threshold, the monitoring data is determined to be reliable data. Preset confidence threshold: refers to the lower limit value used to determine whether the template proximity value is sufficient to support the direct use of monitoring data as reliable data. The preset confidence threshold is set based on the distribution results of multiple quality mode templates in historical comparisons, the stable range of the target quality mode, and the requirements of formal early warning for misjudgment control. When the cost of misjudgment is high, the preset confidence threshold is increased; when the cost of retesting is high, the preset confidence threshold is appropriately decreased.

[0048] When the dominant mismatch item is in a complete state, and the deviation values ​​corresponding to the continuous state, adjacent state, and associated state are all no greater than the preset deviation threshold, the monitoring data is determined to be repairable missing data. When the dominant mismatch item is a continuous state, an adjacent state, or an associated state, and the deviation value corresponding to the complete state is not greater than the preset deviation threshold, the monitoring data will be identified as repairable abnormal data. When the template proximity value is not greater than the preset confidence threshold, or the number of mismatches is greater than the preset number threshold, the monitoring data will be identified as data to be retested.

[0049] Preset quantity threshold: refers to the maximum number of mismatches that can be reached. It is used to determine whether the monitoring data enters the repair process or the retesting process. The preset quantity threshold is set based on the number of simultaneous mismatches that can be tolerated in the complete state, continuous state, adjacent state, and associated state. When multiple states are mismatched at the same time, it is more likely to correspond to equipment abnormalities or sudden changes in operating conditions. The preset quantity threshold is reduced.

[0050] One embodiment of the present invention is as follows: the preset repair channel includes a first repair channel, a second repair channel and a third repair channel; The first repair channel is used to perform historical record extraction, stage record filtering and trend continuation processing on repairable missing data to generate the first candidate repair value; The second repair channel is used to perform synchronous record extraction, adjacent record filtering and difference correction processing on repairable missing data or repairable abnormal data to generate a second candidate repair value. The third repair channel is used to perform corresponding record extraction, related record filtering, and offset conversion processing on repairable abnormal data or data to be retested, and generate a third candidate repair value.

[0051] The steps for generating the first, second, and third candidate repair values ​​are as follows: The first repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data, retrieves the historical records of the same monitoring location in the same construction stage arranged in reverse order of reception time, selects the first historical record, the second historical record, and the third historical record that are closest to the current time, calculates the first change value between the first historical record and the second historical record, and calculates the second change value between the second historical record and the third historical record. When the first change value and the second change value change in the same direction, compare the first time interval between the first historical record and the current time and the second time interval between the second historical record and the current time to determine the main change value corresponding to the smaller time interval. Then compare the difference between the first change value and the second change value. When the difference is not greater than a preset close value, take the middle value between the first change value and the second change value as the continued change value. When the difference is greater than the preset close value, take the main change value as the continued change value. Preset proximity value: refers to the upper limit of the allowable proximity of the difference between the first change value and the second change value. It is used to determine whether two historical changes can be merged into the same continuous trend. The preset proximity value is set based on the natural dispersion of the continuous changes at the same monitoring location within the same construction stage and the time continuity requirements corresponding to the first time interval and the second time interval.

[0052] When the first change value and the second change value do not change in the same direction, the change value with the smaller absolute value is taken as the continued change value, and the continued change value is superimposed on the first historical record to obtain the first candidate repair value; The specific content of the first candidate repair value obtained by superimposing the continuous change value onto the first historical record is as follows: First, read the historical value corresponding to the first historical record, then determine the direction of the continuous change value. If the continuous change value is positive, add the continuous change value to the historical value. If the continuous change value is negative, subtract the absolute value of the continuous change value from the historical value to obtain the continuous repair value. Then, write the continuous repair value to the missing or abnormal position corresponding to the current monitoring data to form the first candidate repair value.

