Gas explosion building risk assessment method and system based on digital twinning

Abstract

The invention discloses a gas explosion building risk assessment method and system based on digital twinning, and particularly relates to the technical field of building risk assessment, and the method comprises the steps: firstly constructing a unified time reference, a quality label and a semantic aperture baseline, and generating traceable features in a rolling manner according to partitions; establishing a ventilation and diffusion twinborn body on the basis, carrying out observation assimilation, and extracting two normalized indexes: ignition readiness and hazard focusing degree; a continuous risk score is calculated according to the data, and is mapped into a discrete level through a multi-level uplink threshold, downlink hysteresis and minimum residence time, and jitter of a critical threshold is inhibited; an atomic instruction is synthesized by combining linkage and a constraint table, issuing and receipt checking are completed, and consistency acceptance and small-step correction are conducted on field measurement and twinborn prediction through a fixed verification window; the parameters, the calibers and the timestamps of the whole process are written into an immutable log, and evaluation, linkage and auditing closed loops are achieved.

Description

Technical Field

[0001] This invention relates to the field of building risk assessment technology, and more specifically, to a method and system for assessing the risk of gas explosions in buildings based on digital twins. Background Technology

[0002] Gas is widely used in urban buildings, including restaurants, commercial complexes, and underground spaces. Once a gas leak occurs, it is highly susceptible to localized accumulation and secondary deflagration due to factors such as ventilation conditions, spatial connectivity, and ignition sources. Existing risk assessments largely rely on static rules or offline simulations: heterogeneous data sources, asynchronous timing, and inconsistent semantic definitions make it difficult to align and trace assessment inputs; the lack of clear coupling between the sufficiency of combustible mixtures and ignition conditions, and between hazardous energy accumulation and personnel exposure, results in a lack of physical direction for risk quantification; classification often uses hard thresholds without hysteresis or dwell time design, leading to frequent jumps in the yellow / orange critical range and unstable linkage control; there is a lack of unified execution tables with mutual exclusion and dependency constraints across subsystems (ventilation, shut-off, sprinkler, explosion venting), and a lack of auditable correspondence between post-treatment data and scenario evidence and model parameters.

[0003] The aforementioned issues result in insufficient real-time performance, robustness, and traceability of online assessments, making it difficult to meet the safety management needs of environments with high population density and tightly coupled equipment. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for risk assessment of gas explosion buildings based on digital twins to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for risk assessment of gas explosion buildings based on digital twins includes the following steps: Establish a unified time benchmark, quality label, and semantic baseline, and generate traceable feature sequences with timestamps and spatial indexes in a rolling manner according to the partition; construct a digital twin of ventilation and diffusion based on the traceable feature sequences, perform observation assimilation, and extract two normalized indicators: ignition readiness and hazard focus. Ignition readiness is obtained by multiplying concentration proximity by ignition source activity. Concentration proximity is the normalized value of the volume fraction of combustible gas in a zone relative to the lower explosive limit. Ignition source activity is obtained by taking the maximum value after normalizing evidence of open flame or hot work and evidence of electrical disconnection and arcing. Hazard focus is the product of equivalent combustible equivalent and normalized occupancy density. Exposure weights are introduced for zones such as evacuation routes, shafts, and control rooms to improve impact. Using ignition readiness and hazard focus as the only inputs, a continuous risk score is calculated. The continuous risk score is then mapped to a discrete risk level through multiple uplink thresholds, downlink hysteresis bands, and minimum dwell time. The minimum dwell time is determined by the preset base dwell time, the continuous risk score, and the real-time occupancy density.

[0006] In a preferred embodiment, based on the discrete risk level, atomic instructions are synthesized under the constraints of the linkage and constraint table and the mutual exclusion matrix to perform linkage control, complete instruction issuance and acknowledgment verification; within a fixed verification window, the effect is accepted based on the consistency between digital twin prediction and on-site measurement, and bounded small-step correction is performed on the twin parameters.

[0007] In a preferred embodiment, the ignition readiness and hazard focus are smoothed according to the assessment step size to suppress instantaneous spikes, and the smoothed indicators are used for the calculation of continuous risk scores and subsequent classification.

[0008] In a preferred embodiment, when key observations are missing, noise is excessive, or semantic caliber drift occurs, the system enters a conservative classification channel. The uplink threshold is not changed, but the reset criteria and minimum dwell time are temporarily tightened only in the downgrade direction. After the uncertainty is resolved, the normal parameters are automatically restored.

[0009] In a preferred embodiment, the discrete risk classification sets up multi-level uplink thresholds of yellow, orange, and red, and corresponding downlink hysteresis bands, with the equal sign belonging to the uplink direction; when the continuous risk score is in the critical neighborhood of yellow and orange and the reset criterion is not met, the high level is maintained and the level is not downgraded.

[0010] In a preferred embodiment, the minimum dwell time is determined by a preset base dwell time, a continuous risk score, and a normalized value of the real-time occupancy density, and increases with the increase of the continuous risk score and the real-time occupancy density.

[0011] In a preferred embodiment, the linkage and constraint table registers the trigger level, preconditions, priority, mutual exclusion and dependency relationships, and secondary evidence threshold for each linkage action; in mutually exclusive conflict scenarios, the priority of security benefits is used for adjudication, and conditional actions are not triggered when secondary evidence is insufficient.

[0012] In a preferred embodiment, the atomic instruction includes the target device tag number, target status or set value, planned start time stamp, idempotent token, sequence number, acknowledgment timeout and maximum number of retries, and supports automatic switching of alternative paths in case of execution failure.

