Surrounding rock mechanics parameter real-time inversion method, system, electronic device and computer program product

Abstract

This invention discloses a real-time inversion method, system, electronic equipment, and computer program product for surrounding rock mechanical parameters, relating to the field of numerical analysis technology in geotechnical engineering. It acquires the surrounding rock monitoring response, engineering state, and reference parameter state, generates candidate parameter states, and determines the data consistency residuals and independent physical consistency residuals before and after the update. When data consistency improves while independent physical consistency deteriorates and exceeds the allowable boundary, a counterfactual regression state is constructed for the updated parameters, and forward mechanical calculations are re-executed to determine the set of conflicting parameters. The reference values ​​of conflicting parameters are restored, while the updated values ​​of non-conflicting parameters are retained, forming a mixed parameter state. After verification, this mixed parameter state is fed back to the next monitoring window as a valid parameter state. This invention also discloses an inversion system, electronic equipment, and computer program product for updating surrounding rock mechanical parameters in underground engineering, avoiding the propagation of non-physical parameter states and the overall loss of valid updates.

Description

Technical Field

[0001] This invention relates to the field of numerical analysis technology in geotechnical engineering, specifically to methods, systems, electronic equipment, and computer program products for real-time inversion of surrounding rock mechanical parameters. Background Technology

[0002] During the construction of tunnels, underground caverns, and roadways, the load conditions, boundary constraints, and deformation states of the surrounding rock will change with the excavation stage. In engineering, the field monitoring responses such as displacement, strain, or stress are usually input into the numerical model. By iteratively adjusting the constitutive parameters, the model's predicted response gradually approaches the monitoring response, and the surrounding rock mechanical parameters used in the construction stage are updated accordingly.

[0003] Chinese patent document CN111666671A discloses a real-time inversion method for creep parameters of surrounding rock mass. First, a constitutive model and the creep parameters to be inverted are determined based on geological surveys and rock mechanics tests, constructing a numerical calculation model of the surrounding rock mass. Then, multiple sets of parameters within their value ranges are selected for numerical calculation to obtain a dataset of creep parameters and mechanical responses. A recurrent neural network surrogate model is then trained using this dataset. An objective function is established between the predicted response and the measured response based on the field-measured mechanical response, and this function is subjected to stochastic global optimization to determine the current creep parameters. When a new measured mechanical response arrives, the above parameter optimization is repeated to update the inversion results. Here, model inversion refers to using a surrogate model to reduce redundant numerical calculations, and using the difference between the predicted and measured responses as the basis for searching parameters.

[0004] However, if the number of monitoring channels is small, the effects of different parameters on the observed response are similar, or the numerical model has approximation errors, multiple surrounding rock mechanics parameters may undergo compensatory changes. That is, different combinations of parameter values ​​will reduce the monitoring fitting error. In this case, a small data residual does not necessarily mean that the obtained results are continuous. For example, other responses not involved in the fitting, constitutive evolution, boundary states, and states between adjacent construction stages. If the entire set of updates is accepted only based on the fitting target, parameter states that deviate from physical constraints may be passed on to later windows. If all parameters are uniformly restored after an anomaly is detected, and update results that did not cause physical deviations are also revoked, it will increase redundant calculations and weaken the effectiveness of continuous inversion.

[0005] Therefore, existing technologies still need to address the technical problem of how to reliably handle local parameter update distortion and avoid the overall loss of effective updates when there is a discrepancy between monitoring the improvement in fit and the evaluation of physical state. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method, system, electronic equipment, and computer program product for real-time inversion of surrounding rock mechanical parameters. First, it generates a current candidate parameter state based on the current monitoring window, current engineering state, and current reference parameter state. Then, it generates data consistency residuals and independent physical consistency residuals before and after parameter updates. When data consistency improves while independent physical consistency deteriorates and exceeds the corresponding allowable boundary, a counterfactual backoff state is constructed for one or more tested parameter units. While maintaining the current engineering state and other updated parameters, the forward mechanical calculation is re-executed for each counterfactual backoff state. The set of conflicting parameters is determined based on the changes in the two types of residuals before and after backoff. Subsequently, only the conflicting parameters are restored to their reference values, while the updated values ​​of the non-conflicting parameters are retained, forming a mixed parameter state. Finally, a complete review is performed on the mixed parameter state, and upon successful review, it is written into the current valid parameter state and fed back to the next monitoring window. This solves the technical problems addressed in the background art.

[0008] (II) Technical Solution

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

[0010] The real-time inversion method for surrounding rock mechanical parameters includes: acquiring the monitoring response, engineering state, and reference parameter state; generating candidate parameter states based on the parameter update response set; and determining the data residuals and physical residuals between the two states based on the response set and a physical verification constraint set generated independently of the candidate states.

[0011] When the data residual of the candidate parameter state decreases and the physical residual increases beyond the limit, select the parameter unit from the updated parameters and restore it to the reference value. Keep the candidate values ​​of the remaining updated parameters, construct the counterfactual backoff state and perform the forward mechanical calculation.

[0012] Parameter units whose physical residuals are recovered and whose data residuals are not out of bounds are included in the conflict parameter set. The parameters in this set are restored to the reference values, and the remaining updated values ​​are retained to form a mixed parameter state.

