Bim twin bio-driven shield tunnel monitoring data stream processing system

Abstract

The application discloses a BIM twin biological joint driving shield tunnel monitoring data stream processing system and relates to the technical field of shield tunnel monitoring processing. An edge gateway accesses shield machine operation parameters, structural environment monitoring and inspection data. After field alignment, uniform monitoring data packets are packaged, containing a globally unique identifier, a measurement point incremental serial number, double timestamps, a space-time object key and an object mapping version number. Disconnected cache is supplemented and power is equalized for storage. The cloud corrects alignment according to event time to form a time window and a ring number statistical frame. The statistical frame is online governed, and credibility, reason labels and uncertainty intervals are output. Based on credibility gating, early warning is shunted, and a late event correction generates a new version of the event and an evidence summary package. The credible event incremental update twin object state library is isolated from the update structure state and the sensor link state, can support playback and traceability and operation and maintenance closed loop, reduce false alarms and missed alarms, and maintain twin consistency and auditability.

Description

Technical Field

[0001] This invention relates to the field of shield tunnel monitoring and processing technology, specifically a shield tunnel monitoring data stream processing system driven by BIM twinning. Background Technology

[0002] Shield tunneling is commonly used in underground engineering projects such as urban subway sections and integrated utility tunnels. During the construction, assembly, grouting, operation, and maintenance phases of shield tunnels, it is essential to continuously obtain structural and environmental parameters such as surface settlement, convergence deformation, segment stress, and leakage. Simultaneously, operational parameters of the shield machine, including earth pressure, thrust, torque, grouting, and attitude, must be acquired for risk warning, construction control, and monitoring. Existing shield tunnel health monitoring and safety early warning systems employ multi-sensor data acquisition, remote transmission, a central database, and a human-machine interface for safety assessment. Building Information Modeling (BIM)-based monitoring visualization can associate monitoring point numbers with model profiles or component locations, and determine the status through threshold comparisons. A digital twin-based monitoring data processing scheme is used to group and merge monitoring data blocks at standard time intervals.

[0003] In practical engineering, abnormal underground communication links, power outages and restarts of equipment, and inconsistent time bases at various acquisition terminals can cause data loss, out-of-order delays, and sensor drift, jamming, or transient spikes due to environmental and installation conditions, leading to short-term abnormal changes in the data stored. Studies have shown that missing data in large-scale tunnel boring machine datasets can affect the analysis and interpretation of monitoring results. When several conditions overlap, reliance on fixed threshold comparisons or merging by time period still makes alarm and model status updates susceptible to data quality and update timing, resulting in unstable on-site handling and increased risks of review and work stoppages. For example, when there is a time discrepancy or delay between high-frequency operating parameters and low-frequency deformation monitoring, the threshold-based judgment results are difficult to correspond to the actual working conditions, increasing the reliance on alarm interpretation and verification. Missing segments and abnormal spikes are directly entered into the database for statistical purposes, which can lead to breakage and abrupt changes in trend calculations. When monitoring values ​​are directly mapped to model components for status display, if the validity of the data cannot be reflected synchronously, the model risk identification will deviate from the actual on-site status, further increasing management uncertainty and false alarms and missed alarms in sensitive sections such as underpasses or adjacent pipelines.

[0004] The resulting technical problem is:

[0005] When multi-source monitoring data of shield tunnels is accessed as a continuous data stream, if inconsistent time references cause out-of-order delays, missing data, or sensor zero drift, jamming, spikes, or other reasons that cause changes in data quality, how can we determine the validity and reliability of the monitoring data before the release of safety warnings and digital twin status, and determine whether the anomaly is caused by the quality of the acquisition link or sensor data, or by changes in structure or operating conditions, so that the alarm results and model status are consistent? Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a BIM twin-driven shield tunnel monitoring data stream processing system. This system generates time windows and ring number statistical frames through cloud-based event time-based correction and alignment; it processes these statistical frames online and outputs credibility, cause labels, and uncertainty ranges; it provides credibility-based gating, diversion, and early warning, and generates new event versions and evidence summary packages for late events; it incrementally updates the twin object state library with credible events and isolates updates to structural and sensor link states, reducing false alarms and missed alarms while maintaining twin consistency, thus solving the technical problems described in the background section.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] The BIM twin-driven shield tunnel monitoring data stream processing system includes: an edge gateway generating unified monitoring data messages containing a globally unique identifier, a measuring point incrementing sequence number, equipment acquisition time, platform access time, and a spatiotemporal object key; retransmission after connection failure; and cloud storage using the globally unique identifier and the measuring point incrementing sequence number.

[0011] The time deviation of the equipment is estimated based on the measurement points and the equipment acquisition time is corrected to obtain the correction event time. The unified monitoring data messages are aggregated according to the correction event time to generate time window statistical frames and ring number statistical frames, and the allowable late time correction is set.

[0012] Data quality management was performed on both time window statistical frames and ring number statistical frames, and missing data was filled in to generate a quality result object. The quality result object includes a confidence score, a set of cause labels, an effective sample rate, and an uncertainty interval.

[0013] Based on the quality result object, perform credibility gating and diversion, and generate data quality events, observation warnings or structural risk alarms. Aggregate and write them to the event ledger by spatiotemporal object key. When late correction causes changes in statistical frames, generate version replacement.

[0014] Establish a twin object state library; incrementally write according to the new version of the event ledger; the structural twin state is updated by observation warnings and structural risk alarms; the sensor link state is updated by data quality events; and replay according to the correction event time.

[0015] Furthermore, the unified monitoring data message includes a globally unique identifier, a measurement point incrementing sequence number, device acquisition time, platform access time, device and acquisition link status flags, a spatiotemporal object key, and an object mapping version number. Before encapsulation, the edge gateway performs field alignment, unit dimension unification, and basic legality verification on the fields. When the network is interrupted, the message is written to the local cache queue and retransmitted according to the measurement point incrementing sequence number after the link is restored.

[0016] Furthermore, the cloud receiving side uses a globally unique identifier and the incremental sequence number of the measurement point as an idempotent key to perform deduplication and merging on duplicate messages formed by disconnection and retransmission, and then writes them into the unified monitoring data message storage area. The device acquisition time and platform access time are retained along with the merging result and stored in association with the spatiotemporal object key and the object mapping version number.

[0017] Furthermore, for each measuring point, the device time deviation is maintained. The slow drift is estimated by the difference between the access time and the device acquisition time of the sliding statistics platform, and the device time deviation is updated to obtain the correction event time. When the device and acquisition link status flags change, the device time deviation is re-initialized, and the device acquisition time and correction event time are written into the unified monitoring data message at the same time.

[0018] Furthermore, based on the correction event time, the unified monitoring data messages are stream-aligned and aggregated to generate time window statistical frames. Different statistical periods are set for the tunnel boring machine operating parameters and structural monitoring data. The allowable delay time is dynamically adjusted based on link health indicators such as delay distribution, packet loss, retransmission, and buffer depth. Late data arriving within the allowable delay time is used to correct the statistical results of the corresponding time window statistical frames.

[0019] Furthermore, when an increment in the ring number is detected, a ring number event window is triggered, and the mileage that crosses the ring length threshold is recorded at the same time. The key statistics during the previous ring are solidified into a ring number statistical frame. The key statistics include the mean, fluctuation and slope. The ring number and mileage identifier, missing rate and effective sample rate are written into the ring number statistical frame, and the ring number statistical frame is associated with the spatiotemporal object key and written into the statistical frame storage area.

[0020] Furthermore, a data quality governance operator chain is executed on the time window statistical frames and ring number statistical frames. The operator chain includes missing detection, spike detection, drift detection, jamming detection, over-range and rate of change constraint verification, and spatial consistency check. The device and acquisition link status flags are introduced into the weighted calculation of the credibility score, and the detected anomaly categories are written into the cause label set.

[0021] Furthermore, missing data is filled in according to the duration of the missing data: short-term missing data is filled in by interpolation, medium-term missing data is filled in by establishing a regression relationship between neighboring measurement points and measurement points in the same loop, and the regression relationship is established by the same loop samples within the same loop number statistical frame, and long-term missing data is written with a data quality degradation marker instead of generating a fill value, and the fill value marker and uncertainty interval are written into the quality result object.

[0022] Furthermore, a three-state mechanism for credibility gating is established: the smaller value of the credibility score and the effective sample rate is compared with a low credibility threshold; if the value is lower than the low credibility threshold, a data quality event is generated; if the smaller value is between the low credibility threshold and the high credibility threshold, an observation warning is generated and the number of consecutive confirmations is recorded; if the smaller value is higher than the high credibility threshold, structural risk fusion judgment is initiated.

[0023] Furthermore, the structural risk fusion judgment is based on a combination of multiple evidence rules, including the trend and rate criteria of the main monitoring indicators, the spatial consistency criteria of the same ring and adjacent rings, the operating condition segmentation criteria, and the historical baseline deviation criteria. When the rule combination is satisfied, a structural risk alarm is generated and an evidence summary package is generated. The evidence summary package contains the triggering basis, the key statistical frame index, and the data range identifier.

[0024] Furthermore, data quality events, observation warnings, and structural risk alarms are deduplicated and aggregated according to spatiotemporal object keys and spatiotemporal neighborhoods to form measurement point-level, ring-level, and mileage segment-level events, and their status flows according to creation, confirmation, handling, closure, and recurrence; when late data arriving within the allowed late time changes the statistical frame, a new version of the event is generated and the event level and evidence summary package are updated.

[0025] Furthermore, a twin object state library is established, using spatiotemporal object keys and component identifiers as indexes to maintain component-level, ring-level, and mileage segment-level states. Version management is performed on object mapping version numbers, and mapping changes are recorded as events. Only observation warnings, structural risk alarms, and ring number statistics frames are written to the structural twin state, while data quality events are written to sensor health status and link health status.

[0026] Furthermore, the twin object state library is written using incremental update packages, which include the effective time, the set of changed fields, the source event identifier, and the state version number.

[0027] When a new version of an event is generated, the twin state is synchronously corrected in a version overwrite manner, and the twin state at any time is reconstructed by replaying the corrected event time. The continuously occurring data quality events are transformed into maintenance tasks such as sensor calibration, line troubleshooting, encrypted inspection and retesting suggestions, and are recorded in association with the spatiotemporal object key.