[0013] In a preferred embodiment, the effect is accepted based on the consistency between digital twin prediction and field measurement within a fixed-length verification window. If the standard is not met, the strength is extended or increased without changing the linkage and constraint table entries and the hierarchical threshold. Bounded small-step correction is only performed on parameters such as equivalent leakage strength, removal coefficient and connectivity gain. At the same time, the parameter name, old value and new value, amplitude and effective time are recorded in a versioned manner.

[0014] In a preferred embodiment, the following modules are included: The data access and semantic baseline module is used to unify the time base, quality labels and calibers, and to output feature sequences with version binding in a rolling manner; The twin modeling and parameter estimation module is used to construct ventilation and diffusion twins and assimilate observations, outputting ignition readiness and hazard focus and their metadata; The risk assessment and classification module calculates continuous risk scores based solely on ignition readiness and hazard focus, and outputs discrete risk levels according to threshold, hysteresis, and dwell time. The linkage control and closed-loop verification module synthesizes atomic instructions based on linkage and constraint tables and mutual exclusion matrices, issues and verifies receipts, and performs consistency acceptance and small-step correction within the verification window.

[0015] The technical effects and advantages of this invention are as follows: This invention operates on a unified time base and semantic caliber baseline, standardizing and versioning multi-source data according to "partitions and time steps" to form a traceable chain of evidence. Risk assessment relies solely on two physically interpretable normalized indicators: ignition readiness and hazard focus, directly corresponding to the critical thresholds of whether ignition is possible and whether the hazard is concentrated. The grading mechanism employs an upward-regressing threshold, downward hysteresis, and minimum dwell time combination, eliminating jumps in the critical range. Dwell time is linked to risk score and occupancy density, automatically extending when personnel are dense or risk increases, conforming to the time lag pattern of on-site evacuation and organizational response.

[0016] On the execution side, the system maintains unchanged uplink sensitivity for uncertainties such as missing observations, excessive noise, or caliber drift, only tightening downgrade reset and dwell parameters, adhering to the principle of "not acting rashly due to insufficient information" to stabilize the classification level. Inter-subsystem (ventilation, cutoff, sprinkler, explosion venting) linkages are driven by constraint tables and priority / mutual exclusion relationships, with conflicts adjudicated based on safety benefits. Execution processes retain receipts and immutable logs. Twin predictions and on-site measurements are continuously aligned within the verification window, and bounded small-step corrections are performed on a few parameters such as equivalent leakage and connectivity gain when necessary, without affecting the classification thresholds. Using partitions as unified calculation and control units, parameters and thresholds can be versioned and migrated, facilitating rapid deployment in scenarios such as restaurants, commercial complexes, and underground spaces. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart illustrating the gas explosion building risk assessment method based on digital twins of the present invention. Figure 2 This is a schematic diagram of the structure of the gas explosion building risk assessment method based on digital twins of the present invention. Detailed Implementation

[0018] 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.

[0019] Example 1: This invention provides a method for risk assessment of gas explosion buildings based on digital twins, such as... Figure 1 As shown, it includes the following steps: Step S1: Data Acquisition, Asset Access and Preprocessing This step establishes a unified data and semantic baseline required for subsequent assessments, ensuring that inputs entering the digital twin and risk grading stages are aligned, traceable, and reproducible in terms of time, space, and scope. The processing sequence is as follows: on-site access and location, time consistency and quality governance, semantic and scope baseline establishment, indicator stabilization, feature generation and degradation management, and output and audit archiving. Each step is interconnected and indispensable.

[0020] The on-site process first involved the integration and spatial positioning of multi-source data. The data included combustible gas concentration, ventilation and differential pressure, fan and valve tag numbers, gas emergency shut-off valve status, sprinkler and explosion relief device tag numbers, ignition source-related quantities (electrical load, hot work permit, open flame detection), number of occupants and population density, door and window openings, and boundary conditions such as outdoor wind, temperature, and humidity. To ensure spatial identifiability, the system established a two-way mapping between unique device numbers and zones. A zone is the smallest computational unit for ventilation-diffusion calculations and control, requiring that gas and aerodynamic conditions within this unit be approximated by representative parameters and correspond to sensors or actuators. Zones are typically aligned with rooms; for large spaces, shafts, or normally open interconnected areas, they can be subdivided or merged based on airflow connectivity and control boundaries. All data was accompanied by source identification and measurement point calibration information (range, resolution, accuracy, last calibration time) before entering the message bus for subsequent traceability.

[0021] Time consistency is achieved using the master clock as the reference. The system employs NTP or GPS time synchronization to resample data from different sampling periods to a unified time grid. Linear interpolation is used for continuous quantities with nearest-neighbor timestamps, while zero-order hold is used for step or hold-type quantities, with an "interpolation" marker retained at the beginning of the values. Sudden high-frequency segments are first de-jittered before downsampling according to the target period to prevent short-term spikes from affecting trend judgment. Thus, any statistical or derived calculation is performed based on the same time base, avoiding systematic errors caused by sampling asynchrony.

[0022] Quality governance is implemented after data entry and resampling. Each data point is assigned mutually exclusive quality labels according to rules: OK (valid), MISSING (missing data), STUCK (abnormal constant value), and NOISY (excessive noise or frequent jumps). Short-term missing data (not exceeding twice the nominal period) is compensated using forward limiting but retains the "compensated data" label; long-term missing data remains in the missing state without fitting, allowing downstream systems to identify information gaps and adopt conservative strategies. Outlier identification is completed within a sliding time window, using the median and median absolute deviation (MAD) to construct robust thresholds; quantities that should have natural fluctuations but whose variance continuously approaches zero are marked as STUCK. Samples marked NOISY or STUCK are not included in parameter statistics, but their original values ​​and labels are simultaneously stored in the database for audit review.