[0013] When a conflicting parameter set is formed, a mixed parameter state is submitted; when no conflicting parameter set is formed, the data residual of the candidate parameter state is not greater than the data residual of the reference parameter state, and none of the physical residual components cross the corresponding physical allowable boundary, a candidate parameter state is submitted; when no conflicting parameter set is formed, the data residual of the candidate parameter state is greater than the data residual of the reference parameter state, or any physical residual component crosses the corresponding physical allowable boundary, the reference parameter state is maintained; if the submitted state passes the review of both types of residuals, it is considered a valid parameter state; otherwise, the reference parameter state is maintained.

[0014] Furthermore, the monitoring response is divided into a parameter update response set and a physical verification constraint set before generating candidate parameter states. The parameter update response set participates in the construction of the parameter update target, while the physical verification constraint set does not participate in the construction of the target. The division result is saved together with the window identifier and the engineering state version, and remains unchanged during the residual comparison of candidate parameter states, counterfactual backoff states, and mixed parameter states.

[0015] Furthermore, the decrease in data residuals relative to the reference parameter state and the increase in each physical residual component are calculated; when the decrease exceeds the data fluctuation threshold, at least one increase exceeds the corresponding physical fluctuation threshold, and the corresponding physical residual component crosses the physical allowable boundary, a counterfactual backoff state is constructed.

[0016] When the data residual of the candidate parameter state is not greater than the data residual of the reference parameter state and none of the physical residual components have crossed the corresponding boundary, the candidate parameter state is submitted; otherwise, the reference parameter state is maintained.

[0017] The real-time inversion system for surrounding rock mechanical parameters includes: a parameter incremental update module, which receives monitoring responses, engineering status, and reference parameter status, and generates candidate parameter status based on the parameter update response set;

[0018] The dual consistency calculation module determines the data residuals and physical residuals between the reference parameter state and the candidate parameter state based on the parameter update response set and the physical verification constraint set that did not participate in the generation of candidate parameter states.

[0019] The consistency conflict triggering module triggers the counterfactual backoff state construction module when the data residual of the candidate parameter state decreases and the physical residual increases and exceeds the physical allowable boundary.

[0020] The counterfactual rollback state construction module selects parameter units from the updated parameters, restores the parameter units to the reference values, and keeps the remaining updated parameters as candidate values ​​to form a counterfactual rollback state.

[0021] The forward mechanics solution module performs forward mechanics calculations on counterfactual backoff states.

[0022] The conflict parameter set determination module includes parameter units whose physical residuals are restored to within the physical allowable boundary and whose data residuals do not exceed the data allowable boundary into the conflict parameter set;

[0023] The selective rollback and verification module restores the parameters in the conflict parameter set to the reference value and retains the remaining updated values ​​to form a mixed parameter state. When a conflict parameter set is formed, the mixed parameter state is set to the commit state. When the trigger condition is not met, the candidate parameter state is set to the commit state. When the trigger condition is met but a conflict parameter set is not formed, the reference parameter state is set to the commit state, and the data residual and physical residual verification is performed on the commit state.

[0024] The parameter status storage and feedback module saves the submission status as a valid parameter status when the review is passed, and retains the reference parameter status when the review fails.

[0025] An electronic device includes a processor, a memory, a network interface, and a communication bus. The processor, memory, and network interface are connected via the communication bus. The network interface receives monitoring responses, engineering status, and mechanical model configurations. The memory stores reference parameter status and a computer program. When the computer program is executed by the processor, the electronic device implements the real-time inversion method for surrounding rock mechanical parameters.

[0026] A computer program product includes a computer program that, when executed by a processor, implements the real-time inversion method for surrounding rock mechanical parameters.

[0027] (III) Beneficial Effects

[0028] This invention provides a real-time inversion method for surrounding rock mechanical parameters, which has the following beneficial effects:

[0029] By determining the parameter update response set and physical verification constraint set before parameter updates and keeping their uses unchanged within the current round, independent physical consistency is prevented from being absorbed by the update objective. This allows for the detection of parameter compensation scenarios where data fitting improves while constitutive states, boundary responses, or state continuity deteriorate. By comparing the data consistency residuals and independent physical consistency residuals between the current reference parameter state and the current candidate parameter state, and by combining data fluctuation boundaries, physical fluctuation boundaries, and physical allowable boundaries for triage, numerical fluctuations, ordinary update failures, and consistency conflicts can be distinguished, thereby reducing false triggers and overall rollback.

[0030] By constructing a counterfactual backoff state for the tested parameter unit, only the reference value of the corresponding parameter is restored, while other updated parameters are kept as candidate values. The forward mechanical calculation is then re-executed under the same engineering conditions, thus isolating the actual influence of the parameter or parameter group on the two types of residuals. This provides a repeatable mechanical basis for determining the conflict parameter set.

[0031] By incrementally searching for parameter groups when single-parameter rollback fails to restore consistency, and by combining state versioning, activity state index switching, maintaining the current valid parameter state in case of anomalies, and cross-window re-judgment, multi-parameter joint compensation, calculation timeout, and state submission anomalies are effectively handled, thus preventing unverified intermediate states from propagating to subsequent monitoring windows. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the logical architecture of the online inversion system for surrounding rock mechanical parameters of the present invention;

[0033] Figure 2 This is the main flowchart of the online inversion method for surrounding rock mechanical parameters of the present invention;

[0034] Figure 3 This is a schematic diagram illustrating the classification of data consistency and independent physical consistency change states in this invention;

[0035] Figure 4 This is a schematic diagram illustrating the construction of the parameter-level counterfactual fallback state of the present invention;

[0036] Figure 5 This is a flowchart illustrating the generation process of the conflict parameter set in this invention.