[0028] (III) Beneficial Effects

[0029] This invention provides a BIM twin-driven shield tunnel monitoring data stream processing system, which has the following advantages:

[0030] The edge gateway aligns and unifies the fields of tunnel boring machine operating parameters, structural environment monitoring, and inspection data into a unified monitoring data message. This message contains a globally unique identifier, an incremental sequence number of the measuring point, a spatiotemporal object key, and an object mapping version number. It also supports disconnection caching and retransmission, as well as idempotent data entry. Retransmissions are not duplicated, and measuring point migrations and model updates are traceable.

[0031] The cloud-based system corrects the acquisition time of equipment based on the time deviation of the measurement point maintenance equipment, generating correction event times. Time window statistical frames are generated by aligning and aggregating these correction event times. Ring numbers are incremented and fixed, and the statistical frames are associated with mileage. Subsequent references are made using a unified time axis and ring number units. The time window statistical frames and ring number statistical frames undergo checks for missing data, spikes, drift, jamming, and spatial consistency. Over-range and rate-of-change constraints are verified. Combined with equipment and acquisition link status flags, a quality result object is generated. This quality result object includes a reliability score, cause label, effective sample rate, and uncertainty interval, serving as the input for reliability gating and triage.

[0032] Based on the quality result object, credibility-gated traffic is distributed to generate data quality events, observation warnings, and structural risk alarms. Data is deduplicated and aggregated according to the spatiotemporal object key and its lifecycle is traversed. For late data within the late duration, a new version of the event is generated by correcting the statistical frame and the evidence summary package is updated. The twin incremental write is synchronized with the new version of the event. A twin object state library with spatiotemporal object key and component identifier index is established. The twin is written using incremental update packages and is overwritten and synchronized with the new version of the event. The structural twin is generated by observation warnings, structural risk alarms, and ring number statistical frames. Sensor health and link health are generated by data quality events. The twin is generated by replaying the correction event time for traceability and operation and maintenance closed loop generation. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the shield tunnel monitoring data stream processing system of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 This invention provides a BIM-based twin-driven shield tunnel monitoring data stream processing system, comprising:

[0036] Step 1: At the shield tunneling site or tunnel section, encapsulate multi-source monitoring information into a unified monitoring data message, and complete cached retransmission and idempotent storage under network interruption or link fluctuation conditions, so as to provide stable and traceable input for event time correction and streaming alignment in Step 2.

[0037] Shield tunnel construction typically takes place in enclosed underground spaces. The shield machine control system, structural monitoring sensors, environmental monitoring sensors, and inspection and maintenance records belong to different acquisition ports and different maintenance teams. Their acquisition frequencies, field naming, unit dimensions, time bases, and network link statuses differ. If the data is directly uploaded and entered into the database in its original format, link jitter can cause messages from the same measurement point to arrive repeatedly or be missing. Temporary replacement of sensors or adjustment of building information model component identifiers on-site can also result in multiple sets of numbering systems for the same physical location. These differences are amplified in steps two through five, manifesting as inconsistent time alignment, multiple measurement points occupying the same component, repeated generation of alarm events, and repeated overwriting of twin states.

[0038] Therefore, Step 1 uses the unified monitoring data message as the only data object, and uses the spatiotemporal object key, object mapping version number, device acquisition time, platform access time, globally unique identifier, and measurement point incrementing sequence number as mandatory fields to compress multi-source differences to a controllable field level, so that subsequent steps can be processed within the same semantic framework.

[0039] Within the same tunnel section, the shield machine's operating parameters are collected from the control system acquisition channel, settlement and convergence are collected from the structural monitoring acquisition channel, temperature, humidity and harmful gases are collected from the environmental monitoring acquisition channel, and routine inspections are collected from the manual input channel; the field names, unit dimensions and anomaly expression methods output by these channels are not consistent.

[0040] Near the control cabinet in the tunnel boring machine's operating room, the edge gateway identifies each input source channel through a pre-registered list of acquisition channels. Each source channel corresponds to an acquisition channel mapping table within the edge gateway. The mapping table clearly defines the one-to-one correspondence between the channel's output fields and the fields of the unified monitoring data message, and specifies the rules for filling missing fields and intercepting abnormal fields. Since it is common practice to maintain the original wiring and replace the equipment when replacing sensors or acquisition devices on-site, the edge gateway, upon recognizing a new device identifier on the same wiring port, does not directly reuse the old field interpretation. Instead, it uses the device's self-test status flag and the acquisition link status flag to jointly determine whether the device has entered an available state. When the device has not yet entered an available state, the edge gateway still encapsulates the channel data into a unified monitoring data message, but writes an enumeration value indicating that it has not entered an available state into the device and acquisition link status flags, so that subsequent steps can identify that the message belongs to an abnormal source at the data link layer.

[0041] During field alignment, the edge gateway simultaneously generates a field source fingerprint to record which acquisition channel, original field name, and acquisition port each unified field originated from. The field source fingerprint is stored in the database along with the unified monitoring data packet. Subsequently, when a sudden change occurs in the value of the same field, the source fingerprint can be used to trace back to the original channel and field wiring location. The generation of the field source fingerprint employs optional implementation paths: one is to use a fixed-order field concatenation and perform cyclic redundancy check to obtain a short fingerprint; the other is to use cryptographic hashing to obtain a long fingerprint. Both paths are fixed in the edge gateway's configuration file to avoid changes in fingerprint rules during operation.

[0042] The edge gateway completes field alignment using a data acquisition channel mapping table, expresses the availability status of the acquisition end using device and data acquisition link status flags, and then records the field source fingerprint along with the message.

[0043] When used, the field semantics of the unified monitoring data message remain consistent, preventing ambiguity in subsequent steps due to differences in field names. The unified monitoring data message carries the availability information of the acquisition end, enabling subsequent steps to distinguish between two sources: structural changes and acquisition end insecurity.

[0044] After field alignment, the edge gateway needs to address errors caused by inconsistent unit dimensions. In the field, the output units often change after a sensor of the same type is replaced, or the inspection records use a customary unit. If the unit dimensions are not standardized, subsequent time window statistics and reliability gating will treat unit differences as numerical abrupt changes.

[0045] To this end, the edge gateway maintains a unit-dimension mapping table for each unified field. The mapping table provides conversion rules from the original unit to the target unit and defines the basic validity boundaries for that field. These basic validity boundaries are not used to assess risk, but rather to intercept formatting errors that are clearly impossible for the sensor to generate, such as empty strings, non-numeric characters, missing units, and mismatches between unit labels and field types.

[0046] After unifying unit dimensions and performing basic legality verification, the edge gateway binds the message and object affiliation together. A spatiotemporal object key is generated by mapping the object version number. This allows the same message to correspond to a unique tunnel section location and a unique building information model component affiliation in subsequent steps. The spatiotemporal object key is generated by a deterministic encoding function.

[0047]

[0048] In the formula: Project identifier : Code used to identify engineering projects, with values ​​taken as a pre-registered string sequence; range identifier : Code used to identify tunnel sections, with values ​​taken from a pre-registered string sequence: mileage identifier : Used to identify the mileage position corresponding to the message, the value can be an integer or a fixed-point decimal string: ring number identifier : Used to identify the shield tunnel ring number, positive integer:

[0049] Component identification Used to identify building information model components; string sequence: measurement point identifier. Used to identify physical measurement points; values ​​are string sequences: data type identifier. : Used to identify data categories, with values ​​taken from a preset enumeration string to distinguish between various types such as settlement, convergence, temperature, earth pressure, thrust, and torque; object mapping version number. : A positive integer used to identify the mapping version upon which the spacetime object key is generated; spacetime object key : This is a deterministic encoded output, with values ​​being fixed-length strings; deterministic encoding function. : Take the input concatenated byte sequence, encode it according to a predetermined character set, and then output the string according to a fixed-length mapping rule;

[0050] In object mapping version number The edge gateway operates on the premise of measurement point movement and model changes: the site is located at the same mileage marker. and ring number marking When redeploying measurement points, or identifying components in the Building Information Model (BIM) After being replaced, the edge gateway does not overwrite the old version mapping, but instead maps the object to the version number. Adding a new version number to the message creates two mapping trajectories at the data level, preventing new measurement point data from overwriting the historical data of old measurement points.

[0051] The edge gateway uses a unit dimension mapping table to unify the measurement benchmark and perform basic legality verification. Then, it generates a spatiotemporal object key using a deterministic encoding function and writes it into the object mapping version number.

[0052] In order to make It is feasible and the output is reproducible. The following implementation constraints are added:

[0053] Field normalization: Project identifier Interval identifier Component identification Measurement point marking Data type identifier First, remove leading and trailing whitespace, use a unified character set encoding, and prohibit the inclusion of delimiters; mileage identifiers. With ring number identification Use decimal strings uniformly with fixed decimal place rules; object mapping version number. Use a decimal positive integer string.

[0054] Separators and Concatenation: Concatenation Symbols Remove the delimiter characters that are not allowed in the fields, and use empty strings as placeholders for missing fields while retaining the delimiter positions, thereby ensuring that the field order and the number of fields remain constant.

[0055] Fixed-length output: Output a fixed-length string. Implementation path one involves first concatenating the original byte sequence according to rules, then performing fixed-length encoding and truncating / padding; implementation path two involves performing a cryptographic hash on the original byte sequence and converting it into a fixed-length string. Both paths satisfy the deterministic and fixed-length requirements.

[0056] Conflict handling: In the event of a very low probability of key conflicts in the project, use the spacetime object key. +The original set of fields is used as the review condition, not based on It can be used as a unique truth value on its own; when a conflict occurs, the version number is mapped to the object. The parallel trajectory retains two records and is marked as pending verification on the audit end.

[0057] When used, the numerical meaning of the unified monitoring data messages remains consistent, and subsequent steps will not cause format changes due to differences in units during statistics and discrimination; the unified monitoring data messages obtain a unique object attribution index, and subsequent steps can aggregate all data of the same component according to the same spatiotemporal object key.