[0023] Semantic and caliber baselines were implemented after time and quality were guaranteed. Key indicators (such as LEL, equivalent ventilation rate, ignition source activity, and occupancy density) were uniformly defined in terms of units, dimensions, value ranges, and calculation formulas, and were solidified as the "golden standard" with version numbers. This application adopts the lower explosive limit (LEL) given by standards or catalogues, and the volume fraction of combustible gas c corresponding to... Concentration proximity Used for normalized comparisons. The equivalent ventilation rate, under steady-state approximation, can be expressed by the following formula: calculate, For air intake, This represents the air output; it is used to reflect relative ventilation capacity and does not replace the mass conservation equation.

[0024] To avoid semantic discrepancies between fields with the same name in different systems, external interfaces generate access fingerprints, recording field structure, unit, service level, error code, retry and idempotency strategies, version number, release date, and upstream and downstream dependencies. Any changes to definitions or interfaces are permanently saved as semantic snapshots with an effective date, facilitating cross-version comparison and replay.

[0025] To suppress false judgments triggered by noise and minute changes in aperture within the critical range, stabilization control is implemented upfront at the indicator layer. Anchor thresholds, upper and lower hysteresis bands, and minimum dwell times are set for key indicators involved in the judgment. Short-term round-trip fluctuations of indicators within the hysteresis band do not change the established judgment result; state transition is only allowed after crossing the reset boundary and continuously meeting the dwell time requirement. Considering the sensitivity to the direction of deterioration in engineering, a smaller bandwidth is used for the uplink threshold and a larger bandwidth for the downlink threshold, ensuring agile uplink and robust downlink. This reduces the back-and-forth jumps at the yellow / orange boundary and maintains consistency with the upper-level state machine.

[0026] With a unified time base and aperture, the system generates features in a rolling fashion at fixed steps (e.g., 1–5 s) for direct use by twins and hierarchies. Each zone computes... Based on the duration of exceeding the limit, the rate of increase or decrease in concentration is extracted and estimated according to the zone volume and airflow / pressure difference. Electrical load, open flame detection, and hot work permits are weighted and normalized to determine ignition source activity (range 0-1), and occupancy density and its temporal gradient are obtained based on counting gates or acquired data. All derived quantities are bound to the source measurement point, caliber version, and access fingerprint to form a feature sequence with timestamps and spatial indexes, which can be directly referenced by S2.

[0027] When key measurement points are consistently missing, noise is excessive, or semantic comparison judgments exhibit high-risk drift, this step enters degrade mode and explicitly propagates boundary conditions. The probability-side input uses a mildly upward-adjusted limiting, and the output takes... The severity of consequences is assessed by raising the lower limit of the input and adjusting the output. .For example The range is 0.05–0.15. The range is 0.2–0.4. An uncertainty flag is set synchronously to prompt downstream users to tighten thresholds or extend dwell time in the degradation direction. Using a limiting rather than replacement method preserves the directionality of information and avoids equating unknown risks with no risk. The triggering, maintenance, and deactivation of degradation are recorded along with the caliber and fingerprint version.

[0028] The output of this step is archived as a chain of evidence. For each time step and each partition, the system outputs cleaned and aligned data frames, feature vectors bound to the caliber and fingerprint version, quality and uncertainty status, and the immediate status of stabilization control (whether it is in the hysteresis band, resident time progress), and writes them to an immutable log. Any subsequent changes in risk scores and levels can be traced back to the specific original quantity, derived caliber, and version, thereby ensuring that the parameter identification, classification judgment, and linkage execution of subsequent S2–S4 operate on a unified, clear, and auditable baseline.

[0029] Step S2: Digital Twin and Parameter Estimation This step, building upon the unified time base, quality label, caliber version, and characteristic sequence already completed in S1, constructs an online-running ventilation and diffusion twin. It performs physically consistent calculations and observational assimilation of the combustible gas volume fraction in each zone, generating two normalized outputs as the sole input for subsequent grading: ignition readiness. With focus on harm To facilitate project implementation and ensure reproducibility in audits, this step clarifies the model equations, the meaning and dimensions of the parameters, specifies smoothing and conservative treatments, and solidifies all definitions and versions with timestamps.

[0030] Twins are defined by the volume fraction of combustible gas in partition z. For state variables, their evolution is described using a first-order approximation based on mass conservation: ;in, For partition volume; The volumetric flow rate of adjacent zone j into zone z is determined by the fan operating conditions, valve opening, and pressure difference. The equivalent leakage source strength within the zone can be equal to the leakage volume flow rate when the leaking gas is approximately a single combustible component. The overall dilution / adsorption coefficient is used to characterize the equivalent removal effect excluding ventilation. Numerical implementation is discretized according to the evaluation step size, aligned with the time grid of S1, and uses post-cleaning concentration, airflow, and pressure difference as observations for assimilation. The connectivity structure is initially determined by equipment status and door / window opening, and then updated in small steps within this feasible region. , and To avoid noise being misjudged as structural changes, the parameter update step size is linked to the observation weights and quality labels; when an observation is marked as NOISY or STUCK, its weight is reduced and the step size is limited; when it is MISSING, the relevant channels are frozen and the missing measurement interval is recorded to maintain the most recent reliable state.

[0031] After the twin and observations are consistent, ignition readiness is defined. As a combined measure of the satisfaction of combustible mixture and the satisfaction of ignition conditions. Based on the lower explosive limit (LEL) of the corresponding gas, the instantaneous volume fraction of combustible gas in each zone is... Normalization to concentration proximity The activity of ignition sources was classified according to the on-site data. Its value is determined based on evidence such as the proportion of electrical load, open flame detection, and hot work permits. Ignition readiness is expressed as a product of the given relationships: This form ensures that if conditions on either side are insufficient... It is suppressed only when the combustible mixture approaches or exceeds the lower limit and the ignition source is active. Only then does it significantly increase, matching the physical threshold from dissipation to ignition at the scene.