[0037] Figure 6 This is a flowchart of the selective rollback and mixed parameter state verification of the present invention;

[0038] Figure 7 This is a schematic diagram illustrating the cross-monitoring window parameter status feedback and anomaly degradation of the present invention;

[0039] Figure 8 This is a schematic diagram of the hardware topology of the electronic device of the present invention;

[0040] Figure 9 This is a schematic diagram illustrating the current application classification and independent verification relationship of the monitoring response of this invention;

[0041] Figure 10 This is a flowchart of the search process for the minimum conflict parameter set in this invention;

[0042] Figure 11 This is a schematic diagram illustrating the global anomaly handling and security degradation of this invention. Detailed Implementation

[0043] 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.

[0044] Please see Figures 1-11 This invention provides a method, system, electronic equipment, and computer program product for real-time inversion of surrounding rock mechanical parameters.

[0045] like Figure 1 As shown, the online rock mechanics parameter inversion system 100 establishes data interfaces with external monitoring data sources 10, engineering state data sources 20, mechanical model configuration data sources 30, and initial parameter and parameter boundary data sources 40. Internally, the system 100 includes a monitoring window construction module 110, an observation mapping and response partitioning module 120, a mechanical forward solution module 130, a parameter incremental update module 140, a dual consistency calculation module 150, a consistency conflict triggering module 160, a counterfactual rollback state construction module 170, a conflict parameter set determination module 180, a selective rollback and verification module 190, a parameter state storage and feedback module 200, and an anomaly handling module 210. Each module can be encapsulated as a function component within the same process or as multiple service components scheduled by the same computational entity. Regardless of the encapsulation method, consistency conflict triggering, counterfactual rollback state construction, mechanical re-solution, conflict parameter determination, selective rollback, and mixed state verification are all controlled and completed by the same rock mechanics parameter inversion calculation entity, thus concentrating the method steps on a single entity and forming a continuous data state closed loop.

[0046] Combination Figure 1 and Figure 2 It is understood that the external monitoring data source 10 only provides monitoring records with timestamps, channel identifiers, physical quantity types, monitoring values, units, and quality identifiers; the engineering status data source 20 provides construction stages, load states, boundary condition versions, and model versions; the mechanical model configuration data source 30 provides control equations, constitutive relations, discrete models, observation mappings, and solution configurations; and the initial parameter and parameter boundary data source 40 provides the initial parameter states used in the first monitoring window and the physical upper and lower bounds of each candidate parameter. The above data sources do not require specific monitoring device structures or specific communication protocols; their function is to provide traceable input to the system 100. After completing S101 to S106, the system 100 outputs the current valid parameter states and corresponding state records, without directly driving the support equipment or construction machinery. This limits the technical boundaries of this implementation to the numerical inversion, conflict attribution, and state feedback processes of the surrounding rock mechanical parameters.

[0047] Combination Figures 9 to 11 It can be seen that the current monitoring response is divided into update purpose and independent verification purpose before the parameter update begins; when a single parameter counterfactual rollback is insufficient to restore independent physical consistency, candidate parameter groups are constructed in increments according to the parameter group size; when anomalies occur in the data, solution, parameter or status submission process, the anomaly handling module 210 switches between local repair and security degradation. Figures 9 to 11 The aforementioned methods of isolation, minimum conflict parameter set search, and global anomaly handling are further elaborated upon, with each section utilizing [the relevant technologies / methods]. Figures 1 to 8 The data sources, modules, and state relationships are already defined in the code, and no new entity devices are introduced.

[0048] The current monitoring window is a data range corresponding to a defined engineering state, used for one parameter update and one consistency verification. The current monitoring window can be formed according to a fixed duration, a sliding duration, a construction event, or an engineering state transition event. For monitoring data spanning two construction stages, two load states, or two sets of boundary conditions, the monitoring window construction module 110 splits the data based on the engineering state transition time, ensuring consistency in the forward mechanical calculation conditions within the same window. This is used to avoid mixing monitoring responses from different engineering states into the same parameter attribution process.

[0049] The current project status is a combination of the construction stage identifier, load status, boundary condition status, model version, and observation mapping version corresponding to the current monitoring window. The current project status is provided by the project status data source 20 and written into the current window record by the monitoring window construction module 110. The current project status is not the same as the general business status; it must be able to be converted by the mechanics forward solution module 130 into solution boundary, load input, initial state, or stage status.

[0050] The current reference parameter state is the complete state of surrounding rock mechanics parameters before the start of this round of incremental parameter updates. In the first effective monitoring window, the current reference parameter state originates from the initial parameters and parameter boundary data source 40; in subsequent monitoring windows, the current reference parameter state originates from the current effective parameter state saved after verification in the previous window. The current reference parameter state can be considered both the starting point of incremental parameter updates and the benchmark for restoring the tested parameters during counterfactual rollback. The current candidate parameter state is the parameter state formed by the incremental parameter update module 140 after updating the parameters based on the current reference parameter state and the parameter update response set, but which has not yet been accepted as a valid state by the parameter state storage and feedback module 200. The current candidate parameter state includes candidate values ​​of parameters updated in this round and reference values ​​of parameters not updated in this round. The candidate state requires double consistency calculation and necessary selective rollback; it cannot directly overwrite the current reference parameter state.