[0053] The second repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data or repairable abnormal data, retrieves the synchronous records of adjacent positions within the same construction stage and the synchronous records of adjacent positions at the previous moment, calculates the position difference of each synchronous record relative to the current monitoring data, calculates the offset difference of each synchronous record relative to the current monitoring data, retains the synchronous records whose change direction is consistent with the current monitoring data and whose offset difference is less than the position difference, averages the offset differences corresponding to the retained synchronous records to obtain the adjacent correction value, compares the adjacent correction value with the preset adjacent correction interval, when the adjacent correction value falls within the preset adjacent correction interval, the adjacent correction value is retained, when the adjacent correction value exceeds the preset adjacent correction interval, the interval boundary value consistent with the direction of the adjacent correction value is truncated as the target adjacent correction value, if the target adjacent correction value is positive, the target adjacent correction value is added to the current monitoring data, if the target adjacent correction value is negative, the absolute value of the target adjacent correction value is subtracted from the current monitoring data to obtain the second candidate repair value; Calculate the position difference of each synchronized record relative to the current monitoring data: The position difference is obtained by subtracting the layout coordinates of the corresponding position of the current monitoring data from the layout coordinates of the adjacent positions.

[0054] Calculate the offset difference of each synchronized record relative to the current monitoring data: the offset difference is obtained by subtracting the current value of the current monitoring data from the current value of each synchronized record.

[0055] Preset neighbor correction range: This is the range within which neighbor correction values ​​are allowed to apply to the current monitoring data. It is set based on the maximum allowable difference between adjacent locations in the same construction stage, the spatial correlation strength of the monitoring locations in the current area, and the upper limit of the neighbor correction that the second candidate repair value should not exceed.

[0056] The third repair channel reads the current monitoring data, target quality mode, and current construction stage context corresponding to the repairable abnormal data or data to be retested. It retrieves the current corresponding record and the previous corresponding record of the related project within the same construction stage, calculates the correlation change value between the current corresponding record and the previous corresponding record, calculates the historical correspondence difference between the current monitoring data and the previous corresponding record, and superimposes the correlation change value and the historical correspondence difference to form a correlation conversion value. The correlation conversion value is compared with a preset correlation correction interval. When the correlation conversion value falls within the preset correlation correction interval, the correlation conversion value is retained. When the correlation conversion value exceeds the preset correlation correction interval, the boundary value of the interval with the same direction as the correlation conversion value is truncated as the target correlation conversion value. If the target correlation conversion value is positive, the target correlation conversion value is added to the current monitoring data. If the target correlation conversion value is negative, the absolute value of the target correlation conversion value is subtracted from the current monitoring data to obtain the third candidate repair value.

[0057] The specific steps for superimposing the associated change value with the corresponding historical difference to form the associated conversion value are as follows: First, determine whether the direction of the associated change value and the corresponding historical difference is consistent. If the direction is consistent, then add the two in the same direction to form the associated conversion value. If the direction is inconsistent, then subtract the smaller absolute value from the larger absolute value and retain the direction of the larger value to form the associated conversion value.

[0058] Preset correlation correction range: This is the range that limits the correction range of the correlation conversion value to the current monitoring data. It is set based on the historical correspondence range between the correlation item and the current monitoring data, the maximum acceptable offset of the correlation change within the current construction stage, and the stable output requirements of the third candidate repair value.

[0059] The first repair channel has a structure consisting of a historical record extraction structure, a stage record filtering structure, a trend continuation structure, and a first output structure. The second repair channel has a structure consisting of a synchronous record extraction structure, a neighboring record filtering structure, a difference correction structure, and a second output structure. The third repair channel has a structure consisting of a corresponding record extraction structure, a related record filtering structure, an offset conversion structure, and a third output structure.

[0060] One embodiment of the present invention is as follows: the method for determining the first confidence level, the second confidence level, and the third confidence level is as follows: Read the first historical record, second historical record, third historical record, and current construction stage context corresponding to the first candidate repair value; calculate the first repair difference between the first candidate repair value and the first historical record; calculate the first amplitude difference between the first change value and the second change value; calculate the first time difference between the first time interval and the second time interval; and assign the first repair difference, the first amplitude difference, and the first time difference to the first confidence order in ascending order; and determine the first confidence level based on the first confidence order. Read the synchronous record, adjacent correction value, target adjacent correction value and current construction stage context corresponding to the second candidate repair value, calculate the second offset difference between the second candidate repair value and each synchronous record, count the number of valid synchronous records that are consistent with the change direction of the second candidate repair value, calculate the second correction difference between the adjacent correction value and the target adjacent correction value, and assign the second offset difference, the number of valid synchronous records and the second correction difference to the second confidence order from best to worst, and determine the second confidence level based on the second confidence order; Read the current corresponding record, the previous corresponding record, the associated change value, the historical corresponding difference, and the target quality pattern corresponding to the third candidate repair value. Calculate the third corresponding difference between the third candidate repair value and the current corresponding record. Compare the directional consistency between the associated change value and the historical corresponding difference. Determine the matching length of the third candidate repair value falling into the state interval corresponding to the target quality pattern. Assign the third corresponding difference, directional consistency, and matching length in descending order of quality as the third confidence order. Determine the third confidence level based on the third confidence order.