[0032] Ignition source activity index The quantification method is as follows: At each partition z and each assessment time t, ignition source activity is constituted by two most representative types of evidence: one is evidence of the presence of open flames or hot work, and the other is indirect evidence of electrical interruption and arc risk. These two types of evidence are normalized to... and Then, the supremum of both is taken as the exponent output: .

[0033] The normalized definition of evidence of open flame or hot work is as follows: A zone is equipped with a UV / IR flame detector or an equivalent infrared radiation channel, and its instantaneous intensity is denoted as... Take the reference strength from the device catalog or the on-site playback calibration. Obtain the sensor normalization value To cover scenarios where there are no detectors but a hot work permit exists, the hot work permit system outputs a Boolean value within the permit's validity window. The two are integrated into That is, as long as the sensor shows an open flame, or the area is within the validity period of a hot work permit, it is considered as evidence of an open flame or hot work; if there is no detector and no permit, the channel is set to 0. During the installation and adjustment phase, the intensity percentile (such as 90 percentile) of the open flame can be determined and calibrated and versioned based on the detector response threshold or historical playback.

[0034] The normalization of electrical interruption and arc evidence adopts a simple approach of window counting and amplitude limiting. Let the time sliding window length be... ,For example =20s, count the number of events that may cause an electric arc or generate a strong spark within this window: high current interruption, circuit breaker tripping, arc detection device alarm, welding machine or cutting machine power-on, etc. Record the number of normal interruptions as... The number of arc and fault alarms was Given an equivalent counting coefficient (Used to emphasize the weight of arc-type events), and reference count. (Determined by the typical operating conditions of the area).

[0035] definition ;in , Exported from event logs of power monitoring / protection devices and distribution cabinets; To reflect the importance of an arc event, a value of 3–5 is used; It can be based on the normal turnover intensity of this area, at The typical number of interruptions is selected and fixed with each version. If there is also an arc fault indication current or voltage distortion channel on site, segments that continuously exceed the threshold can also be converted into equivalent events and included. The threshold and conversion rules are fixed in the access fingerprint.

[0036] Hazard Focus Determine whether energy and exposure accumulate in the space surrounding personnel and critical components once ignited. Calculate the equivalent combustible gas equivalent on a unified time base. ;in The volume fraction of combustible gas in a zone refers to the proportion of combustible gas in the air of that zone, and is a component of the gas mixture. For partition volume, ∈[0,1] represents the exposure weight, used to enhance the impact of key areas such as evacuation routes, shafts, and control rooms. Different areas have different preset exposure weights. The upper limit is based on experience with similar buildings. unify it as Simultaneously, the occupancy density threshold is given based on the actual site conditions. Real-time occupancy density Unification Real-time occupancy density refers to the actual number of people present per unit of effective area within a certain zone at the current moment, which can be counted by time clocks or cameras.

[0037] The hazard focus intensity is the combined strength of both factors: .

[0038] When the combustible equivalent or the number of people exposed on either side is insufficient It will not be accidentally raised; when both sides are simultaneously elevated It rapidly approaches 1, thus providing a direct quantification of the spatial overlap between dangerous energy and the victim.

[0039] To suppress transient spikes without sacrificing trend response, both parameters are exponentially smoothed at a uniform evaluation step size Δt to obtain nominal values. (t), (t): ;in ∈[0.7,0.9] is determined by the field control lag and fixed to the version. This is the smoothed nominal output. If the S1 label contains critical missing measurements, excessive noise, or aperture drift, it will enter the conservative channel according to the principle of caution and no degradation, and the output will be... ;in For example, an additive conservative increment of ignition readiness. ∈[0.05,0.15], The lower limit of the safety margin for hazard to focus, for example ∈[0.2,0.4]; Both are calibrated in offline playback and output along with the timestamp, uncertainty flag and caliber version. After the uncertainty is resolved, they automatically return to the nominal channel.

[0040] The output of this step is two normalized parameters organized by region and time, along with their metadata, which serve as the sole input for S3 risk score and level determination.

[0041] Step S3: Risk Calculation and Classification This step, on the time axis with a unified evaluation step size Δt, uses only two normalized parameters, namely the ignition readiness, as input after smoothing / conservatism and outputting the S2 parameter. With focus on harm The system first obtains a continuous risk score R(t), and then maps it to a discrete-level state under threshold and hysteresis constraints. (representing green, yellow, orange, and red in that order), and a dwell timer is used to suppress round-trip transitions in critical intervals. To ensure the interpretability and auditability of the project, the function form, parameter meaning, value range, and equality sign attribution are clearly defined in this step, and all parameters are fixed with the version number and timestamp.

[0042] The continuous risk score is calculated using a coupled weighted form that is monotonically increasing with respect to both inputs: ; in ∈[0,1] represents the linear weighting coefficients, for example =0.5, which determines the relative sensitivity of the risk score to ignition readiness and hazard focus; ≥0 represents the coupling enhancement coefficient, for example =0.5, used to increase the score increase when both factors are simultaneously high, thus reflecting the negative cumulative effect. To avoid numerical accumulation error, the calculation result is... The method is limited to [0,1].

[0043] Discrete grading uses multi-level uplink thresholds and downlink hysteresis bands to... This is mapped to a level status. Let the uplink thresholds for yellow, orange, and red be respectively... , , ,satisfy For example =0.2, =0.5, =0.8; Assume a uniform downlink reset bandwidth of ,For example =0.02, and minimum dwell time In any partition, if the current state is G (green), when R(t,z)≥ Immediately rise to Y (yellow); if the current value is Y, when R(t,z)≥ When it rises to O (orange); if the current value is O, when R(t,z)≥ The alert level is raised to R (red). The equal sign indicates a higher level to ensure timely response to any deterioration.