[0058] Underground communication links are subject to interruptions and jitter, and unified monitoring data packets may be repeatedly sent at the sending end or repeatedly received at the receiving end. At the same time, on-site retransmission usually occurs within a short period of time after the link is restored, and retransmitted packets and newly generated packets will arrive interleaved. Without establishing a clear retransmission order and deduplication rules, subsequent steps will treat duplicate packets as multiple observations, resulting in duplicate entries in time window statistics, duplicate generation of alarm events, and repeated triggering of twin incremental updates.

[0059] The edge gateway establishes a local cache queue for each acquisition channel, and uses the incremental sequence number of the measurement points as the natural order within the queue. To avoid cache loss due to power outages or restarts, the local cache queue is written to the edge gateway's non-volatile storage medium in the form of a persistent log, and the write batch is recorded with the log segment number.

[0060] The edge gateway divides the queue into unacknowledged segments and acknowledged segments: after unified monitoring data packets are written, they first enter the unacknowledged segment; when the edge gateway completes the transmission to the cloud receiving side and receives the acknowledgment, the edge gateway marks the log segments with a sequence number less than or equal to the last measurement point confirmed in the acknowledgment as acknowledged segments and releases the storage space.

[0061] To ensure that the retransmission order is not interrupted by new messages, the edge gateway adopts a retransmission-first-then-transmission rhythm during the link recovery phase: it starts sending messages sequentially from the smallest unacknowledged measurement point of the unacknowledged segment, incrementing the sequence number, until the unacknowledged segment is acknowledged and enters the acknowledged segment, then resumes sending new messages in the generation order. If a new message is generated during the retransmission process, the edge gateway still writes the new message to the end of the unacknowledged segment, so that the retransmission and new transmission share the same sequence space. Optional implementation paths for this rhythm include: one is that the sending scheduler inside the edge gateway polls the unacknowledged segment at fixed time slices; the other is that the sending scheduler triggers continued retransmission based on the arrival of an acknowledgment event. Both paths maintain the order rule of unacknowledged segments first.

[0062] The edge gateway uses persistent logs as a local cache queue, uses dual-track confirmation of unacknowledged segments and acknowledged segments as the retransmission sequence, and sends data in a retransmission-then-send rhythm based on link recovery.

[0063] In use, the unified monitoring ensures that the order of data packets during network interruption and recovery is traceable, and the input obtained in subsequent steps will not be passively duplicated due to retransmission and new transmission. Edge gateway acknowledgment provides a basis for releasing cache space, avoiding packet loss due to blindly deleting historical packets.

[0064] Furthermore, after receiving the unified monitoring data packets, the cloud-based receiving side needs to perform deduplication and merging during the data entry stage, ensuring that even if the same packet arrives repeatedly, it only forms a single data entry semantic. To this end, the cloud-based receiving side generates an idempotent key fingerprint using a globally unique identifier and an incrementally increasing sequence number of the measurement points, and uses this idempotent key fingerprint as the data entry index. The idempotent key fingerprint uses a cryptographic hash function to calculate a digest of the concatenated input, in the form of:

[0065]

[0066] Where: idempotent fingerprint value : Outputs a hash digest, which is a hexadecimal string representing a fixed-length sequence of bytes;

[0067] Cryptographic hash function : Take the input byte sequence and iteratively calculate the digest according to a predetermined compression function, and take the value as a deterministic function; For deterministic cryptographic hash functions, the output digest length must be at least 128 bits; identical inputs will always produce identical outputs. When used for deduplication indexing, it should possess collision-resistant properties; globally unique identifier. : A unique identifier for the message generated by the edge gateway, with a value being a string sequence; incremental sequence number of the measurement point. : An incrementing sequence number generated by the edge gateway for the same measurement point, with a value that is a positive integer; idempotent fingerprint value In this context, the idempotent key remains a binary tuple. idempotent fingerprint value This is the index fingerprint of the tuple, used as a database primary key or partition key to improve retrieval efficiency. The insertion logic is based on... Review and verify to avoid relying solely on .

[0068] The cloud-based receiving side first queries the idempotent fingerprint value when inserting data into the database. Does it already exist? If not, write it into the unified monitoring data message and record the idempotent fingerprint value; if it already exists, only update the receiving trajectory corresponding to the idempotent fingerprint value. The receiving trajectory records the channel identifier, platform access time, and link status flag of this arrival, which is used to trace whether the repeated arrival occurred during the edge retransmission or the link retransmission stage. After the data is entered into the database, the cloud receiving side sends a confirmation receipt to the edge gateway. The receipt must contain at least the incrementing sequence number of the last confirmed measurement point and the corresponding globally unique identifier set, so that the dual-track confirmation mechanism can promote the release of the confirmed segment.

[0069] To ensure compatibility with parallel mapping caused by changes in object mapping version numbers, the cloud receiving side uses idempotent fingerprint values. In addition to deduplication, the spatiotemporal object key is also used as the index dimension: when the same measurement point incrementing sequence number generates different spatiotemporal object keys under different object mapping version numbers, the cloud receiving side does not regard it as a duplicate message, but as a new belonging message after the mapping change, thus preserving the parallel trajectories of the old and new mappings; subsequent steps can select the current version trajectory for time alignment and credibility calculation based on the object mapping version number, and the historical version trajectory is used for tracing.

[0070] The cloud-based receiving side generates an idempotent fingerprint value using a cryptographic hash and deduplicates it before storing it in the database. It then records the receiving trajectories of duplicate arrivals and sends acknowledgments to the edge gateway. Simultaneously, it retains parallel trajectories for the spatiotemporal object keys under different object mapping version numbers. In use, unified monitoring data packets form single-transaction semantics at the database storage level, preventing subsequent steps from generating duplicate alarms or duplicate twin status updates due to duplicate packets. The receiving trajectories provide source explanations for duplicate arrivals, facilitating on-site investigation of the triggering points for retransmission or link retransmission strategies.

[0071] Step 2: Based on the unified monitoring data messages entered into the database in Step 1, establish the alignment order between the acquisition time of the calibration equipment and the streaming data while preserving the original evidence chain, and solidify the time window statistical frames and ring number statistical frames as the input carriers for online data quality governance in Step 3.

[0072] The edge gateway collects tunnel boring machine operating parameters, structural and environmental monitoring data, and generates unified monitoring data messages, which carry the device acquisition time. With platform access time Equipment data collection time Slow drift or sudden jumps can occur due to temperature, power supply fluctuations, and restart / reset effects; platform access time Sudden increases in latency can occur due to backhaul link congestion, retransmission queue backlog, and route switching. If platform access time is used directly... Sorting and aggregation may result in data from the same ring number being split into different statistical windows; if data is directly collected by device time... In the sorting and aggregation process, systematic offsets can occur when aligning across measurement points. These offsets can be misinterpreted as structural changes during the missing data detection, drift detection, and spatial consistency checks in step three, causing passive fluctuations in the credibility score. Therefore, step two first maintains the device time deviation at the message level, then uses the correction device acquisition time to advance the water level line and solidify the statistical frames, thereby converging the arrival order uncertainty into a correctable difference within the allowable delay duration.

[0073] This involves using the unified monitoring data message entered into the database in step one as input, and utilizing the device collection time within the same message. With platform access time Arrival delay samples are constructed, and device time skew is estimated using robust quantiles. Then, the device acquisition time is... The correction process obtains the acquisition time from the calibration device. This sub-step output retains the original timestamp and correction trace, allowing audit traceability to still return to the original message from step one.

[0074] Disconnection and retransmission will cause multiple messages to arrive at the same measurement point in a concentrated manner, resulting in platform access time. The clustered distribution of the data indicates that if the mean is used to estimate the bias, the clustered distribution will skew the bias. Therefore, the cloud-based receiver first filters samples based on device and acquisition link status flags, selecting only messages with stable links and available acquisition terminals for bias estimation; subsequently, within the same spatiotemporal object key... Map version number to the same object Construct an arrival delay sample set within the specified range, and define the sample values ​​as platform access times. With equipment data collection time The difference ensures that the sample set does not contain components belonging to different categories or data types.

[0075] After the sample set is constructed, the cloud-based receiver sorts the sample values ​​and extracts the quantiles. The quantile positions are determined by the device bias quantile parameters. Control: When the link is stable, select the quantile point close to the median to suppress occasional spikes; when the link has a long tail of queues, the quantile point moves to the side with the larger value to cover the long tail.

[0076] The specific method for calculating quantiles is as follows: Arrange the sample values ​​in ascending order, and take the quantile parameter whose sorted index is not less than the total number of samples and the equipment deviation. The smallest sample value of the product is taken as the quantile result. The quantile result is then exponentially smoothed with the equipment time deviation of the previous cycle, so that the equipment time deviation slowly tracks with the slow drift, while not being pulled off by a single cluster of transmissions.

[0077]

[0078] Where: Equipment time deviation : Represents the offset of the acquisition terminal clock relative to the cloud clock. The value is a signed time quantity, and its range is [value range missing]. Equipment update time deviation : This represents the device time deviation after a smooth update, with a value range of [value missing]. Smoothing factor : is the exponential smoothing fusion coefficient. Control quantile estimation affects the time bias of the updated equipment. The impact strength; when the acquisition end is available and not in clock reset state, take the engineering configuration value; within a certain number of message cycles when clock reset state is established, take a smaller engineering configuration value;

[0079] quantile function : A function for finding quantiles in a sample set, where each value is an element in the sample set; Equipment deviation quantile parameter : represents the quantile location parameter. Arrival Delay Sample Set This is the set of differences between the platform access time and the device acquisition time, and the values ​​are a set of signed time quantities; among them, the cloud receiving side first filters samples by status flags and then uses them as the same spatiotemporal object key. Map version number to the same object Internal structure arrival delay sample set Then, using the quantile function Center estimation and smoothing factor Device update time deviation As a supplement: quantile function Executable definition for sample set: Sort in ascending order ,Pick ,definition .in The range of values ​​remains unchanged from the existing text.