[0061] The steps for determining the matching length of the third candidate repair value falling into the state interval corresponding to the target quality mode are as follows: read the upper and lower boundaries of the state interval corresponding to the target quality mode, then read the numerical range of the third candidate repair value, and determine whether there is an overlap between the numerical range of the third candidate repair value and the state interval corresponding to the target quality mode. If there is an overlap, calculate the length of the overlapping range as the matching length. If there is no overlap, record the matching length as zero and use it to participate in the determination of the third trust order.

[0062] The preset early warning model is a Bayesian network model. The Bayesian network model receives the first candidate repair value, the second candidate repair value, the third candidate repair value, and the first confidence level, the second confidence level, and the third confidence level. First, it determines the first risk state, the second risk state, and the third risk state based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. Then, it performs state correction on the first risk state, the second risk state, and the third risk state based on the first confidence level, the second confidence level, and the third confidence level to obtain the corresponding first risk probability, the second risk probability, and the third risk probability. The maximum value among the first risk probability, the second risk probability, and the third risk probability is determined as the quality constraint risk value.

[0063] One embodiment of the present invention is as follows: The Bayesian network model includes an input node, a state node, a correction node, and an output node. The input node is used to receive a first candidate repair value, a second candidate repair value, a third candidate repair value, and a first confidence level, a second confidence level, and a third confidence level. The state node is used to represent a first risk state, a second risk state, and a third risk state, respectively. The correction node is used to perform reliable corrections on the first risk state, the second risk state, and the third risk state based on the first confidence level, the second confidence level, and the third confidence level. The output node is used to output the first risk probability, the second risk probability, the third risk probability, and the quality constraint risk value. The first risk state is determined by the degree of deviation of the first candidate repair value from a first historical record; the second risk state is determined by the degree of offset of the second candidate repair value from a synchronized record; and the third risk state is determined by the degree of correspondence of the third candidate repair value with the current corresponding record.

[0064] One embodiment of the present invention is as follows: the first risk state, the second risk state, and the third risk state each include a low-risk state, a medium-risk state, and a high-risk state; when the first repair difference between the first candidate repair value and the first historical record is not greater than the upper limit of the first deviation interval, the first risk state is determined to be a low-risk state; when the first repair difference is greater than the upper limit of the first deviation interval but not greater than the upper limit of the second deviation interval, the first risk state is determined to be a medium-risk state; when the first repair difference is greater than the upper limit of the second deviation interval, the first risk state is determined to be a high-risk state; when the second offset difference between the second candidate repair value and each synchronized record is not greater than the upper limit of the third deviation interval, the second... The risk status is determined as low risk. When the second offset difference is greater than the upper limit of the third deviation interval but not greater than the upper limit of the fourth deviation interval, the second risk status is determined as medium risk. When the second offset difference is greater than the upper limit of the fourth deviation interval, the second risk status is determined as high risk. When the matching length of the third candidate repair value falling into the corresponding state interval of the target quality mode is not less than the first matching length threshold, the third risk status is determined as low risk. When the matching length is less than the first matching length threshold but not less than the second matching length threshold, the third risk status is determined as medium risk. When the matching length is less than the second matching length threshold, the third risk status is determined as high risk.