[0051] The parameter update response set is a set of responses determined by the observation mapping and response partitioning module 120 from the current monitoring window before the parameter update begins, used to construct the target for this round of parameter updates. The parameter update response set may consist of a portion of monitoring channels, a portion of spatial locations, a portion of response types, or a portion of time slices. Once this set is used for this round of parameter updates, it remains unchanged until the end of this round to avoid changing the fitting target afterward based on the update results.

[0052] The physical verification constraint set is a set of constraints that are not directly used as the target for the current round of parameter updates, but are used to evaluate whether the current candidate parameter state or counterfactual backoff state meets the independent physical requirements. It may include reserved monitoring responses, internal state continuity between adjacent monitoring windows, constitutive state acceptability, boundary responses not involved in the update, and the allowable response range given by the output of an independent reference mechanical model or material testing. The physical verification constraint set must be able to be calculated or read from the current engineering state and parameter state, and maintain data independence from the parameter update response set in terms of data usage.

[0053] Data consistency residuals are the differences between the predicted responses and the corresponding monitored responses in the parameter update response set. These residuals are used to evaluate the explanatory power of a parameter state for the current fitted target. Data consistency residuals can be normalized according to the noise scale of each channel or can retain the component forms of multiple channels. Their allowable boundaries are derived from monitoring repeatability errors, steady-state residual distributions, or calibrated data error ranges, rather than arbitrarily set, sourceless constants.

[0054] Independent physical consistency residuals represent the degree to which a parameter state violates the physical verification constraint set. These residuals can consist of multiple components, such as reserved response residuals, cross-window state continuity residuals, constitutive state residuals, boundary response residuals, or independent model residuals. Independent physical consistency residuals do not solely rely on the numerical convergence residuals of the same mechanics solver; while solver convergence residuals are only used to determine the validity of the current forward solution, independent physical consistency is used to determine whether the parameter state still holds true under physical conditions that do not participate in updating the objective.

[0055] It should be noted that consistency conflict refers to a situation where, compared to the current candidate parameter state, the current reference parameter state causes the data's consistency residuals to decrease, and at least one independent physical consistency residual is large and exceeds the corresponding physical allowable boundary. This type indicates that although the parameter update improves the fitting objective, it will lose at least a portion of the updates that may have disrupted the independent physical verification conditions through parameter compensation.

[0056] The parameter unit under test is a parameter or set of parameters selected from the parameters updated in the current round, which are collectively restored to the reference value in a counterfactual rollback state. Single-parameter test units are used to locate a single conflicting parameter; parameter group test units are used to handle situations where multiple parameters compensate together and a single-parameter rollback cannot restore consistency.

[0057] The counterfactual rollback state is a parameter state formed based on the current candidate parameter state, where only the parameters in the tested parameter unit are restored to their corresponding values ​​in the current reference parameter state, while the updated parameters outside the tested parameter unit remain as current candidate values. By recalculating this state under the same engineering conditions, the changes in the two types of consistency residuals after the update of the tested parameter unit can be evaluated, thereby transforming the abstract parameter influence into repeatable mechanical solution results.

[0058] The conflict parameter set is the set of parameters for which the current update should be revoked, determined based on the resolution results of each counterfactual rollback state. When a tested parameter unit recovers its reference value, and its independent physical consistency returns to the allowable range or achieves an improvement exceeding the numerical fluctuation, while data consistency remains within the allowable range, then the parameters in that tested parameter unit can be included in the conflict parameter set.

[0059] A hybrid parameter state is a complete parameter state formed by restoring the parameters in the conflicting parameter set to their current reference values ​​and retaining the updated parameters outside the conflicting parameter set as current candidate values. A hybrid parameter state is not a simple concatenation; it requires re-performing forward mechanical calculations under the current engineering conditions and undergoing verification through data consistency, independent physical consistency, parameter range checks, and solution validity.

[0060] The current valid parameter status is the parameter status submitted and activated by the parameter status storage and feedback module 200 after complete verification. This status serves as the current reference parameter status for the next monitoring window. Current candidate parameter statuses, counterfactual rollback statuses, and mixed parameter statuses that have not yet completed verification must not replace the current valid parameter status.

[0061] Online inversion refers to the process where, after a new effective monitoring window arrives, the system 100 can re-update parameters, attribute consistency conflicts, and provide status feedback based on the current project status and the previous effective parameter status. Online inversion does not mean that the operation must be completed within a fixed millisecond or second time limit. Whether a specific runtime limit is met depends on the window arrival period, model size, number of parameters, and computing resources.

[0062] To ensure that the minimalist labels in the attached figures correspond uniquely to the terminology in the text, Figure 3 In this context, physical consistency refers to independent physical consistency. "Double excellent" means that both data consistency and independent physical consistency have improved. "Data excellent, physical poor" means that data consistency has improved while independent physical consistency has deteriorated. "Data poor, physical excellent" means that data consistency has deteriorated while independent physical consistency has improved. "Double poor" means that both types of consistency have deteriorated. Figure 6 In this context, data verification and physical verification refer to data consistency verification and independent physical consistency verification, respectively. Figure 9 The update response and verification constraint in the text refer to the parameter update response set and the physical verification constraint set, respectively. Figure 10 In this context, a single-parameter failure indicates that all single-parameter counterfactual rollback states have failed to meet the physical recovery and data preservation conditions. Figure 11 "Maintain the most recently valid" means maintaining the current valid parameter status of the most recently submitted and fully reviewed parameter. The above abbreviations are only used to shorten the text in the accompanying figures and do not change their technical meaning in the main text.