[0044] Minimum stay time It is determined by the continuous risk score and real-time occupancy density, and the specific formula is as follows. ;in To preset the basic dwell time, for example =2; For continuous risk scores, This is the real-time occupancy density normalized value.

[0045] The more people there are, the higher the expected loss from premature downgrading (relaxation). Under the same "signs of decline," the confirmation time should be extended to ensure the risk has indeed stabilized and decreased, avoiding secondary hazards from large-scale exposure. High occupancy will prolong the actual evacuation / on-site shelter completion time; the system should maintain its current control state (e.g., ventilation, cut-off hold) for longer to cover the lag in crowd behavior and organizational response. Crowd activity can cause airflow disturbances and access control opening and closing, leading to short-term fluctuations in gas concentration and airflow measurements; longer dwell times are needed to filter out false declines / accidental low values, ensuring that downgrading is based on stable rather than instantaneous readings.

[0046] Degradation in the opposite direction occurs only under strict reset conditions: when the state is yellow, orange, or red, respectively, it requires... continuously below or The cumulative time reached Only then can the color change from yellow to green, from orange to yellow, or from red to orange; if at any point during this continuous timing period... If the value returns to the corresponding reset threshold, the timer is immediately reset to zero and maintains its original level. To further suppress jitter near the orange threshold, when... Located in the interval Furthermore, when the current state is 0, the system remains orange without degrading until the reset threshold is explicitly met and the dwell time is completed. If necessary, a very small uplink dwell time can be set in the uplink direction, but the default value is 0 to ensure agile response in the deteriorating direction.

[0047] The hierarchical state S(t,z) operates as a finite state machine, and its transitions depend only on the previous state and the current state. Threshold parameters and dwell time results. The system maintains a dwell timer for each channel that needs to be downgraded; when entering any new level, the corresponding downlink timer is reset; when the corresponding downlink reset threshold is met, the timer increments in steps of Δt until it reaches the threshold. The trigger threshold, equality rule, dwell time length, and current state transition conditions mentioned above. , , Snapshots are recorded with timestamps and archived together with the parameter baseline version number to ensure that any level change can be traced back to clear data and rule basis.

[0048] When the uncertainty propagated by S2 in a certain partition is true (e.g., missing key observations, excessive noise, or semantic drift), this step does not change the uplink threshold. , , Only temporarily tightening the reset criteria in the downgrade direction, adhering to the principle of not aggressively downgrading due to insufficient information, the reset bandwidth and dwell time are strictly controlled. and ,in >0, A value >0 represents a minor tightening, with its scope and range fixed within the parameter baseline. Once the uncertainty is resolved, the parameter automatically returns to its normal value, and the relevant effective and recovery times are written to an immutable log.

[0049] To ensure comparability across time periods and versions, all parameters involved in this step are given a unified definition and archived in a versioned manner. The evaluation step size Δt is in seconds; α and β control the shape of the risk function; , , This is the uplink threshold; and Control the robustness of downgrades; , This is a tightening measure for uncertain scenarios. The system outputs at each evaluation step. In addition, the above parameter versions, equal sign attribution rules, and input quantities are also provided. , The output includes the version number and timestamp; this output serves as the sole trigger for S4 linkage and constraint execution, as well as closed-loop self-consistency verification, ensuring that the control strategy and evaluation conclusions remain consistent in time and semantics.

[0050] Step S4: Linkage and constraint execution and closed-loop self-consistency verification This step uses the risk score and discrete level of S3 as the sole triggering basis. Under the constraints of the Linkage and Constraint Table (LCT) and the mutual exclusion matrix, a control strategy is generated. Command issuance and feedback verification are completed, and the effect is accepted and minor adjustments are made by comparing on-site measurements with digital twin predictions. Once safety requirements are met, the system is maintained and reset according to reversible / irreversible properties. To ensure project traceability, all relevant parameters, thresholds, versions, and timestamps are permanently stored in the output.

[0051] Strategy generation is based on a set of actions. Each action is pre-registered in the LCT with its trigger level, preconditions, priority, whether it is subject to a secondary evidence threshold, and its mutual exclusion and dependency relationships with other actions. The trigger level declares at which the action is eligible to be triggered (values ​​are G, Y, O, or R, corresponding to green, yellow, orange, or red); the preconditions are Boolean judgments of the site boundaries (e.g., whether the exhaust duct is unobstructed, whether the fire water supply pressure is up to standard, whether the evacuation route is unobstructed), and the action only enters the candidate list when the judgment is "true"; the priority is represented by positive integers, with smaller values ​​indicating higher priority, and is used to decide whether to allow or exclude mutually exclusive actions at the same time; the secondary evidence threshold only applies to conditional actions, such as sprinklers needing to reach the open flame evidence threshold, and explosion relief needing to reach the potential overpressure evidence threshold. The evidence is represented by a normalized index and compared with the threshold to determine whether triggering is allowed. The mutual exclusion matrix is ​​used to declare the relationship that two actions cannot be concurrent; the dependency relationship is used to declare the order of actions (e.g., cutting off ventilation before increasing it) and the minimum waiting time between two actions. At each evaluation step, the system reads the current level and scenario evidence, filters candidate actions that meet the trigger level and preconditions, and examines them in descending order of priority. If mutually exclusive actions are encountered, only the one with the greater safety benefit and higher priority is retained. For actions with secondary evidence, they are only included in the final combination if the corresponding evidence is not lower than a threshold. The resulting action set is organized into an execution sequence, and each action is assigned a planned start time and hold duration. If dependencies exist, the relative timing is arranged according to the dependency order and minimum waiting time.