[0065] One embodiment of the present invention is as follows: a first confidence level, a second confidence level, and a third confidence level are used to correct the first risk state, the second risk state, and the third risk state. The correction method is as follows: when the corresponding confidence level is not lower than a preset high confidence threshold, the corresponding risk state remains unchanged; when the corresponding confidence level is lower than the preset high confidence threshold but not lower than a preset medium confidence threshold, if the corresponding risk state is a high risk state, the high risk state is corrected to a medium risk state, and if the corresponding risk state is a medium risk state, the medium risk state remains unchanged; when the corresponding confidence level is lower than the preset medium confidence threshold, if the corresponding risk state is a high risk state, the high risk state is corrected to a medium risk state, if the corresponding risk state is a medium risk state, the medium risk state is corrected to a low risk state, and if the corresponding risk state is a low risk state, the low risk state remains unchanged; after the state correction is completed, the median of the low risk probability interval corresponding to the corrected low risk state, the median of the medium risk probability interval corresponding to the corrected medium risk state, and the median of the high risk probability interval corresponding to the corrected high risk state are read and used as the first risk probability, the second risk probability, and the third risk probability, respectively.

[0066] One embodiment of the present invention is as follows: the output steps for formal early warning, review early warning, partial retest, or observation events are as follows: Read the quality constraint risk value, the first risk probability, the second risk probability, the third risk probability, and the first confidence level, the second confidence level, and the third confidence level; determine the target risk probability and target confidence level corresponding to the quality constraint risk value; and determine the target risk state corresponding to the target risk probability. The method for determining the target risk probability and target confidence level corresponding to the quality constraint risk value is as follows: compare the first risk probability, the second risk probability, and the third risk probability, select the risk probability with the largest value as the target risk probability, and extract the confidence level corresponding to the largest risk probability as the target confidence level.

[0067] The steps for determining the target risk state corresponding to the target risk probability are as follows: read the first risk state, second risk state, or third risk state corresponding to the target risk probability, and determine the risk state that is of the same origin as the maximum risk probability as the target risk state.

[0068] The target risk probability is compared with the preset formal risk threshold, preset review risk threshold and preset observation risk threshold, and the target confidence level is compared with the preset high confidence threshold and preset medium confidence threshold to obtain the formal judgment result, review judgment result, retest judgment result or observation judgment result; Preset formal risk threshold: This is the lower limit of risk that triggers a formal warning. It is set based on the minimum acceptable probability of a severe risk condition.

[0069] Preset review risk threshold: This is the lower limit of risk that triggers a review warning. It is set based on the intermediate risk probability that requires manual confirmation or encrypted sampling.

[0070] Preset observation risk threshold: This is the lower limit of risk that triggers the observation event. It is set based on the probability of mild risk that should be recorded but not intervened in at the moment.

[0071] Preset high confidence threshold: This is the lower limit for determining that the target confidence level is sufficient to directly support the formal output. It is set based on the stability of the repair channel and the upper bound of the confidence level distribution.

[0072] The preset medium confidence threshold is the lower limit for determining that the confidence level of the target is sufficient to support the review output. It is set based on the confidence level that can still be tracked while retaining a certain degree of uncertainty. The preset formal risk threshold is greater than the preset review risk threshold, the preset review risk threshold is greater than the preset observation risk threshold, and the preset high confidence threshold is greater than the preset medium confidence threshold.

[0073] When the formal judgment result is valid, a formal warning is output and a formal record is generated. When the review judgment result is valid, a review warning is output and a review record is generated. When the retest judgment result is valid, a partial retest is output and a retest record is generated. When the observation judgment result is valid, an observation event is output and an observation record is generated. If the formal judgment result is successful, the probability of the target risk is not lower than the preset formal risk threshold and the confidence level of the target is not lower than the preset high confidence threshold.

[0074] If the review and judgment result is successful, the probability of the target risk is not lower than the preset review risk threshold and the target confidence level is not lower than the preset medium confidence threshold but lower than the preset high confidence threshold.

[0075] If the retest result is successful, the probability of the target risk is not lower than the preset retest risk threshold and the target confidence level is lower than the preset medium confidence threshold.

[0076] The formal warning specifically involves sending an immediate action instruction to the output area and increasing the frequency of subsequent data collection.

[0077] The review and early warning specifically involves sending key verification instructions to the output area and maintaining encrypted sampling. Local re-monitoring specifically involves: issuing a re-monitoring request for the current monitoring location and adjacent locations, and re-collecting monitoring data. The specific observation event involves recording the current risk status and continuously tracking subsequent monitoring data according to the output duration.

[0078] Based on official records, review records, retest records, or observation records, determine the corresponding output area, output duration, and subsequent acquisition frequency, and write the output area, output duration, and subsequent acquisition frequency into the construction stage context corresponding to the subsequent monitoring data.