[0063] Security degradation refers to the anomaly handling module 210 freezing the current round of parameter updates, rejecting the current candidate parameter state, maintaining the most recently valid parameter state, recording the cause of the anomaly, and entering the processing state of the next monitoring window when the current anomaly cannot be partially repaired through counterfactual rollback, parameter group rollback, or re-solution. Security degradation does not write unverified candidate states into the active state, nor does it mean permanently stopping online inversion; when the corresponding recovery conditions are met, the system 100 re-enters the normal operation process from the most recently valid parameter state.

[0064] like Figure 1 As shown, this embodiment uses online inversion of surrounding rock mechanical parameters during tunnel construction as an application scenario. The online inversion system 100 for surrounding rock mechanical parameters is deployed in a computing environment capable of accessing monitoring data, engineering status data, and mechanical model configuration. The system 100 can run on a single computer or a set of computing nodes controlled by the same task scheduling logic; its underlying layer must have at least a processor, memory, and data interface to receive monitoring records, load the mechanical model, execute multiple forward solutions, and save the parameter status. The system 100 does not require monitoring devices, communication networks, and construction equipment to have specific mechanical structures; it only requires that external data can be input in the form of verifiable data records.

[0065] like Figure 1 As shown, the external monitoring data source 10 can be a monitoring database, the output interface of a data acquisition platform, or a file data source, which provides monitoring records to the monitoring window construction module 110. Each monitoring record contains at least a timestamp, channel identifier, physical quantity type, value, unit, and data quality identifier, and each monitoring record corresponds to an important program. The monitoring window construction module 110 can consist of a data reading thread, a time alignment program, a window indexing program, and a validity verification program, and writes its output to the working storage area of ​​the system 100. The timestamp, channel identifier, and quality identifier are retained at the input end. Subsequent operations such as observation mapping, window splitting, and anomaly rollback can all be traced back to the original record, preventing the parameter status from being unreproducible due to the unknown source of the record.

[0066] like Figure 1As shown, the engineering state data source 20 provides the monitoring window construction module 110 and the forward mechanical solution module 130 with construction stage identifiers, boundary condition versions, load state versions, and model state identifiers. For example, during a certain monitoring period, due to excavation stage switching, unloading step changes, or boundary constraints updates, a new state record is generated at the corresponding effective time. Subsequently, the monitoring period spanning state switching (i.e., each monitoring period) is split into two different monitoring windows, so that each forward mechanical calculation matches only one set of boundary and load conditions, and the response differences generated before and after parameter updates are caused by parameter changes, rather than being mixed by different engineering states.

[0067] like Figure 1 As shown, the mechanical model configuration data source 30 stores the governing equations, constitutive relations, discrete model, model observation locations, calculated response types, solution tolerances, and model versions. The mechanical forward solver module 130 reads the current engineering state and model version, and then loads the corresponding solver configuration. The mechanical forward solver module 130 can encapsulate finite element solution, finite difference solution, boundary element solution, meshless solution, or validated reduced-order mechanical solver logic; regardless of the solver path used, its common inputs include the current engineering state and the complete surrounding rock parameter state, and its common outputs include predicted responses that can be correlated with monitoring responses and a solver validity indicator.

[0068] like Figure 1 As shown, the initial parameter and parameter boundary data source 40 stores the initial parameter status required for the first monitoring window, the Chinese name, physical meaning, unit, lower bound, upper bound, and source identifier of each parameter. The upper and lower bounds of the parameters can be provided by rock mechanics tests, engineering surveys, design calculations, historical stable parameter status, or confirmed material response ranges. When reading these records using the parameter status storage and feedback module 200, the source identifier and version number are also read out. Subsequent parameter rollback can clearly obtain the reference value to which the parameter is restored. By using parameter boundaries with traceable sources, the path where parameter updates cross the physically feasible domain and are still accepted by the data fitting results can be blocked.

[0069] like Figure 1As shown, the observation mapping and response partitioning module 120 is located between the monitoring window construction module 110 and the forward mechanical solution module 130, and can be composed of a channel mapping table, a spatial interpolation program, a direction conversion program, and a response set partitioning program. Specifically: the channel mapping table matches the monitored channels with nodes, integration points, boundary positions, or output functions in the model; the direction conversion program transforms the response in the model coordinate system into the response in the monitoring coordinate system; and the response set partitioning program, before the start of the current parameter update, partitions the valid responses in the current window into a parameter update response set and a physical verification constraint set. In this way, the partitioning can be completed and the partitioning results fixed before the update begins, preventing the selection of verification data afterward based on candidate parameter results.

[0070] Combination Figure 9 It can be seen that after the monitoring window construction module 110 forms the current monitoring window, the observation mapping and response partitioning module 120 partitions its purpose before the parameter update begins: the parameter update response set is input to the parameter incremental update module 140 along the update branch, the physical verification constraint set is input to the double consistency calculation module 150 along the verification branch, and the mechanical forward solution module 130 provides the predicted response corresponding to the two branches. Figure 9 In this context, "pre-partitioning" and "invariant in this round" mean that the partitioning result is determined before the start of the current round and remains consistent when comparing the current candidate parameter state, counterfactual backoff state, and mixed parameter state, thus avoiding the need to change the verification object after the fact based on the calculation results.