[0052] The policy issuance uses atomic instructions as the smallest unit. An atomic instruction includes the target device tag number, target status or setpoint, planned start timestamp, idempotent token and sequence number, confirmation timeout, and maximum number of retries. Alternative paths can also be specified to handle execution failures. Idempotent tokens ensure repeated delivery does not result in repeated execution; sequence numbers ensure execution in a predetermined order; confirmation timeout limits the maximum window for waiting for on-site confirmation; the maximum number of retries defines the upper limit for automatic retries; and alternative paths immediately switch to equivalent actions on the security side if the primary path fails. After the instruction is issued, the system collects confirmations within the confirmation timeout period. Confirmations include at least whether it was successful, response delay, execution time, and final tag number status. If no confirmation is received within the timeout or a failure confirmation is received, the system immediately executes according to the alternative path and reduces the remaining number of retries. When all retries have been exhausted and still unsuccessful, the current or higher risk level is maintained, any automatic degradation or automatic reset actions are prohibited, and a manual intervention request is sent to on-duty personnel, recording the reason, time, and evidence snapshot in the audit log.

[0053] After the action takes effect, the system enters the effect monitoring and twin consistency verification stage to verify whether the treatment has achieved the expected results and quantify the deviation from the model prediction. Acceptance is based on a fixed-length time window, which starts timing from the beginning of the strategy. Within the window, the system calculates the average amount of concentration trajectory deviation (using the mean square deviation within the window or an equivalent stable deviation metric) and compares it with an error threshold. Simultaneously, it checks whether the proximity of the concentration in each zone to the lower explosive limit is lower than the safe proximity threshold. When the deviation is not higher than the threshold and the proximity is consistently lower than the safe proximity threshold at the end of the window, the treatment is considered successful. To measure the speed of success, the system defines a safe success time, which is the shortest time required for all zones included in the assessment to continuously maintain a position below the safe proximity threshold for at least one minimum duration from the start of the strategy. If the success conditions are not met within the predetermined period, the system will determine that the treatment effect is insufficient. Without changing the LCT entries, mutual exclusion relationships, and level thresholds, it will extend the current treatment duration or increase the treatment intensity (e.g., upgrading the exhaust to a higher level) and trigger a small-step correction process for the twin parameters.

[0054] Small-step corrections are applied only to continuously adjustable environmental and diffusion parameters in S2, and the adjustment range is constrained by an upper limit to ensure the stability and interpretability of the control logic. The evidence used for correction is the average residual of "measured concentration minus predicted concentration" for each zone within the window. When the residual is consistently positive and exceeds the residual threshold, it indicates that the model underestimates the pollution source or overestimates ventilation dilution. The system will then appropriately increase the equivalent leakage intensity or decrease the equivalent removal capacity for that zone. When the residual is consistently negative and exceeds the threshold, it indicates that the model overestimates the pollution source or underestimates dilution. The system will then appropriately increase the equivalent ventilation capacity or equivalent removal coefficient, or make a small gain correction to the flow rate. Each correction has a proportional coefficient and a maximum step size limit to control the magnitude of each change. All changes are recorded with parameter name, old value, new value, change magnitude, effective time, and applicable zone. It is also clearly stated that LCT, mutual exclusion matrix, hierarchical threshold, and evidence threshold will not be touched to prevent the pursuit of short-term fitting from compromising strategy consistency.

[0055] Action holding and reset follow differentiated rules. For reversible actions (such as exhaust fan speed and damper opening), after the level is downgraded according to rule S3 and the reset conditions are continuously met to reach the minimum dwell time, the action gradually returns to the lower level or the initial state according to its own holding time. For irreversible actions or actions requiring manual reset (such as gas cut-off and explosion relief triggering), even if the level is downgraded, the action will not automatically reset. It can only be set after three conditions are met simultaneously: First, the safety proximity threshold is stably met within a minimum holding time; second, the concentration trajectory deviation is not higher than the error threshold; and third, the on-duty personnel explicitly agree to reset through a manual confirmation signal. This differentiated treatment avoids premature removal of critical protection before the treatment is closed.

[0056] During operation, uncertainties and abnormal situations also need to be handled. When there are uncertainties in the ground system annotation input (such as missing key observations or caliber drift), this step does not change the triggered uplink action, but adopts more conservative conditions in the degrading and reset directions, such as increasing the minimum dwell time or extending the action hold duration, to reflect the principle that insufficient information does not support aggressive degrading. When manual coverage is detected, the automatic reverse action related to the covered equipment will be suspended, and the system records the operator, the reason for coverage, and the duration. When a network or power supply anomaly occurs, the local logic on site enters a pre-agreed disconnection safety state (such as disconnection hold, ventilation hold), and the central side maintains no degradation and retryes periodically until the link is restored.

[0057] To ensure cross-version comparability and end-to-end auditability, the parameters and units involved in this step are specified in the manual as follows: the trigger level is G, Y, O, or R; the precondition is a Boolean value; the priority is a positive integer; the secondary evidence threshold and evidence strength are normalized to 0-1; the plan start time, confirmation timeout, maximum waiting time, minimum hold time, action hold time, acceptance window length, and compliance deadline are all in seconds; the maximum number of retries is a non-negative integer; the proportional coefficient and maximum step size are expressed in m³ / s, 1 / s, and dimensionless gain for equivalent leakage intensity, equivalent removal coefficient, and connectivity, respectively. The LCT version number, mutual exclusion matrix version number, action and device firmware version, parameter baseline version, and the timestamps and trigger reason snapshots for each policy synthesis, trimming, distribution, receipt, acceptance, and correction are all written to an immutable log and cross-referenced with the caliber and model version of S1 and S2. Through the above constraints and records, any decision can be traced back to the valid data and control baseline at that time; any abnormal rebound, delay in handling, or model deviation can be located to specific parameters and links and corrected within boundaries.