[0080] Sliding window maintenance: For each measurement point label With data type identifier The system maintains a queue of delayed samples, consisting of delayed samples from the most recent packets. Sample addition and removal are controlled by either a fixed upper limit on the number of samples or a fixed time span; either method is acceptable, but the queue must be bounded. Status flag enumeration: The device and acquisition link status flags must include at least four types of enumerated values: acquisition end available state, retransmission congestion state, retransmission occurrence state, and clock reset state. The acquisition end available state determines whether a sample enters the queue. The congestion state is used to determine whether a sample enters the system. With safety margin The value range, the clock reset state is used to trigger the smoothing factor. The degradation strategy is as follows. When in use, quantile smoothing estimation suppresses the pull of offsets from the supplementary clusters and long tail pairs, so that subsequent alignment does not mistake link fluctuations for structural fluctuations; the offset update has the ability to track gradual changes, so that slow clock drift at the acquisition end is continuously absorbed without introducing abrupt jumps.

[0081] Furthermore, after the equipment time deviation is updated, this deviation needs to be applied to the equipment acquisition time of the unified monitoring data message. The data acquisition time of the calibration device is obtained.

[0082] The correction action must simultaneously meet two requirements: it can be used for windowing and it can be returned to the original chain of evidence. To this end, the cloud receiving side adds a correction device acquisition time field to the message, while retaining the original device acquisition time. With platform access time Record the time deviation of the updated equipment used in this calibration. Mapping version number to object .

[0083] When the acquisition terminal restarts or resets, it causes a delay in the device's acquisition time. During the bounce, the cloud receiving side uses the incremental sequence number of the measurement point. The continuity and device / acquisition link status flags are used to determine whether a clock reference reset has occurred; if the determination is successful, the smoothing factor for the device time deviation is updated. By taking a smaller value, the acquisition time of the calibration device gradually transitions within several messages, avoiding the clock bounce from directly spreading to the boundary of the statistical window.

[0084]

[0085] Where: the data acquisition time of the calibration device : This is the corrected event timestamp, with values ​​representing a date-time series; device acquisition time. : The raw event timestamp generated by the acquisition end, with values ​​representing a date-time series; update the device time deviation. Offset, with a value of ;

[0086] Among these features, the cloud-based receiving side adds a correction device acquisition time to the unified monitoring data message. and the original equipment acquisition time and platform access time and calculate And update the device's time deviation. and object mapping version number As a calibration trace; calibration device acquisition time With a unified timeline, subsequent windowing will unfold simultaneously according to the same event time, including the original device acquisition time and platform access time. The retention and correction actions can be traced back to the message from step one.

[0087] Furthermore, to correct the device's data acquisition time The main timeline is combined with the platform access time. The allowed late arrival time and water level time are advanced, and after the water level time is advanced, the time window statistics frame and ring number statistics frame are fixed. This sub-step limits out-of-order arrivals to within the allowed late arrival time and binds the statistics frame to its source message range, providing recalculated input for step three.

[0088] Message arrival delays exhibit a long tail, making it difficult to balance premature fixation with prolonged delay. Therefore, the cloud receiver adjusts the timing based on the data collection time of the calibration equipment. With platform access time Construct a corrected arrival delay sample set, where the sample values ​​are the platform access times. Acquisition time with calibration equipment The difference is used to exclude messages whose status flags indicate they are in a retransmission congestion segment, so that the sample set can characterize the normal latency of the current link.

[0089] The allowable delay duration is obtained by adding a safety margin to the quantile: First, the quantile value of the corresponding late arrival quantile parameter is selected from the corrected arrival delay sample set as the upper bound quantile value of the delay. Then, a safety margin is added to include the tail delay after short-term retransmissions or route switching. The safety margin is driven by the link retransmission flag and the buffer depth, and changes with the link state to avoid fixed configuration mismatch.

[0090] In the formula: Allowable lateness duration : is the upper bound for out-of-order late arrivals, a non-negative time quantity, and its value ranges from 0 to 1. Quantile function : A function for finding quantiles in the sample set, where each value is an element in the sample set, from the corrected to the delayed sample set. The upper bound quantile estimate is obtained; the late quantile parameter is obtained. : represents the quantile location parameter. Correcting the arrival delay sample set : The set of non-negative time values ​​between the platform access time and the calibration device acquisition time; safety margin. : This is the additional delay coverage, a non-negative time amount. ;

[0091] Furthermore, after the allowed lateness duration is determined, the cloud receiving side advances the water level time based on the largest correction device acquisition time among the most recently received messages. When the statistical window ends earlier than or equal to the water level time, the cloud receiving side considers the impact of late messages within that window to have converged and can proceed to the consolidation phase. The water level time is calculated as follows:

[0092]

[0093] Where: water level time : Timestamp for statistical window fixation; most recent maximum calibration device acquisition time : The maximum acquisition time of the calibration device in the most recently received messages, representing the position of the event time progression; Allowable delay duration : Same as the definition above, with different values ;

[0094] Among them, the cloud receiving side first receives the delayed sample set after correction. Through quantile function With safety margin Determine the allowable lateness duration Then, based on the most recent maximum calibration device acquisition time... Deducting allowed late time Time to obtain water level line This will lead to the standardization of statistical windows. When using them, the allowed lateness duration will be specified. Adjusting to changes in the long tail of the link, the statistical window is fixed with interpretable late boundaries, waterline time. Provide fixed judgment conditions to limit the impact of late messages to a clear window.

[0095] After the water level line is advanced, the cloud receiving side corrects the equipment's data acquisition time. As a windowed timeline, press the spacetime object key. Filter samples and solidify time window statistical frames. Short windows are used for high-frequency operating parameters, and long windows are used for low-frequency structural monitoring. The window width is indicated by the data type. Determine and map the version number to the object. Synchronous recording is included in the frame header. In addition to the mean, quantiles, and extreme values, the missing rate and effective sample rate are also recorded within the frame for missing detection and confidence assessment in step three. The trend slope is calculated using the median slope process: first, sample pairs are sorted according to the acquisition time of the calibration device, and then the median of the slope of the sample pairs is taken as the slope within the window, so that single-point spikes do not dominate the slope direction.

[0096] Form sample pairs from valid samples within the statistical frame. ,in To calibrate the device acquisition time , For all... Calculate the slope

[0097]

[0098] Take all The median is used as the trend slope. When the number of sample pairs is too large, sampling at fixed intervals is permitted as an equivalent implementation method, but the sampling intervals must be recorded for verification.

[0099] The ring number statistics frame is triggered by a change in the ring number identifier. The trigger signal comes from the ring number register of the tunnel boring machine control system or is obtained by the mileage identifier crossing a single ring length threshold. After triggering, the key statistics within the previous ring are fixed, and the globally unique identifier interval and the incremental sequence number interval of the unified monitoring data messages participating in the statistics within that ring are written into the frame header, so that step three can locate the source range of the statistics frame. To avoid object mapping version numbers... During the change, ownership mixing occurred, and the framing process used object-mapped version numbers. As an additional filtering condition, the new version belongs to a new statistical frame sequence, while the old version sequence remains traceable but does not participate in the current version's solidification.

[0100] Among them, the cloud receiving side is at the water level time. After advancement, the data acquisition time of the calibration equipment will be adjusted. With spacetime object key The frames are created by separately storing the time window statistics frames and the ring number statistics frames, and recording a globally unique identifier in the frame header. Increasing sequence number of interval and measuring point Range, and also with object-mapped version numbers Filtering avoids version mixing. When in use, the statistical frame header carries the source range, allowing step three to trace back the sample range and recalculate statistics in missing data detection and drift detection; the median slope process reduces the pull of spikes on trend judgment, making trend judgment and operating condition switching easier to distinguish.

[0101] Step 3: Follow the order of first characterizing usability and the causes of anomalies, and then outputting gating quantities to avoid filling in gaps in the facts and to avoid credibility deviating from the explanation of causes.

[0102] Jitter, disconnections, and restarts at the acquisition end during underground construction can cause missing and out-of-order remnants in the statistical frames; loose sensor connections, range switching, and changes in temperature, humidity, and preload can cause spikes, slow drift, and jamming. This is because the statistical frames have been calibrated according to the equipment's acquisition time. Alignment makes trends and abrupt changes clearer, but it still cannot distinguish whether their cause is structural change or data quality fluctuation. If data quality fluctuations are not converted into a calculable set of credibility scores and cause labels before proceeding to step four, the multi-evidence rules in step four are prone to being hindered by missing evidence, sudden spikes, or deadlocks, leading to review pressure or omissions in manual verification. The statistical frame solidified in step two is used as input, and the header of the statistical frame contains a globally unique identifier. Increasing sequence number of interval and measuring point The interval allows for backtracking to a unified monitoring data message range to pinpoint missing information. It outputs missing segment records, the effective sample rate, and a set of cause labels, serving as boundary conditions for completion and uncertainty generation.

[0103] Simultaneously, statistical frame sample counting and measurement point increment sequence numbers are used. The continuity of location is missing. A sample within a frame is considered usable only if it has a numerical value, passes basic validity checks, and the device and acquisition link status flags are in an available state; other samples are counted as missing samples.

[0104] The effective sample rate is written into the quality result object, and each sequence number jump forms a missing segment record. The missing segment record includes the start and end sequence numbers, the corresponding collection time range of the calibration device, and the status snapshot at that time. When the statistical frame is a ring number statistical frame, the missing segment record is also bound to the ring number identifier for easy reference in subsequent spatial consistency.

[0105]

[0106] Where: effective sample rate : Represents the proportion of available samples within a statistical frame. Available sample count To count the number of samples within a frame that satisfy the usable state and have valid values. Integer; missing sample count : To count the number of samples that are missing within a frame or intercepted as illegal. Integers; where the effective sample rate is calculated from the available sample count and the missing sample count of the statistical frame. And in ascending order of measurement point number The missing segment record generated at the jump position is written to the quality result object.

[0107] When used, the effective sample rate By using a unified standard to express the availability of statistical frames, gating can be performed in step four without recalculation. Missing segment records provide traceable sequence numbers and time boundaries, enabling completion and verification to locate the gap interval.

[0108] After excluding missing segments, the abnormal segmentation of valid samples is performed: spikes and spikes are identified by the steep changes of neighboring samples on the time axis of the calibration device; drift is identified by the cumulative offset in the same direction across statistical frames; jamming is identified by the change in statistical frame coverage time while the numerical quantization step remains unchanged; and over-range is identified by the triggering of the basic legality boundary.

[0109] Instead of directly deleting samples, the anomalies are written into the cause label set, so that step four can interpret the gating results based on the cause labels.