[0071] like Figure 1 As shown, the parameter incremental update module 140 receives the current reference parameter state, the parameter update response set, and the predicted response returned by the mechanics forward solution module 130. The parameter incremental update module 140 can consist of an objective function construction program, a parameter constraint program, an iterative update program, and a step size control program. This module is not limited to using a specific optimization algorithm; its necessary action is to start from the current reference parameter state, change the parameters to be updated in this round within the physical range of the parameters, and generate a complete current candidate parameter state. After each trial update state is formed, the parameter incremental update module 140 calls the mechanics forward solution module 130 to recalculate the predicted response, thereby ensuring that the parameter update is based on the actual model response under the current engineering state.

[0072] like Figure 1As shown, the dual consistency calculation module 150 receives the parameter update response set, the physical verification constraint set, the model output corresponding to the current reference parameter state, and the model output corresponding to the current candidate parameter state, respectively. This module can consist of a data residual calculation program, a physical residual component calculation program, a residual normalization program, and a change direction comparison program. The dual consistency calculation module 150 retains the residual components corresponding to different physical verification constraints, and does not compress all residuals into a single comprehensive score using unsourced weights. This structure enables the consistency conflict triggering module 160 to determine which specific type of independent physical verification has deteriorated, thus providing a clearer physical indication for subsequent conflict parameter attribution.

[0073] like Figure 1 As shown, the consistency conflict triggering module 160 receives and stores the residuals before and after the update and the corresponding fluctuation boundary information output by the dual consistency calculation module 150. Specifically, if the improvement in data consistency exceeds the data fluctuation range, and at least one independent physical consistency deteriorates beyond the allowable range, the consistency conflict triggering module 160 pushes a conflict attribution triggering instruction to the counterfactual rollback state construction module 170. If both types of issues are improved, the process can directly proceed to candidate state review; if data consistency deteriorates, the parameter incremental update module 140 needs to adjust the update step size or reject the current update. Furthermore, this splitting process avoids confusing ordinary non-convergence, measurement fluctuations, and parameter compensation conflicts into the same anomaly. Figure 1 As shown, the counterfactual fallback state construction module 170 reads the current reference parameter state, the current candidate parameter state, and the list of parameters updated in this round, and constructs a counterfactual fallback state for one or more tested parameter units one by one. This module can consist of a parameter index management program, a state copying program, a reference value replacement program, and a task queue program. After each counterfactual fallback state is formed, it is sent to the mechanics forward solution module 130 as an independent solution task. By restoring only the tested parameter unit while keeping other updated parameters as candidate values, the impact of the current round update of the tested parameter unit can be isolated from the updates of other parameters.

[0074] like Figure 1 As shown, the conflict parameter set determination module 180 receives the data residuals and independent physical residuals corresponding to each counterfactual rollback state. It determines whether a tested parameter unit needs to enter the conflict parameter set according to predefined deterministic rules. These rules simultaneously check whether its physical consistency has been restored or significantly improved, and whether its data consistency is within acceptable limits. The selective rollback and verification module 190 generates a mixed parameter state based on the conflict parameter set and then calls the forward mechanics solution module 130 again for a complete secondary verification. In this way, the identification of conflict parameters and the acceptance of the final state can both rely on the re-solution results rather than just parameter gradients, sensitivity rankings, or abstract contribution values.

[0075] like Figure 1 As shown, the parameter status storage and feedback module 200 uses versioned status records to save the current reference parameter status, current candidate parameter status, each counterfactual rollback status, conflicting parameter set, mixed parameter status, and currently valid parameter status. This module can consist of a status database, a version index, and a commit control program. Only when the mixed parameter status is verified and the status record is completely written will the active status index point to the new currently valid parameter status; if the write fails, the verification fails, or the process is abnormal, the active status index continues to point to the original current reference parameter status. This commit method avoids the misuse of partially written or unverified statuses in the next window.

[0076] like Figure 1 As shown, the exception handling module 210 continuously receives status identifiers submitted by various modules, such as window validity, observation mapping integrity, forward solution validity, parameter range status, iterative convergence status, counterfactual solution status, and storage submission status. Based on the exception occurrence node, the exception handling module 210 performs operations such as window discarding, step size shrinking, task retry, parameter group rollback, overall rollback, or state preservation. The exception handling module 210 does not consider simply issuing a prompt as the end point of processing.

[0077] Even if a local solution fails, system 100 can still retain the state of the previous valid parameters and continue processing in subsequent windows.

[0078] Combination Figure 11 It is known that anomalies such as communication interruption, data loss, solution failure, parameter out-of-bounds errors, deterioration of double consistency, and rollback timeout are all fed into the anomaly handling module 210. The anomaly handling module 210 first determines whether the current anomaly can be partially repaired through the counterfactual rollback state construction module 170, the selective rollback and verification module 190, and resolving. If the partial repair is successful and verification is passed, the parameter state storage and feedback module 200 saves or maintains the most recently valid parameter state. If partial repair is not possible or partial verification fails, the system enters a security degradation phase, sequentially freezing updates, rejecting the current candidate parameter state, maintaining the most recently valid parameter state, and recording the anomaly. This state transition ensures that the output of the anomaly handling module always points to the verified state, rather than to an intermediate state where verification is not yet complete.