[0058] Example 2 The present invention discloses a gas explosion building risk assessment system based on digital twins, which is based on the method in Example 1, specifically as follows: Figure 2 The following modules are shown: The data access and semantic baseline module, under the premise of unified time base and caliber, generates feature sequences that can be directly used for digital twins and risk classification with a fixed evaluation step size. The system calculates the duration of exceeding limits and the concentration change rate for each zone, and combines this with zone volume and airflow / pressure difference estimation to form normalized quantities such as traceable ignition source activity and occupancy density. All derived quantities are bound to the source measurement point, caliber version, and access fingerprint, forming a feature output with timestamps and spatial indexes, serving as the sole data entry point for subsequent modules. When critical observations are missing, noise is excessive, or semantic comparison shows high-risk drift, an explicit degradation mode is entered: the threshold is gently increased on the probability side, the lower limit is raised on the severity side, and an uncertainty flag is set to prompt downstream users to tighten thresholds or extend residence time in the degradation direction. The triggering, maintenance, and deactivation of degradation are recorded along with the caliber and access fingerprint version to ensure that the directionality of information is not obscured.

[0059] This module's output is archived in the form of a chain of evidence: for each time step and each partition, it writes cleaned and aligned data frames, feature vectors bound to the caliber and fingerprint version, quality and uncertainty status, and the immediate status of stabilization control (whether it is in the hysteresis band, resident time progress), and solidifies them into an immutable log to ensure that subsequent parameter identification, hierarchical judgment and linkage execution run on a unified, clear and auditable baseline.

[0060] The twin modeling and parameter estimation module is used to standardize the dimensions, value ranges, and calculation formulas of key indicators (such as LEL, equivalent ventilation rate, ignition source activity, and occupancy density) after ensuring time and quality, and solidify them with version numbers as the golden standard. To avoid the meaning discrepancies of fields with the same name in different systems, external interfaces form access fingerprints, recording field structure, units, service levels, error codes, retry and idempotency strategies, and upstream and downstream dependencies. Any changes to the standard or interface are solidified as semantic snapshots and marked with the effective time to support cross-version comparison and playback. To suppress misjudgments caused by noise and minor changes in standard in the critical range, stabilization control is implemented at the indicator layer in advance: anchor thresholds, upper and lower hysteresis bands, and minimum dwell time are set for key indicators participating in the judgment, so that short-term round-trip fluctuations of indicators within the hysteresis band do not change the judgment result, and state transition is allowed only after crossing the reset boundary and continuously meeting the dwell time. Upward actions can use a smaller bandwidth to maintain agility, and downward actions use a larger bandwidth to improve robustness, thereby reducing round-trip jumps at the yellow / orange boundary and maintaining consistency with the upper-level state machine. Based on this, this module outputs two normalized inputs, along with their caliber and version information, under a unified evaluation step size, for the risk assessment module to directly reference.

[0061] The risk assessment and grading module generates continuous risk scores based solely on the two normalized inputs mentioned above, and maps these scores to discrete graded states S(t)∈{G,Y,O,R} under threshold and hysteresis constraints. To ensure comparability across time periods and versions, all parameters are standardized and versioned: the assessment step size is in seconds, and α and β control the shape of the risk function. , , The uplink threshold, reset bandwidth, and minimum dwell time are used to control degradation robustness. The system outputs at each evaluation step with parameter version, input caliber version, timestamp, and equal sign attribution rule, which serve as the sole triggering basis for subsequent linkage and constraint execution.

[0062] When a partition is in an uncertain state, the uplink threshold is not changed, and the reset criteria are only temporarily tightened in the downgrade direction. The reset bandwidth and dwell time are tightened in accordance with the principle of not rushing downgrade due to insufficient information. After the uncertainty is resolved, the parameters automatically return to normal values, and all effective and recovery times are written to the immutable log.

[0063] The linkage control and closed-loop verification module uses risk score and discrete level as the sole triggering basis. Under the constraints of the linkage and constraint table (LCT) and its mutual exclusion matrix, it generates control strategies, completes atomic command issuance and acknowledgment verification, and implements effect acceptance and small-step correction through comparison of on-site measurements and digital twin predictions. After meeting safety requirements, it maintains / resets actions based on their reversible / irreversible attributes, while simultaneously fixing parameters, thresholds, versions, and timestamps to ensure project traceability. The atomic command, as the smallest execution unit, includes the target device tag number, target status or setpoint, planned start timescale, idempotent token and sequence number, confirmation timeout, and maximum number of retries. It can also specify alternative paths to handle execution failures. Acknowledgments are collected within the confirmation timeout period. If a timeout or failure occurs, the alternative path is immediately switched and the remaining retries are reduced until all retries are exhausted. If unsuccessful, the current or higher risk level is maintained, automatic downgrading / reset is prohibited, and manual intervention is requested. The cause, timescale, and evidence snapshot are recorded.