[0079] One embodiment of the present invention is as follows: the output area is determined based on the target risk status, the current area, and formal records, review records, retest records, or observation records; when a formal warning is output, the current monitoring location and its adjacent locations are determined as the output area; when a review warning is output, the current monitoring location is determined as the output area, and the support location or drainage location corresponding to the current monitoring location is determined as the extended verification area; when a partial retest is output, the current monitoring location and its adjacent locations with synchronous records are determined as the output area; when an observation event is output, the current monitoring location is determined as the output area.

[0080] One embodiment of the present invention is as follows: the output duration is determined jointly based on the target risk probability and the target confidence level; when a formal warning is established, the first preset duration is determined as the output duration; when a verification warning is established, the second preset duration is determined as the output duration; when a local retest is established, the third preset duration is determined as the output duration; when an observed event is established, the fourth preset duration is determined as the output duration; wherein, the first preset duration is longer than the second preset duration, the second preset duration is longer than the third preset duration, and the third preset duration is longer than the fourth preset duration; if the target confidence level is lower than the preset confidence threshold, a retest waiting time is added to the corresponding preset duration to ensure that subsequent monitoring data can cover the retest feedback process.

[0081] One embodiment of the present invention is as follows: the subsequent collection frequency is determined according to the formal record, review record, retest record, or observation record respectively; when a formal record is generated, the current collection frequency is increased to the first frequency level; when a review record is generated, the current collection frequency is increased to the second frequency level; when a retest record is generated, the current collection frequency is increased to the third frequency level, and the third frequency level is higher than the first frequency level; when an observation record is generated, the current collection frequency is kept unchanged or increased to the fourth frequency level; wherein, the third frequency level is higher than the first frequency level, the first frequency level is higher than the second frequency level, and the second frequency level is higher than the fourth frequency level.

[0082] One embodiment of the present invention is as follows: when writing the output area, output duration and subsequent acquisition frequency into the construction stage context corresponding to the subsequent monitoring data, the output area is first written into the area field, then the output duration is written into the duration field, then the subsequent acquisition frequency is written into the frequency field, and an output source mark is added to the construction stage context to identify the source type of formal warning, review warning, local retest or observation event, so that after the subsequent monitoring data enters the edge node, it can preferentially call the construction stage context consistent with the current output source mark.

[0083] Example: At a deep foundation pit construction site, an edge node is deployed at a preset distance of 30 meters from the monitoring data source. The edge node receives displacement data 10.8, settlement data 4.2, water level data 6.1, support change information, excavation change information, drainage change information, and equipment change information, and generates a construction stage context, in which the current area is area A, the current level is the second level, the current support status is supported, the current drainage status is pumping, and the current communication status is stable. Within a 10-minute short-term analysis window, the completeness of the statistical monitoring data was 0.96, the numerical stability was 0.88, and the signal clarity was 0.91, resulting in an initial quality vector. After correction by a preset quality benchmark, a regional correction value of 0.02, and a communication correction value of 0.01, a context-corrected quality vector was obtained, forming a data quality fingerprint, and the monitoring data was identified as repairable missing data. The first repair channel retrieves the first historical record 10.2, the second historical record 9.8, and the third historical record 9.4 to obtain the first change value 0.4 and the second change value 0.4. Since the difference amplitude of 0 is not greater than the preset close value of 0.1, the middle value of 0.4 is taken as the continued change value and superimposed on the first historical record to obtain the first candidate repair value 10.6. The second repair channel calculates the adjacent correction value of 0.2 based on the synchronization record, resulting in the second candidate repair value of 10.8; the third repair channel forms the correlation conversion value of 0.4 based on the correlation change value of 0.3 and the historical corresponding difference of 0.1, resulting in the third candidate repair value of 11.2. The first confidence level is set to 0.86, the second confidence level to 0.82, and the third confidence level to 0.74. The first candidate repair value, the second candidate repair value, the third candidate repair value, and the first, second, and third confidence levels are input into the Bayesian network model to obtain the first risk probability of 0.42, the second risk probability of 0.56, and the third risk probability of 0.63. The quality constraint risk value is set to 0.63, a review warning is output, and the output area, output duration of 20 minutes, and subsequent collection frequency are written into the construction phase context.