[0110] To standardize the amplitude scale across different data types, a bounded mapping function is used to map the anomaly intensity ratio to a potential energy value, and a potential energy threshold is used to trigger suspected spike, suspected drift, and suspected deadlock labels. The reference amplitude and reference duration values ​​are identified by data type in the engineering configuration. For index solidification: the reference range can be derived from the rate of change constraint of this data type and the statistical frame span, or from the upper limit of the sensor range and the acquisition cycle; the reference duration is determined by the inspection cycle of the construction shift and the minimum allowed continuous observation cycle of this data type, and varies with the object mapping version number. Leaving traces facilitates the use of different reference scales after the measurement points are moved.

[0111] Simultaneously, a spatial consistency pre-check is performed on the ring number statistics frame. If the trend of the reference measurement points in the same or adjacent rings does not support the current anomaly, spatial inconsistency is written and the reference index is recorded. The selection of reference measurement points is subject to the spatiotemporal object key. Mapping version number to object Constraints are in place to prevent neighborhood mixing caused by measurement point migration.

[0112]

[0113] In the formula: bounded mapping function To map nonnegative ratios to A deterministic function, taking values The ratio of anomaly intensities is converted into a unified potential energy scale; the ratio This represents the ratio of the abnormal amplitude to the reference amplitude or the ratio of the duration to the reference duration. ; with bounded mapping functions The ratio of abnormal intensity is converted into potential energy. Peaks, drifts, jams, over-range, and spatial inconsistencies are written into the cause label set and reference index. When used, the cause label set directly displays the source of the anomaly. Subsequent gating does not require determining the cause based on numerical fluctuations. The potential energy scale unifies the intensity caliber of different data types, and the weight configuration is not limited by differences in units.

[0114] As an example: During a break in the assembly of a certain section, the on-site monitor discovered that a segment convergence data acquisition device was displaying the same reading for an extended period, but the device's self-test status indicator remained active. Step two solidified the time window statistical frame covering this period. Step three output that the effective sample rate did not decrease, but identified a stuck pattern where the quantization step remained unchanged and the coverage time increased. A "stuck suspected" label was written into the cause label set, and this label was written along with the quality result object to a record under the same spatiotemporal object key. Based on this, on-site personnel replaced the data acquisition device's power adapter and reset the data acquisition device. In subsequent statistical frames, the "stuck suspected" label disappeared, and the quality result objects continued to be generated continuously along the same index, facilitating confirmation that the action taken had been reflected at the data level.

[0115] Based on the missing segment records and the set of cause labels, hierarchical completion is performed to generate uncertainty intervals, following the order of filling in the morphology of short missing records, filling in the correlation of medium missing records, and leaving gaps for long missing records. Short missing records rely on the morphology of samples before and after the same measurement point, medium missing records rely on the correlation of the same or adjacent rings with reference measurement points, and long missing records are not forcibly filled but are expressed as unavailability through uncertainty intervals. Finally, the combined results of the confidence score, effective sample rate, uncertainty interval, set of cause labels, and completion method identifier are output.

[0116] For short missing segments, continuous valid samples still exist on both sides. Therefore, morphological constraint interpolation is used: endpoint values ​​and slope directions are extracted on both sides of the missing segment as boundary conditions to generate a completion sequence within the missing segment, ensuring the completion sequence is continuous with the trend directions on both sides at the endpoints. In engineering implementation, either piecewise cubic Hermitian interpolation or monotonic preservation interpolation can be used. When the trend directions on both sides are opposite and the missing segment is too long, it is downgraded to endpoint preservation, and the missing completion is written into the cause label set to avoid generating corner artifacts at trend reversal points.

[0117] The missing data persists for a long time, and relying solely on endpoints is insufficient to maintain its shape. Therefore, reference-based completion is adopted: based on the reference index of sub-step 301, the version number of the same object is mapped. The system extracts statistics from similar measurement points within the same or adjacent rings, and calculates the minimum absolute deviation regression coefficient for the band boundary. Both the regression coefficient and the completed value are constrained by the basic legality boundary. The solver can choose to use the simplex method for linear programming or the interior point method for convex optimization with band boundaries. When the reference sample is insufficient, the regression completion is not output, but the gap is retained and the uncertainty interval is expanded. After completion, the completion method identifier, reference index, and completion interval index are written to the quality results object, making the source of the completion traceable.

[0118] Short missing values ​​are filled using morphological constraint interpolation and identified by the completion method. Medium missing values ​​are filled using reference-related minimum absolute deviation calculation and identified by the reference index and completion interval index. In use, short missing value completion maintains the continuity of the endpoint trend direction, reducing the interference of the completion sequence on subsequent trend criteria. Medium missing value completion writes the reference source and solution constraints into the index, giving the completed value an engineering-related interpretation path.

[0119] Furthermore, the resulting missing potential energy, peak potential energy, drift potential energy, and stuck potential energy are weighted and aggregated into a confidence score, which is then coupled with the uncertainty radius in the output. (Missing potential energy) Peak potential energy Drift potential energy , potential energy All are bounded mapping functions The values ​​are obtained by mapping the corresponding intensity ratios, and the range of values ​​is [missing information]. The weighting coefficients are used to express the emphasis of different anomalies on credibility. The credibility score uses an exponential decay form, so that credibility decreases monotonically as the potential energy increases, thus satisfying the monotonicity requirement of gating.

[0120]

[0121] Where: Credibility score : A numerical value characterizing the reliability of statistical frame observations, with values ​​ranging from 0 to 10. As one of the confidence-gated inputs in step four; weighting coefficient : represents the missing potential energy weight, with a value of Weighting coefficients : This represents the peak potential energy weight, with a value of [value missing]. Weighting coefficient : This represents the drift potential energy weight, with values ​​ranging from 0 to 10. Weighting coefficient To control the potential energy weight, the value is set as follows: Weighting coefficient to : Identified by data type in the project configuration table The configuration is based on the sensitivity of this anomaly pattern to misjudging this data type, and the missing potential energy. : This is the missing value intensity normalization factor, taking values ​​of 100%. Peak potential energy : is the peak intensity normalization factor, with a value of Drift potential energy : This is the normalized value for drift intensity, taking various values. ; potential energy stuck : This is the normalized value of the jamming strength. ;

[0122] The uncertainty interval is centered on the completed observed value, and upper and lower bounds are formed by extending the uncertainty radius upwards and downwards. The uncertainty radius varies with... It increases with size, and fractional saturation ensures that the interval does not expand unbounded:

[0123]

[0124] Where: Uncertainty radius : represents the half-width of the uncertainty interval, with a value of ;Minimum radius : This represents the lower bound of the uncertainty radius, and its value is [value missing]. Preserve the minimum uncertainty introduced by measurement resolution; maximum radius : This represents the upper bound of the uncertainty radius, and its value is [value missing]. Credibility score As defined above, the value is... , which is used as a coupling variable in the calculation of the radius of uncertainty;

[0125] Shape parameters : is the fractional saturated shape parameter, with a value range of . ; , , : Take the minimum resolvable change range or a multiple thereof corresponding to the sensor resolution; Choose the smaller of the upper limit of the sensor's measurement range and the allowable envelope in engineering. Used to control from Towards Approximation speed, identified by data type in the project configuration table. Configuration.

[0126] The quality result object writing phase uses the spatiotemporal object key. Mapping version number to object Index, write credibility score Valid sample rate Uncertainty radius The system includes a set of cause tags, completion method identifiers, reference indexes, and missing segment indexes, enabling step four to directly access and backtrack the calculation path within the same index system.

[0127] Among them, the credibility score is generated by the potential energy weighted index mapping. And scored based on credibility. With shape parameters Coupling generates uncertainty radius Then, the quality result object is written and the index is kept consistent.

[0128] As a supplement: missing potential energy : Incrementing sequence number of measurement points in the statistical frame Locate the largest consecutive missing segment in the sequence, and take the ratio of the acquisition time span of the correction device covered by the missing segment to the reference duration configured for this data type. and take The reference duration is indicated by the data type. The corresponding acquisition period and statistical window length are derived, or directly given by the engineering configuration table. Peak potential energy : Calculate the maximum absolute value of the difference between adjacent samples in the valid sample sequence of the statistical frame, and use the reference amplitude configured for this data type as the normalized denominator to form the ratio. ,Pick The reference range is derived from the rate of change constraint and the acquisition period, or from the upper limit of the sensor range and the safety factor.

[0129] Drift potential energy The trend slope is calculated using the median slope procedure on the valid samples of the statistical frame. That is, the slope is calculated for all sample pairs and the median is taken as the in-window slope. Then, a ratio is formed using the reference slope configured for this data type. ,Pick The reference slope can be obtained by converting the design allowable rate of change or construction control indicators.

[0130] Stalled potential energy Within a statistical frame, the numerical range is calculated. If the range does not exceed the quantization step corresponding to the sensor resolution, the state is considered to be in an invariant state. The ratio of the duration of this invariant state to the reference duration is taken as the ratio. ,Pick The quantization step is given by the sensor specifications or derived by inversely from the mode of adjacent differences in historical stable segments.

[0131] When using it, the credibility score The uncertainty radius remains monotonically decreasing for multiple anomalies, ensuring that the fourth-step gating can be filtered using a single quantity without deviating from the causal label; saturation The extension allows the interval boundary to be automatically expanded while maintaining a controllable upper bound when the confidence level of the same observation decreases.

[0132] Step 4: Starting with the quality result object and proceeding along... and Merging is used to implement credibility gating and diversion, structural risk fusion and judgment, event orchestration and event version update based on quality result objects.

[0133] Shield tunnel monitoring data in the same spatiotemporal object key The alarm results may reflect changes in structural status, link jitter, retransmission congestion, and abnormal sensor behavior. If out-of-bounds values ​​are directly interpreted as structural risks during alarm generation, spikes, jams, and long missing retransmissions will generate a dense array of alarms in a short period, leading to system shutdowns for review. If the overall alarm threshold is raised, the risk of missed alarms will be amplified in sections passing under buildings and adjacent pipelines. Therefore, step four first focuses on the quality results. , , Construction gating margin The observations are categorized into data quality events, early warning observations, and structural decision candidates. Multi-evidence fusion is then performed only on structural decision candidates, within an allowed delay period. Use event version number within the boundary Record the revisions to the conclusions so that on-site handling is always based on a reproducible chain of evidence.