[0079] like Figure 2As shown, in step S101, the monitoring window construction module 110 first retrieves monitoring records from the external monitoring data source 10 for the current processing period of the lake area; secondly, it reads engineering status records intersecting with this period from the engineering status data source 20. During this process, the monitoring window construction module 110 establishes window boundaries according to the effective time of the engineering status records and assigns each monitoring record to a unique window. If the timestamp of a monitoring record falls near the state switching boundary, the monitoring window construction module 110 assigns it to either the pre-switching window or the post-switching window according to a pre-determined time allocation rule. This time allocation rule is set before processing begins to ensure that repeated runs yield the same window results.

[0080] like Figure 2 As shown, in the construction of the monitoring window, the monitoring window construction module 110 aligns the time of each monitoring channel within the window. For channels with inconsistent sampling times, the most recent time matching, interval averaging, preserving the previous value, or physically permissible interpolation methods are used; however, the interpolation span cannot exceed the maximum time interval determined based on the sampling period and response change rate. If the number of valid records for a channel in the current window is less than the valid conditions of the current window, the monitoring window construction module 110 marks that channel as unavailable. For the remaining channels that are insufficient to form a parameter update response set and a physical verification constraint set, the exception handling module 210 is activated to update the information of the current window and retain the previous valid parameter state to prevent the system from forcibly generating a new parameter state when the amount of information is too small.

[0081] like Figure 2 As shown, window validity can be determined using the missing rate, time skewness, and quality indicators. Let the total planned record count for the current window be... The number of valid records is The missing proportion can be expressed as

[0082]

[0083] in, Indicates the current window's missing percentage, dimensionless; This indicates the planned number of records obtained based on the sampling configuration and window length, expressed in records. This indicates the number of records verified by timestamp, quality identifier, and value range, expressed in records. The monitoring window construction module 110 will... With upper limit of missing Compare, The determination is made through historical full-window sampling deletion experiments, minimum coverage requirements for monitoring channels, or model observability analysis. The pure numbers 0 and 1 represent the boundary meanings of no missing data and complete missing data, respectively, in the above relationships.

[0084] like Figure 2 As shown, in an exemplary operating condition, when the current window may switch to a later construction phase, the monitoring window construction module 110 does not forcibly retain the original window for a fixed duration, but instead splits it into two sub-windows according to the switching time. If a sub-window is too short to meet the valid recording conditions, this sub-window is placed under the new changed condition, and only the previous valid state exists. In this operating condition, because window splitting prevents the mixing of responses to different boundary states, the subsequent changes in data residuals mainly reflect changes in parameter states, rather than non-parametric changes caused by the switching of construction states.

[0085] like Figure 2 As shown, after the observation mapping and response partitioning module 120 receives the valid window, it maps the monitoring channel to the model output by referring to the mapping configuration in the mechanical model configuration data source 30. For cases where the model nodes and monitoring locations do not completely coincide, model shape function interpolation, local response averaging, or pre-calibrated linear mapping can be used. For responses with inconsistent directions, a coordinate transformation matrix is ​​preferred to convert them to the monitoring direction. The input to the mapping relationship is the model state, and the output is the predicted response with the same physical quantity, unit, and direction as the monitoring record. By clearly defining the observation mapping, direct comparison of non-corresponding model quantities and monitoring quantities can be avoided.

[0086] like Figure 2 As shown, after the observation mapping and response partitioning module 120 obtains the effective window, the monitoring channel is mapped to the model output according to the mapping configuration in the mechanical model configuration data source 30. For cases where the model nodes and monitoring locations do not completely coincide, model shape function interpolation, local response averaging, or pre-calibrated linear mapping are used; for responses with inconsistent directions, a coordinate transformation matrix is ​​used to convert them to the monitoring direction. In this way, the input to the mapping relationship is the model state, and the output is the predicted response with the same physical quantity, unit, and direction as the monitoring record. Therefore, it is clear that the observation mapping no longer needs to directly compare two inconsistent model quantities and monitoring quantities.

[0087] like Figure 2 and Figure 9 As shown, before this round of parameter update, the observation mapping and response partitioning module 120 divides the effective response into a parameter update response set and a physical verification constraint set.

[0088] For example, the system can be divided by monitoring channels, with one group of channels used for parameter updates and another group of channels with different spatial locations or response types reserved for verification; alternatively, it can be divided by time slices, with responses before the window used for parameter updates and responses after the window used for verification; or displacement response can be used as the update target, and independent strain responses, constitutive state continuity, or boundary responses used for physical verification. Regardless of the division method, the division result is written into the window record before the start of the current round and remains unchanged after the candidate state is formed. This pre-division mechanism ensures that independent physical verifications are not directly minimized by the parameter update target, thus enabling the detection of compensatory parameter updates.

[0089] Figure 9 The update response in the upper branch corresponds to the parameter update response set, and its predicted response and monitoring response jointly enter the parameter incremental update module 140; the verification constraint in the lower branch corresponds to the physical verification constraint set, and its predicted quantity and verification quantity jointly enter the double consistency calculation module 150. The mechanics forward solution module 130 outputs the corresponding predicted quantities for the two types of uses, but only the update branch participates in the construction of the parameter update target in this round, thereby maintaining the data usage independence of the verification branch relative to the update target in this round.