[0064] After the action takes effect, the process enters the effect monitoring and twin consistency verification stage: the concentration trajectory deviation is calculated with a fixed verification window and compared with the error threshold, while the safe proximity of each zone to the lower explosive limit is checked; when the deviation is not higher than the threshold and the proximity is consistently lower than the threshold at the end of the window, it is considered to meet the standard; if the conditions are not met within the compliance period, the current treatment duration is extended or the intensity is increased without changing the LCT entries, mutual exclusion relationships and level thresholds, and small-step corrections to twin parameters are triggered, limited to fine-tuning within the boundaries of equivalent leakage, equivalent removal and connectivity gain. In response to uncertain and abnormal situations, this module does not relax the triggered upward actions and adopts more conservative conditions in the degrading and reset directions (such as increasing minimum dwell time or extending action hold). When manual coverage is detected, the relevant automatic reverse actions are suspended and the operator and reason are recorded. When the network or power supply is abnormal, the local logic on site enters a preset disconnection safety state (such as disconnection hold, ventilation hold), and the central side maintains no degradation and retryes periodically until the link is restored. All process parameters and versions, LCT / mutual exclusion matrix versions, action and device firmware versions, and timestamps and trigger reason snapshots of policy synthesis, trimming, distribution, receipt, acceptance, and correction are uniformly written into the immutable log and cross-referenced with the aforementioned data and model baseline.

[0065] 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.

[0066] 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.

[0067] 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.

[0068] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0069] 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 gas explosion building risk assessment method based on digital twinning, characterized in that, Comprising the following steps: Establish a unified time reference, quality label and semantic caliber baseline, and generate a traceable feature sequence with timestamp and spatial index according to the partition; based on the traceable feature sequence, a digital twin of ventilation and diffusion is constructed, observation assimilation is performed, and two normalized indexes are extracted: ignition readiness and hazard focus; Ignition readiness is obtained by multiplying concentration proximity and ignition source activity, concentration proximity is the normalized value of the volume fraction of flammable gas in the partition relative to the lower explosive limit, and ignition source activity is obtained by taking the maximum value of the normalized value of open flame or hot work evidence and electrical disconnect and arc evidence; hazard focus is the product of equivalent flammable equivalent and occupancy density normalized value, and exposure weight is introduced to improve the influence degree in partitions such as evacuation passage, shaft and control room; Taking ignition readiness and hazard focus as the only input, the continuous risk score is calculated, and the continuous risk score is mapped to discrete risk level through multi-level uplink threshold, downlink lag zone and minimum residence time, which is determined by the preset basic residence time, continuous risk score and real-time occupancy density.

2. The digital-twin-based gas explosion building risk assessment method according to claim 1, characterized in that: Based on the trigger basis of discrete risk level, atomic instructions are synthesized under the constraints of linkage and constraint table and mutual exclusion matrix, linkage control is performed, instruction issuing and return verification are completed; effect acceptance is carried out based on the consistency of digital twin prediction and field measurement in the fixed verification window, and bounded small step correction is implemented on the twin parameters.

3. The digital-twin-based gas explosion building risk assessment method according to claim 1, characterized in that: The ignition readiness and hazard focus are smoothed according to the evaluation step length to suppress transient spikes, and the smoothed indexes are used for continuous risk score calculation and subsequent grading.

4. The digital-twin-based gas explosion building risk assessment method according to claim 1, characterized in that: When key observations are missing, noise is too large or semantic caliber drift occurs, enter the conservative judgment level channel, do not change the uplink threshold, only temporarily tighten the reset criterion and minimum residence time in the downgrade direction, and automatically restore the normal parameters when the uncertainty is removed.

5. The digital-twin-based gas explosion building risk assessment method according to claim 1, characterized in that: Discrete risk grading sets yellow, orange, red multi-level uplink threshold and corresponding downlink lag zone, and the equal sign belongs to the uplink direction; when the continuous risk score is located in the critical neighborhood of yellow and orange, and the conditions do not meet the reset criterion, the high level is maintained without downgrading.

6. The digital-twin-based gas explosion building risk assessment method according to claim 1, characterized in that: The minimum residence time is determined by the preset basic residence time, the normalized value of continuous risk score and real-time occupancy density, and is increased with the increase of continuous risk score and real-time occupancy density.

7. The digital-twin-based gas explosion building risk assessment method according to claim 2, characterized in that: The linkage and constraint table registers the trigger level, prerequisite, priority, mutual exclusion and dependency relationship, and secondary evidence threshold for each linkage action; in the mutual exclusion conflict scenario, the safety benefit priority is prioritized, and the conditional action is not triggered when the secondary evidence is insufficient.

8. The digital-twin-based gas explosion building risk assessment method according to claim 2, characterized in that: Atomic instructions include target device bit number, target state or set value, planned start time, idempotent token, sequence number, confirmation timeout and maximum retry number, and support alternative path automatic switching when execution fails.

9. The digital-twin-based gas explosion building risk assessment method according to claim 2, characterized in that: Effect acceptance is based on the consistency between digital twin prediction and field measurement within a fixed-length validation window. When not up to standard, the strength is extended or improved without changing the linkage and constraint table entries and grading thresholds. Only equivalent leakage intensity, removal coefficient, and connectivity gain parameters are implemented with bounded small-step correction. Parameter names, old and new values, amplitudes, and effective times are versioned and recorded.

10. A digital-twin-based gas explosion building risk assessment system for implementing the method of any one of claims 1-9, characterized by, The following modules are included: Data access and semantic baseline module for unified time base, quality label, and caliber, and rolling output of version-bound feature sequence; Twin modeling and parameter estimation module for constructing ventilation and diffusion twins and assimilating observations, outputting ignition readiness and hazard focus degree and their metadata; Risk assessment and grading module for calculating continuous risk scores based on ignition readiness and hazard focus degree and outputting discrete risk levels according to thresholds, hysteresis, and residence time; Linkage control and closed-loop verification module based on atomic instructions synthesized from linkage and constraint tables and mutual exclusion matrices, issuance and receipt verification, and implementation of consistency acceptance and small-step correction within the validation window.

Images