[0084] This embodiment can combine the construction phase context, data quality fingerprint, and three repair channels to obtain the quality constraint risk value, thereby improving the pertinence of the review and early warning output and the reliability of subsequent collection frequency adjustment.

[0085] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent early warning and monitoring of deep foundation pit construction based on edge computing, characterized in that: Includes the following steps: The edge node deployed at the deep foundation pit construction site receives monitoring data and generates a construction phase context based on the monitoring data; Data quality fingerprints are extracted from the monitoring data and the construction phase context, and the monitoring data is divided into reliable data, repairable missing data, repairable abnormal data, and data to be retested based on the data quality fingerprints. Based on the repairable missing data, the repairable abnormal data, and the data to be retested, a preset repair channel is selected to generate a first candidate repair value, a second candidate repair value, and a third candidate repair value. A first confidence level, a second confidence level, and a third confidence level are determined based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. The first candidate repair value, the second candidate repair value, and the third candidate repair value are then input together with the first confidence level, the second confidence level, and the third confidence level into a preset early warning model to obtain a quality constraint risk value. Based on the quality constraint risk value and the first confidence level, the second confidence level, and the third confidence level, output formal warnings, review warnings, partial retests, or observation events, and update the construction phase context of subsequent monitoring data based on the formal warnings, review warnings, partial retests, or observation events.

2. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 1, characterized in that: The edge nodes are deployed at a predetermined distance from the monitoring data at the deep foundation pit construction site; The edge node is used to perform the following processing: receiving the monitoring data, generating the construction stage context based on the monitoring data, extracting the data quality fingerprint, classifying the trusted data, the repairable missing data, the repairable abnormal data, and the data to be retested based on the data quality fingerprint, generating the first candidate repair value, the second candidate repair value, and the third candidate repair value based on the repairable missing data, the repairable abnormal data, and the data to be retested, determining the first confidence level, the second confidence level, and the third confidence level, and outputting the formal warning, the review warning, the partial retest, or the observation event.

3. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 2, characterized in that: The steps for generating the construction phase context are as follows: Construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information are extracted from the monitoring data, and a current operation record is generated based on the construction time information, construction location information, support change information, excavation change information, drainage change information, and equipment change information. The current area, current level, current support status, current drainage status, and current communication status are determined based on the current operation record, and a stage feature record is generated based on the current area, current level, current support status, current drainage status, and current communication status. The construction stage label is determined by matching the stage feature record with the preset stage rules, and a transition state record is generated based on the construction stage label, the current area, and the construction stage context of the previous moment. The construction stage label is maintained, switched, or delayed based on the transition state record, and the construction stage context at the current moment is generated based on the updated construction stage label, the current region, the current communication state, and the construction stage context of the previous moment.

4. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 3, characterized in that: The steps for extracting the data quality fingerprint are as follows: Within a preset short-term analysis window, the monitoring data is grouped according to the reception time. Within the same construction stage defined by the construction stage context, the completeness, numerical stability, and signal clarity of each group of monitoring data are statistically analyzed. The completeness, numerical stability, and signal clarity are then combined in a preset order to obtain an initial quality vector. Each quality value in the initial quality vector is compared with the corresponding benchmark value in the preset quality benchmark corresponding to the construction stage context to obtain the benchmark deviation of each quality value. Then, according to the area correction value corresponding to the current area and the communication correction value corresponding to the current communication state in the construction stage context, the benchmark deviation of each quality value is corrected. The corrected quality values ​​are written into the corresponding positions in the preset order to obtain the context correction quality vector. The quality values ​​in the context-corrected quality vector are arranged from high to low to obtain the current arrangement order. The current arrangement order is then compared with the arrangement order of the previous analysis window, and the order difference between two adjacent quality values ​​in the current arrangement order and the overall offset of the overall average value of the context-corrected quality vector relative to the preset quality benchmark are determined. When the order difference exceeds the preset order condition or the overall offset exceeds the preset offset condition, a vector fluctuation mark is generated. The current arrangement order, the vector fluctuation mark, and the context-corrected quality vector are written into the same vector structure according to a preset field order to form a quality feature vector, and the quality feature vector is then processed to unify it. The unified quality feature vector is compared with a plurality of preset quality pattern templates one by one to determine the template proximity value between each quality pattern template and the quality feature vector. The target template proximity value with the largest value that is greater than a preset proximity threshold is selected from the template proximity values. The unique identifier of the quality pattern template corresponding to the target template proximity value is combined with the quality feature vector to obtain the data quality fingerprint.

5. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 4, characterized in that: The specific content of classifying the monitoring data into reliable data, repairable missing data, repairable abnormal data, and data to be retested based on the data quality fingerprint is as follows: Read the unique identifier, the template proximity value, and the quality feature vector from the data quality fingerprint to determine the target quality pattern corresponding to the monitoring data; Based on the complete state, continuous state, adjacent state and associated state in the target quality pattern, the corresponding fields in the quality feature vector are compared item by item to determine the state with the largest deviation value as the dominant mismatch item, and the number of mismatches in the remaining states whose deviation values ​​exceed the preset deviation threshold is counted. When the dominant mismatch item does not exist and the template proximity value is greater than the preset confidence threshold, the monitoring data is determined as the confidence data; When the dominant mismatch item is the complete state, and the deviation values ​​corresponding to the continuous state, the adjacent state, and the associated state are all not greater than the preset deviation threshold, the monitoring data is determined as the repairable missing data. When the dominant mismatch item is the continuous state, the adjacent state, or the associated state, and the deviation value corresponding to the complete state is not greater than the preset deviation threshold, the monitoring data is determined as the repairable abnormal data. When the template proximity value is not greater than the preset confidence threshold, or the number of mismatches is greater than the preset number threshold, the monitoring data is determined as the data to be retested.

6. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 5, characterized in that: The preset repair channels include a first repair channel, a second repair channel, and a third repair channel; The first repair channel is used to perform historical record extraction, stage record filtering, and trend continuation processing on the repairable missing data to generate the first candidate repair value; The second repair channel is used to perform synchronous record extraction, adjacent record filtering and difference correction processing on the repairable missing data or the repairable abnormal data to generate the second candidate repair value; The third repair channel is used to perform corresponding record extraction, associated record filtering and offset conversion processing on the repairable abnormal data or the data to be retested, and generate the third candidate repair value.

7. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 6, characterized in that: The steps for generating the first candidate repair value, the second candidate repair value, and the third candidate repair value are as follows: The first repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data, retrieves the historical records of the same monitoring location in the same construction stage arranged in reverse order of reception time, selects the first historical record, the second historical record, and the third historical record that are closest to the current time, calculates the first change value between the first historical record and the second historical record, and calculates the second change value between the second historical record and the third historical record; When the first change value and the second change value change in the same direction, compare the first time interval between the first historical record and the current time and the second time interval between the second historical record and the current time to determine the main change value corresponding to the smaller time interval. Then compare the difference between the first change value and the second change value. When the difference is not greater than a preset proximity value, take the middle value between the first change value and the second change value as the continuing change value. When the difference is greater than the preset proximity value, take the main change value as the continuing change value. When the first change value and the second change value do not change in the same direction, the change value with the smaller absolute value is taken as the continued change value, and the continued change value is superimposed on the first historical record to obtain the first candidate repair value; The second repair channel reads the current monitoring data and current construction stage context corresponding to the repairable missing data or the repairable abnormal data, retrieves the synchronous records of adjacent positions within the same construction stage and the synchronous records of the adjacent positions at the previous moment, calculates the position difference of each synchronous record relative to the current monitoring data, calculates the offset difference of each synchronous record relative to the current monitoring data, retains the synchronous records whose change direction is consistent with the current monitoring data and whose offset difference is less than the position difference, averages the offset differences corresponding to the retained synchronous records to obtain the adjacent position correction value, compares the adjacent position correction value with a preset adjacent position correction interval, when the adjacent position correction value falls within the preset adjacent position correction interval, the adjacent position correction value is retained, when the adjacent position correction value exceeds the preset adjacent position correction interval, the interval boundary value consistent with the direction of the adjacent position correction value is truncated as the target adjacent position correction value, if the target adjacent position correction value is positive, the target adjacent position correction value is added to the current monitoring data, if the target adjacent position correction value is negative, the absolute value of the target adjacent position correction value is subtracted from the current monitoring data to obtain the second candidate repair value; The third repair channel reads the current monitoring data corresponding to the repairable abnormal data or the data to be retested, the target quality mode, and the current construction stage context. It retrieves the current corresponding record and the previous corresponding record of the related project within the same construction stage, calculates the correlation change value between the current corresponding record and the previous corresponding record, calculates the historical correspondence difference between the current monitoring data and the previous corresponding record, and superimposes the correlation change value and the historical correspondence difference to form a correlation conversion value. The correlation conversion value is compared with a preset correlation correction interval. When the correlation conversion value falls within the preset correlation correction interval, the correlation conversion value is retained. When the correlation conversion value exceeds the preset correlation correction interval, the interval boundary value with the same direction as the correlation conversion value is truncated as the target correlation conversion value. If the target correlation conversion value is positive, the target correlation conversion value is added to the current monitoring data. If the target correlation conversion value is negative, the absolute value of the target correlation conversion value is subtracted from the current monitoring data to obtain the third candidate repair value.

8. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 7, characterized in that: The methods for determining the first confidence level, the second confidence level, and the third confidence level are as follows: Read the first historical record, the second historical record, the third historical record, and the current construction stage context corresponding to the first candidate repair value; calculate the first repair difference between the first candidate repair value and the first historical record; calculate the first amplitude difference between the first change value and the second change value; calculate the first time difference between the first time interval and the second time interval; and assign the first repair difference, the first amplitude difference, and the first time difference to a first confidence order in ascending order; and determine the first confidence level based on the first confidence order. Read the synchronization record, the neighbor correction value, the target neighbor correction value, and the current construction stage context corresponding to the second candidate repair value; calculate the second offset difference between the second candidate repair value and each synchronization record; count the number of valid synchronization records that are consistent with the change direction of the second candidate repair value; calculate the second correction difference between the neighbor correction value and the target neighbor correction value; and assign the second offset difference, the number of valid synchronization records, and the second correction difference to a second confidence order from best to worst; and determine the second confidence level based on the second confidence order. Read the current corresponding record, the previous corresponding record, the associated change value, the historical corresponding difference, and the target quality pattern corresponding to the third candidate repair value; calculate the third corresponding difference between the third candidate repair value and the current corresponding record; compare the directional consistency between the associated change value and the historical corresponding difference; determine the matching length of the third candidate repair value falling into the state interval corresponding to the target quality pattern; and assign the third corresponding difference, the directional consistency, and the matching length to the third confidence order from best to worst; and determine the third confidence level based on the third confidence order.

9. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 8, characterized in that: The preset early warning model is specifically a Bayesian network model. The Bayesian network model receives the first candidate repair value, the second candidate repair value, the third candidate repair value, and the first confidence level, the second confidence level, and the third confidence level. First, it determines the first risk state, the second risk state, and the third risk state based on the first candidate repair value, the second candidate repair value, and the third candidate repair value, respectively. Then, it performs state correction on the first risk state, the second risk state, and the third risk state based on the first confidence level, the second confidence level, and the third confidence level to obtain the corresponding first risk probability, the second risk probability, and the third risk probability. The maximum value among the first risk probability, the second risk probability, and the third risk probability is determined as the quality constraint risk value.

10. The intelligent early warning and monitoring method for deep foundation pit construction based on edge computing according to claim 9, characterized in that: The output steps for formal early warning, review early warning, partial retest, or observation events are as follows: Read the quality constraint risk value, the first risk probability, the second risk probability, the third risk probability, the first confidence level, the second confidence level, and the third confidence level; determine the target risk probability and target confidence level corresponding to the quality constraint risk value; and determine the target risk state corresponding to the target risk probability. The target risk probability is compared with a preset formal risk threshold, a preset review risk threshold, and a preset observation risk threshold, and the target confidence level is compared with a preset high confidence threshold and a preset medium confidence threshold to obtain a formal judgment result, a review judgment result, a retest judgment result, or an observation judgment result. When the formal judgment result is true, the formal warning is output and a formal record is generated. When the review judgment result is true, the review warning is output and a review record is generated. When the retest judgment result is true, the local retest is output and a retest record is generated. When the observation judgment result is true, the observation event is output and an observation record is generated. Based on the official record, the review record, the retest record, or the observation record, the corresponding output area, output duration, and subsequent acquisition frequency are determined respectively, and the output area, the output duration, and the subsequent acquisition frequency are written into the construction stage context corresponding to the subsequent monitoring data.

Images