[0134] Taking the quality result objects and their corresponding statistical frames as input, the output consists of three types of initial records: data quality events, observation warnings, and structural judgment candidates. First, the observation availability and uncertainty are converted into gating margins. Therefore, subsequent fusion determination is only triggered when the chain of evidence is stable. Specifically, within the same statistical frame, the credibility score... The decrease reflects significant abnormal morphology, and the effective sample rate The decrease reflects an increase in missing and intercepted data, and the radius of uncertainty. Increase the level of evidence completion or broaden the boundaries of evidence. Adjust the credibility score. Valid sample rate logarithmic odds and The normalized penalty is synthesized into the gating margin. This allows for the simultaneous expression of three types of influencing factors within an additive scale, and the binding of the diversion boundary to the data type to accommodate the differences between high and low frequencies.

[0135] To ensure the logarithmic function The domain holds true when calculating the gating margin. First, score the credibility. With effective sample rate Perform boundary clipping: when credibility score or effective sample rate Less than the lower bound of the cutoff Take the lower bound as the time limit. When credibility score or effective sample rate Greater than Time to take This avoids the denominator being zero or the logarithm diverging.

[0136]

[0137] Where: Gating margin Three-state current splitter, with a value of This determines whether an event enters the data quality branch, observation branch, or structure branch; credibility score. The original value range is: When participating in the gating margin calculation, the values ​​after boundary clipping are used to satisfy the domain of the logarithmic term; effective sample rate The original value range is: When participating in the gating margin calculation, the value after boundary clipping is used to satisfy the domain of the logarithmic term;

[0138] Uncertainty radius : As a penalty term, it suppresses high-uncertainty observations from entering structural branches; maximum radius : ,right Normalization to ensure the scaling stability of the penalty term; coupling coefficient : Value ,adjust Gating margin Weights in the formula; penalty coefficient : Value Cut off the lower boundary : Input the trimming parameters for the logarithmic terms, with values ​​ranging from 1 to 2. Credibility rating With effective sample rate Crop to To avoid logarithmic divergence; clipping parameters Given by the engineering configuration and in calculating the gating margin Before cropping, the credibility score after cropping With effective sample rate Enter the logarithmic probability term.

[0139] Complete gate control margin After calculation, using the gate threshold one With gate threshold 2 Three-state splitting is formed: Gating threshold one : Lower bound threshold for gating shunt, value Define the entry boundary for data quality event branches; gating threshold two : Upper limit threshold for gating and shunting, value Define the entry boundary of candidate branches in the structure determination; when Classified as a data quality event, when When classified as an observation and early warning system, Enter the structural decision candidate. Gating threshold 1 With gate threshold 2 Configure by data type and map version number to object. Leaving traces ensures that the gating boundary is distinguishable before and after the measurement point migration. Calculate the gating margin. and with a gate threshold Gating threshold 2 right Perform three-state splitting.

[0140] In use, the gating margin incorporates confidence, sample sufficiency, and uncertainty into the same criterion, making the diversion boundary configurable and repeatable. The gating threshold is bound to the data type, ensuring that high-frequency operating parameters and low-frequency structural monitoring use different diversion scales without mixing them.

[0141] Furthermore, the branches resulting from the gating system must be entered into the event ledger so that on-site personnel can understand the nature of the event and take appropriate action by following the same entrance.

[0142] First, determine the event index key generation rules: the event index key is derived from the spacetime object key. The statistical frame consists of the event type and the end time of the statistical frame. The end time of the statistical frame is taken from the acquisition time of the calibration device. The window end value is truncated to a fixed time slot, so repeated triggers within the same time slot will point to the same event index key. The cause label set, missing segment index, and reference index are then written into the event evidence field, creating a closed-loop chain of trigger cause - evidence index - disposition record.

[0143] For the data quality event branch, the set of cause labels is mapped to an enumeration field of handling actions. This enumeration field covers four types of on-site actions: collector power supply check, communication link check, collector reset, and retesting measurement points. Selection records are retained in the event ledger, ensuring a one-to-one correspondence between investigation actions and event entries. For the observation and early warning branch, hysteresis confirmation is used: the same event index key only enters the structural judgment candidate if it remains in the observation and early warning branch for multiple consecutive time slots; as soon as any time slot falls back to the data quality event branch, the observation count is reset to zero, and the set of fallback cause labels is recorded. Hysteresis confirmation can be implemented by relational database triggers or by a message queue consumer maintaining the count and periodically storing it in the database. Both paths maintain the same criterion, allowing different projects to choose based on their existing IT infrastructure.

[0144] Specifically, the process involves generating an event index key and writing it to the event ledger evidence field; enumerating the handling actions for data quality events; and implementing lag confirmation for consecutively established observation warnings and recording the reasons for the decline. The event index key does not include the event type; the event type is presented as an event ledger field along with the event version number. Updated in response to changes. The event index key is derived from the spatiotemporal object key. Object mapping version number The time slot number is determined by the time collected by the calibration equipment. It is obtained by truncating the time slot width configured for the project.

[0145] In use, the event index key converges discrete triggers into traceable entries, allowing the field end to see manageable events rather than scattered messages. The action enumeration is bound to the cause label set, ensuring that troubleshooting actions are triggered based on causes rather than being driven by instantaneous values.

[0146] As an example: During a night shift in a certain section, an observation warning entry appears in the background and then reverts to a data quality event in the next time slot. The shift worker opens the event ledger entry, reads the link anomaly and missing completion tags in the evidence field, and sees that the missing segment index points to a range of increasing sequence numbers for measurement points. The shift worker notifies the inspection personnel to check the communication converter and network cable crimping terminals in the section monitoring room. The inspection personnel reset the communication converter and re-crimp the terminals, then select the communication link check in the interface to complete the handling action and fill in the description. The same event index key will not revert in subsequent time slots, the observation count will be accumulated again, and the handling record and status change of this entry can be directly viewed during shift handover.

[0147] Furthermore, using the same spatiotemporal object key The statistical frame is used as the calculation object, and the cause label set and uncertainty radius are referenced in the quality result object. As constraints, four types of evidence are extracted: trend evidence comes from the ratio of the trend slope and relative rate of change of the increment within the statistical frame to the historical baseline; consistency evidence comes from the degree of support of the trend direction in the statistical frame pointed to by the reference index; working condition evidence comes from the correspondence between the tunnel boring machine's operating parameters (TBM) operating status, shutdown, grouting status, and structural changes on the time axis; and baseline evidence comes from the deviation strength of the current statistical frame relative to the historical baseline envelope. To unify the scale, a fractional saturation mapping is used to map the evidence strength ratio to... :

[0148]

[0149] In the formula: bounded mapping function : Deterministic fractional saturation function ;ratio The ratio of the strength of evidence, and its range. As The input represents the strength of evidence; the four types of evidence are denoted as trend evidence. Consistency evidence Working condition evidence Baseline evidence And synthesize the risk composite value by complementary product. :

[0150]

[0151] Where: Risk composite value Structural risk synthesis discriminant, value As a core field for structural risk warning levels and evidence summaries; trend evidence : Value This expresses the strength of evidence regarding the cumulative changes of monitoring indicators over time; the ratio is formed by taking the absolute value of the trend slope of the statistical frame and the reference slope of that data type. and take The trend slope is calculated using the median slope process described in step two.

[0152] Consistency evidence : Value This expresses the degree of trend support within the same or adjacent rings; using the statistical frames of the same or adjacent rings pointed to by the reference index as the reference set, the proportion in the reference set that is consistent with the current trend direction is calculated and directly used as... Working condition evidence : Value This expresses the degree of matching between structural changes and operating condition switching; it identifies three types of state segments—propulsion, shutdown, and grouting—from the statistical frames of operating parameters, and maps the strengthening trend of structural indicators to the time overlap ratio of the propulsion / grouting segments. ;

[0153] Baseline evidence : This expresses the strength of evidence for deviations from the historical baseline envelope; using the historical baseline envelope as a reference, the ratio is formed by the ratio of the current statistical frame index's deviation from the envelope to the allowable range of the envelope. and take Complementary product terms The range of values ​​is To achieve nonlinear synthesis of multiple pieces of evidence;

[0154] Based on this, an evidence digest package is generated: the evidence digest package contains the source statistics frame index, reference index, and object mapping version number for each type of evidence. , Cause label set and uncertainty radius This allows reviewers to trace back to the statistical frame and unified monitoring data message interval along the index and recalculate the evidence. The evidence summary package can be stored as a relational database structured field or written to the append log and parsed at the interface layer. Both paths maintain the consistent constraints of traceable source index and recalculated evidence fields.

[0155] Among them, the bounded mapping function Unify the evidence scale and synthesize the risk composite value by complementary product. Then, a version number containing the source index and object mapping is generated. The evidence summary package is bound to the event. When used, multi-evidence synthesis prevents a single anomaly from directly generating a structural risk alarm, ensuring that alarm conclusions are constrained by both spatial and operational condition evidence. The evidence summary package binds statistical frames and message intervals using an index, allowing reviews to be recalculated along the same evidence chain without relying on verbal retelling.

[0156] Furthermore, the generated event candidates are transformed into manageable events. First, for cases where the same event index key is triggered consecutively in adjacent time slots, aggregation is performed based on temporal continuity and spatial adjacency: temporal continuity is determined by the consecutiveness of the time slot sequence, and spatial adjacency is determined by the simultaneous satisfaction of the difference between the ring number identifier and the difference between the mileage identifier; after aggregation, the statistical frame indexes participating in the aggregation are written into the aggregation index field to ensure that the source is not lost.

[0157] Secondly, maintain the event lifecycle status. The four statuses—New, Confirmed, In Progress, and Closed—are all triggered by a handling record. The handling record is generated by on-site personnel selecting handling actions and filling in descriptions on the interface. Let the time slot width be... The reference starting point is Define the time slot number:

[0158]

[0159] in This is the floor function. and For engineering configuration, Can be identified by data type Configuration.

[0160] Subsequently, within the allowed lateness period Changes in conclusions caused by delayed correction within the boundary: When the gating margin corresponding to the same event index key... or risk composite value When an event version number is used to indicate a delayed statistical frame that crosses an existing split boundary, or when the source index set of the evidence digest packet changes. Add one, and archive the old version of the evidence digest package as a historical version; when the late correction only changes minor fields and does not cross boundaries, the event version number is... Remain unchanged but with added revision notes.

[0161] Event version number Mapping version number to object At the same time, records should be kept to avoid misinterpreting changes in attribution caused by the relocation of measurement points as delayed corrections.

[0162] There are two possible implementation paths for this project: one is to use the event index key and the event version number in a relational database. First, a composite key is formed, and transaction locks are used to ensure single-version generation; second, in the append log, the event index key is used as the partition key, and the event version number is used as the partition key. The write operation is monotonically incremented, and then repeatedly written by the consumer according to idempotency rules. Both paths ensure that the version number is monotonically incremented and that any version can be traced back to the evidence digest package and source index.

[0163] This involves aggregating and maintaining the lifecycle state transition based on temporal continuity and spatial adjacency, and allowing for a certain delay duration. Internal basis , Version number of the event triggered by the change in evidence summary Monotonous updates. When in use, deduplication aggregation converges dense triggers into event entries and retains the aggregation index, so that the source chain is not lost. Lifecycle states bind the handling actions to the event entries, so that the handover of shifts can recount the handling process along the state chain.

[0164] Step 5: Using the event ledger as the entry point, assemble structural risk alarms and observation warnings into structural twin state records, and assemble data quality events into sensor twin state records; categorize them according to the data collection time of the calibration equipment. Sort playback and time at water level Previously, a stable boundary was used, and continuously triggered data quality events were mapped to operation and maintenance tasks, and the handling actions were enumerated and written back.

[0165] In shield tunneling and tunnel section monitoring and management, two types of usable information are typically required simultaneously on-site: First, structural status information for construction control and risk management, which needs to be consistently presented across Building Information Modeling (BIM) components and ring number management units; second, data acquisition link and sensor status information for equipment maintenance and data interpretation, specifying whether an anomaly stemmed from link jitter, acquisition terminal restart, jamming, or drift. If these two types of information are combined into a single risk status, data quality events will contaminate the structural status, leading to frequent flipping of the structural twin identifier. Conversely, if the two types of information are completely separated into different systems, on-site personnel will be unable to correlate a risk alarm with its corresponding credibility score. With uncertainty radius Correspondingly, the processing and review would lose their shared chain of evidence. Therefore, step five uses the event ledger from step four as input, allowing only the event ledger to drive the twin object state library to write, and specifying the event version number. As a coverage order, it ensures that new conclusions generated after delayed correction for the same event can overwrite old conclusions, while retaining old conclusions as historical versions. Then, data quality events are transformed into manageable operational tasks, and the operational tasks are written back to the sensor twin state. This ensures that the structural twin state only expresses the evidence chain of structure and operating conditions, and the sensor twin state only expresses the evidence chain of data acquisition links and equipment health, thereby avoiding conceptual confusion.

[0166] By using dual-track writing of structural twin state records and sensor twin state records, the presentation of structural state and data quality processing are ensured to operate without interference; through event version numbers... The coverage order and playback folding rules ensure that the conclusions after late corrections remain consistent and traceable in the twin object state database; by recording and writing back the operation and maintenance tasks and enumerating the handling actions, the link and sensor maintenance actions form a closed loop trace in the twin state, which facilitates subsequent event review and responsibility handover.

[0167] The event entries in the event ledger are translated into state records in the twin object state library, ensuring that the state records can be reconstructed during subsequent playback. This sub-step first establishes index boundaries and version isolation rules, then assembles the fields in the event ledger into incremental update packages and writes them to the twin object state library.

[0168] Building Information Model (BIM) component identifiers may be replaced during model revisions, and measurement points may be migrated due to construction interference. Step one involves mapping version numbers to objects. The old and new mappings have been saved in parallel. If the twin object state library is written only according to the component identifier, model revisions will overwrite the historical state onto the new components, thus losing the traceability boundary. Therefore, the main index of the twin object state library is fixed as the spatiotemporal object key. Mapping version number to object The combination of ring numbers and the use of ring number identifiers as secondary indices allows the same component to form discrete state slices under different ring numbers.

[0169] In terms of state record structure, the twin object state library is divided into two record types: structural twin state records and sensor twin state records. Both share the spatiotemporal object key. Mapping version number to object However, the field sets differ. The structural twin state record must contain at least a risk composite value. Gating margin Credibility score Valid sample rate Uncertainty radius Calibration equipment acquisition time Event version number With evidence summary package index; the sensor twin state record should at least include a set of cause labels, a missing segment index, a reference index, an enumeration of the most recent action, and the acquisition time of the calibration device. With event version number Both types of records share a globally unique identifier. Increasing sequence number of interval and measuring point The interval ensures that any state can be traced back to the range of statistical frames and unified monitoring data messages.

[0170] In terms of project implementation, the twin object state library can be implemented using relational database tables or key-value stores. When using relational database tables, the spatiotemporal object key is used as the storage location. Object mapping version number The ring number identifier and record type are used as a combined unique key, and the event version number is used as the key. As a version column; when using key-value storage, use the spacetime object key. Mapping version number to object Concatenate the primary key and write the ring number identifier and record type into the column family, enabling reads to scan by ring number range. Both paths require the version column to retain historical versions and do not allow overwriting or deletion.

[0171] Among them, the spatiotemporal object key Mapping version number to object Establish a master index for the twin object state database, and distinguish between structural twin state records and sensor twin state records by record type, and by event version number. Keep historical versions.

[0172] When using, the object mapping version number The isolation ensures that measurement point migration and model revision will not cover historical states, and the state traceability boundary is clear; the structural twin state record and the sensor twin state record share an index but the fields are separated, so that the structural state will not be contaminated by data quality events, while preserving the same source evidence chain.

[0173] Transform the event ledger output from step four into an incremental update package that can be written to the twin object state library. This is because the same event index key can be updated within the allowed lateness period. Multiple event version numbers may appear within. If events are appended directly without defining an overwrite order, the twin object state library will retain multiple conflicting current states at the same time. Therefore, the event version number... As an overwrite order, only keys of the same spatiotemporal object are allowed. Mapping version number to object Below, with larger event version numbers Cover smaller event version numbers The current pointer is used to preserve the overwritten version as a historical version for replay.

[0174] The assembly of incremental update packages follows a defined mapping from event type to record type. For data quality events, the incremental update package is only assembled as a sensor twin state record, with the cause label set and action enumeration as the main fields, and the gating margin is also included. With credibility score Write auxiliary fields to ensure that the same data quality event still corresponds to the gating boundary on the interface. For observational warnings and structural risk alerts, incrementally update the package as a structural twin state record and incorporate the risk composite value. Evidence summary package index and uncertainty radius Write the main fields; if the cause tag set contains spatial inconsistencies, write a reference index in the structural twin state record so that the structural determination can be traced by the reviewer.

[0175] Idempotent overwrite can be implemented in two ways in engineering. One approach is to use the insert or update semantics of relational databases: using the spatiotemporal object key. Object mapping version number Locate records by ring number identifier and record type, and compare event version numbers. Only when the newly written event version number Greater than the existing event version number The current pointer is updated periodically; secondly, monotonic write semantics of appending logs are adopted: incremental update packets are updated according to the data collection time of the correction device. Append to write, and record the event version number on the read end. The current state is collapsed. Both paths ensure that duplicate incremental update packages of the same version will not change the result, thus maintaining consistency with the idempotent data insertion semantics of step one.

[0176] Specifically, the incremental update package is written to the structural twin state record or the sensor twin state record according to the event type, and is marked with the event version number. Update the current pointer in the overwrite order while retaining historical versions.

[0177] When using, the event version number The coverage order allows new conclusions after late corrections to replace old conclusions as the current state, while retaining historical versions for retrospection; the fixed assembly mapping of structural twin state records and sensor twin state records prevents the same event from being written into both types of states simultaneously, thus avoiding interpretation conflicts.

[0178] As an example: During a tunnel boring machine (TBM) shift that passes under a sensitive building, the shift operator sees a structural risk alarm on the interface and opens the evidence summary package, then notifies the TBM operator of the alarm information. The TBM operator adjusts the tunneling and grouting operations according to the shift's process requirements and fills out the handling instructions. Simultaneously, the monitor detects a key object in the same time and space. A data quality event occurred, with the cause tag set indicating a link anomaly. The monitor went to the monitoring room to check the data acquisition device's wiring terminals and reset the data acquisition device. After the handling was completed, the duty officer recorded the handling action enumeration in the event ledger. Step five: Write the structural risk alarm to the structural twin status record, and write the data quality event and handling action to the sensor twin status record; when a subsequent late statistical frame triggers the event version number... Add and revise the risk composite value At that time, step five uses a larger event version number. The current pointer of the overriding structural twin status record is updated, and the risk label of the same component on the interface is updated accordingly, while the sensor twin status record retains the original handling record and is not overwritten.

[0179] Furthermore, using historical versions and event ledgers in the twin object state database as input, and according to the data collection time of the calibration device... Reconstruct the structural twin state and sensor twin state at any time, and transform continuous data quality events into operation and maintenance closed-loop tasks, forming a complete chain of handling actions and state changes.

[0180] Read the same spatiotemporal object key from the twin object state library. Mapping version number to object All historical version records, and based on the data collection time of the calibration device. The sequence is sorted to form the replay sequence. Since step four allows for a specified lateness period... Internally generated event version number The revision performs version folding on the same event index key during playback: if multiple event version numbers exist within the same calibration device acquisition time slot. Then select the largest event version number. As the effective version of this slot, it is written to the playback cache; if the playback range crosses the waterline time... The previous interval is considered to be the event version number within that interval. It is now stable and will not be subject to further revisions, thus avoiding repeated jitters during historical playback.

[0181] To reduce playback costs, playback snapshots can be generated at ring number boundaries. The playback snapshot uses the end time of the ring number statistics frame as the snapshot time, and the current pointers of the structural twin state record and the sensor twin state record at that time are fixed as snapshot records. When it is necessary to replay the state at a certain time, the most recent snapshot record is read first, and then the incremental update packets after the snapshot are replayed. Playback snapshots can be written to relational database tables or appended to logs using spatiotemporal object keys. Storage should be partitioned. Both paths require snapshot records to retain the referenced event version number. The evidence summary package index ensures that the snapshot is interpretable.

[0182] Among them, according to the data collection time of the calibration equipment Sorting replays and ranking them by the largest event version number within the same time slot. The folded version is activated, and a playback snapshot is generated at the ring number boundary to shorten the playback chain.

[0183] index key of the same event In the same time slot When there are multiple versions of memory, the effective version number is defined as:

[0184]

[0185] Only use version number during playback. The evidence summary package and the twin state update package. The playback snapshot is generated at the end of the ring number statistics frame, and the snapshot record saves each at that time. and The current pointers of the lower structure twin state record and the sensor twin state record and their corresponding pointers. It also saves the evidence digest package index to ensure that the snapshot is interpretable.

[0186] When used, the version folding rule ensures that the replay results are consistent with the final conclusion of the event ledger and are not affected by intermediate revisions; waterline time. As a stable boundary, it ensures that historical replay remains deterministic within the convergent evidence range, preventing passive flipping; the replay snapshot reduces the cost of long interval replay, enabling on-site review to quickly locate state evolution at the component and ring number granularity.

[0187] Transform data quality events into closed-loop operational tasks and write the handling process back to the sensor twin status record, ensuring a closed loop of anomaly occurrence, investigation, handling, and recovery confirmation within the same indexing system. Triggering conditions utilize the continuity and aggregation results of the event ledger: when the same spatiotemporal object key... When data quality events occur consecutively in adjacent time slots, and the cause tag set includes any of the following: link anomaly, suspected deadlock, or missing data completion, an operation and maintenance task record is generated and bound to the event index key and the current event version number. The operation and maintenance task log includes the task type, a suggested action enumeration, and a description of the on-site location. The on-site location description is taken from the spatiotemporal object key. The section markers, mileage markers, and ring number markers in the system eliminate the need for inspection personnel to find locations from external tables.

[0188] The write-back process follows the principle that actions are selected on-site. After completing the inspection, the inspector selects an action enumeration on the interface and enters a description. The system then writes the action record to the event ledger and simultaneously writes it to the most recent action enumeration field of the sensor twin status record, along with the acquisition time of the calibration equipment where the action occurred. When the gating margin in subsequent statistical frames Rebound and credibility score When the system recovers to the available range, it writes a completion flag to the maintenance task record and a recovery flag to the sensor twin status record. The recovery flag still references the version number of the event that caused the loop closure. This ensures that the closed-loop criteria are traceable.

[0189] In the project implementation path, operation and maintenance task records can be stored in a relational database and used to drive interface reminders with status fields, or they can be written to append logs and pushed by on-site terminals; the enumeration of handling actions can be configured by the project management system, but it must be consistent with the event ledger fields in step four to ensure that the handling write-back does not introduce new terms.

[0190] Specifically, maintenance task records are generated based on the continuous triggering of data quality events and the set of cause tags. Inspection and handling actions are enumerated and their descriptions are written back to the event ledger and sensor twin status records. Meanwhile, a gating margin is used... With credibility score Recovery is used as the criterion for closed-loop completion. In use, the maintenance task log transforms data quality events into executable actions, providing a clear entry point for on-site troubleshooting and binding it to the event index key; the write-back process ensures that the sensor twin status record reflects the latest maintenance actions, allowing subsequent anomalies to be directly linked to previous handling procedures; the closed-loop criterion references the gating margin. With credibility score This ensures that the restoration confirmation is based on a common chain of evidence rather than subjective judgment.

[0191] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0192] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0193] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0194] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0195] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A BIM-based twin-driven shield tunnel monitoring data stream processing system, characterized in that: include, The edge gateway generates unified monitoring data packets, including a globally unique identifier, an incremental sequence number of the measurement point, the device acquisition time, the platform access time, and a spatiotemporal object key. The packets are retransmitted after a connection failure, and the cloud stores them in the database using the globally unique identifier and the incremental sequence number of the measurement point. The time deviation of the equipment is estimated based on the measurement points and the equipment acquisition time is corrected to obtain the correction event time. The unified monitoring data messages are aggregated according to the correction event time to generate time window statistical frames and ring number statistical frames, and the allowable late time correction is set. Data quality management was performed on both time window statistical frames and ring number statistical frames, and missing data was filled in to generate a quality result object. The quality result object includes a confidence score, a set of cause labels, an effective sample rate, and an uncertainty interval. Based on the quality result object, perform credibility gating and diversion, and generate data quality events, observation warnings or structural risk alarms. Aggregate and write them to the event ledger by spatiotemporal object key. When late correction causes changes in statistical frames, generate version replacement. Establish a twin object state library; incrementally write according to the new version of the event ledger; the structural twin state is updated by observation warnings and structural risk alarms; the sensor link state is updated by data quality events; and replay according to the correction event time.

2. The shield tunnel monitoring data stream processing system according to claim 1, characterized in that: The unified monitoring data message includes a globally unique identifier, a measurement point incrementing sequence number, device acquisition time, platform access time, device and acquisition link status flags, a spatiotemporal object key, and an object mapping version number. Before encapsulation, the edge gateway performs field alignment, unit dimension unification, and basic legality verification on the fields. When the network is interrupted, the message is written to the local cache queue and retransmitted according to the measurement point incrementing sequence number after the link is restored.

3. The shield tunnel monitoring data stream processing system according to claim 2, characterized in that: The cloud receiving side uses a globally unique identifier and an incremental sequence number of measurement points as idempotent keys to perform deduplication and merging on duplicate messages formed by disconnection and retransmission, and then writes them into the unified monitoring data message storage area. The device acquisition time and platform access time are retained along with the merging result and stored in association with the spatiotemporal object key and the object mapping version number.

4. The shield tunnel monitoring data stream processing system according to claim 3, characterized in that: For each measuring point, the device time deviation is maintained. The slow drift is estimated by the difference between the access time and the device acquisition time of the sliding statistics platform, and the device time deviation is updated to obtain the correction event time. When the status flag of the device and the acquisition link changes, the device time deviation is re-initialized, and the device acquisition time and the correction event time are written into the unified monitoring data message at the same time.

5. The shield tunnel monitoring data stream processing system according to claim 4, characterized in that: Based on the correction event time, the unified monitoring data messages are stream-aligned and aggregated to generate time window statistical frames. Different statistical periods are set for the tunnel boring machine operating parameters and structural monitoring data. The allowable delay time is dynamically adjusted based on link health indicators such as delay distribution, packet loss, retransmission, and buffer depth. Late data arriving within the allowable delay time is used to correct the statistical results of the corresponding time window statistical frames.

6. The shield tunnel monitoring data stream processing system according to claim 5, characterized in that: When an increment in the ring number is detected, the ring number event window is triggered, and the mileage that crosses the ring length threshold is recorded at the same time. The key statistics during the previous ring are solidified into the ring number statistical frame. The key statistics include the mean, fluctuation and slope. The ring number and mileage identifier, missing rate and effective sample rate are written into the ring number statistical frame. The ring number statistical frame is associated with the spatiotemporal object key and written into the statistical frame storage area.

7. The shield tunnel monitoring data stream processing system according to claim 6, characterized in that: A data quality governance operator chain is executed on the time window statistical frames and ring number statistical frames. The operator chain includes missing detection, spike detection, drift detection, jamming detection, over-range and rate of change constraint verification, and spatial consistency check. The device and acquisition link status flags are introduced into the weighted calculation of the credibility score, and the detected anomaly categories are written into the cause label set.

8. The shield tunnel monitoring data stream processing system according to claim 7, characterized in that: For missing data, imputation is performed according to the duration of the missing data: short-term missing data is imputed by interpolation; medium-term missing data is imputed by establishing a regression relationship between neighboring measurement points and measurement points in the same loop, and the regression relationship is established by the same loop samples within the same loop number statistical frame; long-term missing data is written with a data quality degradation marker but no imputed value is generated, and the imputed value and uncertainty interval are written into the quality result object.

9. The shield tunnel monitoring data stream processing system according to claim 8, characterized in that: Establish a three-state mechanism for credibility gating: compare the smaller value of the credibility score and the effective sample rate with a low credibility threshold; if the value is lower than the low credibility threshold, generate a data quality event; if the smaller value is between the low credibility threshold and the high credibility threshold, generate an observation warning and record the number of consecutive confirmations. When the smaller value is higher than the high confidence threshold, it enters the structural risk fusion judgment.

10. The shield tunnel monitoring data stream processing system according to claim 9, characterized in that: Structural risk fusion judgment is based on a combination of multiple evidence rules, including the trend and rate criteria of the main monitoring indicators, the spatial consistency criteria of the same ring and adjacent rings, the operating condition segmentation criteria, and the historical baseline deviation criteria. When the rule combination is satisfied, a structural risk alarm is generated and an evidence summary package is generated. The evidence summary package contains the triggering basis, the key statistical frame index, and the data range identifier.

11. The shield tunnel monitoring data stream processing system according to claim 10, characterized in that: Data quality events, observation warnings, and structural risk alarms are deduplicated and aggregated according to spatiotemporal object keys and spatiotemporal neighborhoods to form measurement point level, ring level, and mileage segment level events, and the status flows according to creation, confirmation, handling, closure, and recurrence; when late data arriving within the allowed late time changes the statistical frame, a new version of the event is generated and the event level and evidence summary package are updated.

12. The shield tunnel monitoring data stream processing system according to claim 11, characterized in that: Establish a twin object state library, using spatiotemporal object keys and component identifiers as indexes to maintain component-level, ring-level, and mileage segment-level states, and manage the object mapping version number and record mapping changes as events; Only observation and early warning, structural risk alarm and ring number statistics frames are written to the structural twin status, and data quality events are written to the sensor health status and link health status.

13. The shield tunnel monitoring data stream processing system according to claim 12, characterized in that: The twin object state library is written using incremental update packages, which include the effective time, the set of changed fields, the source event identifier, and the state version number. When a new version of an event is generated, the twin state is synchronously corrected in a version overwrite manner, and the twin state at any time is reconstructed by replaying the corrected event time. The continuously occurring data quality events are transformed into maintenance tasks such as sensor calibration, line troubleshooting, encrypted inspection and retesting suggestions, and are recorded in association with the spatiotemporal object key.

Images