[0090] like Figure 2 As shown, the parameter state storage and feedback module 200 reads the current reference parameter state. In the first window, it is provided by the initial parameters and the parameter boundary data source 40; while in the subsequent second window, it is the current valid parameter state submitted in the previous window. For implementations using path-dependent constitutive relations, the parameter state storage and feedback module 200 also needs to read the internal state snapshot corresponding to the starting point of the window, or read the historical index of the reconstructable internal state, so that subsequent reference states, candidate states, and counterfactual backoff states are all solved from the same historical benchmark. This avoids the use of inconsistent historical internal variables for different parameter states and ensures the comparability of backoff comparisons.

[0091] like Figure 2 As shown, in one implementation, the current monitoring window can be a sliding window; adjacent windows retain some overlapping data to maintain the temporal continuity of parameter states. When using a sliding window, the parameter state storage and feedback module 200 still only uses the previously submitted currently valid parameter state as the reference state, and does not use intermediate states that have not yet been reviewed. This limitation prevents overlapping data from causing unreviewed states to propagate repeatedly.

[0092] like Figure 2 As shown, in step S102, the forward mechanical solution module 130 loads the mechanical model configuration according to the current engineering state and performs forward calculations before the update with the current reference parameter state. Let the current window number be... The internal state of the model is The surrounding rock mechanical parameter vector is The current project status is Then the forward mechanical model can be uniformly represented as

[0093]

[0094] in, It represents the residual operator composed of mechanical equilibrium relations, constitutive update relations, boundary conditions, and discrete equations; This represents the set of displacement, stress, strain, or internal state variables corresponding to the current window. This represents a complete vector of surrounding rock mechanical parameters, with each component having a clear physical meaning and unit. This indicates the current engineering status. This expression does not limit the specific numerical solution method, only requiring that the forward mechanics solution module 130 can solve the given... and The effective model state is obtained in time.

[0095] like Figure 2 As shown, the observation mapping and response partitioning module 120 converts the internal states of the model into comparable predicted responses, as follows:

[0096]

[0097] in, The meaning is parameter state The predicted response vector within the current window; The meaning is the observation mapping corresponding to the current window, which specifically includes position interpolation, direction transformation, response extraction and unit transformation. The output of the observation mapping is respectively entered into the prediction component corresponding to the parameter update response set and the prediction component corresponding to the physical verification constraint set.

[0098] like Figure 2 As shown, the parameter incremental update module 140 constructs a data update target using the parameter update response set. In one embodiment, the data update target can be represented as...

[0099]

[0100] in, The meaning is the data update target of the current window, a dimensionless quantity or a quantity that has been uniformly normalized; This represents the predicted response corresponding to the parameter update response set; The meaning is the monitoring response corresponding to the parameter update response set; The error covariance matrix represents the parameter update response set, and its source can be monitoring and calibration data, stability window residuals, or repeated measurement errors.

[0101] If information on the correlation between channels is lacking, a diagonal matrix composed of the variances of each channel can be used; if the noise of a certain channel increases significantly, the covariance weighting will reduce the channel's dominance on parameter updates.

[0102] like Figure 2 As shown, the parameter incremental update module 140 is at the lower bound of the parameter. and parameter upper bound The current candidate parameter state is obtained between these steps. .

[0103] Specifically, updates can employ damped Gaussian-Newton methods, trust regions, constrained gradients, Levenberg-Marquardt methods, or other deterministic iterative methods. For updates requiring gradients, these can be obtained using automatic differentiation, implicit differentiation, adjoint methods, or finite differences. It should be noted that the above differentiation methods are only used to form candidate parameter states and do not change the core logic of subsequent consistency conflict attribution.

[0104] like Figure 2 As shown, for example, when the parameters to be updated in this round are multiple mechanical parameters, the parameter incremental update module 140 can limit the variable to be updated in this round based on existing parameter grouping or joint identifiable analysis, or it can directly update all candidate parameters. However, regardless of the type of the set of parameters to be updated, it outputs a list of parameter indices that have changed in this round, enabling the construction of a counterfactual rollback state. This counterfactual rollback state construction module 170 only performs rollback tasks for parameters that have already been updated in the last time, preventing meaningless counterfactual solutions for parameters that will change in the future.

[0105] like Figure 2 As shown, after each trial update, the trial parameter projection is projected onto the physical boundary by the parameter incremental update module 140, and then the mechanics forward solution module 130 is called to solve again. If the forward solution fails due to the trial parameters, the internal state violates the constitutive acceptable conditions, or the data update target deteriorates significantly, the parameter incremental update module 140 reduces the update step size or increases damping before recalculating; if a valid candidate state cannot be formed even after reaching the iteration limit, the exception handling module 210 can terminate the current update and maintain the current reference parameters. In this way, this degradation path ensures that failure is not mistaken for a consistency conflict.

[0106] like Figure 2As shown, in one implementation using differentiable mechanics, the forward mechanics solution module 130 organizes the parameter-dependent constitutive update, element residuals, and solution process into a differentiable computational chain, while the incremental parameter update module 140 directly obtains the gradient of the predicted response relative to the parameters. In another implementation, the forward mechanics solution module 130 maintains a standard numerical solver, and the incremental parameter update module 140 obtains the update direction through parameter perturbation or adjoint calculation. Both paths form the current candidate parameter state with the same current engineering state, the same parameter update response set, and the same parameter boundary, thus both can enter the subsequent double-consistency conflict attribution process.

[0107] like Figure 2 and Figure 3 As shown, in step S103, the dual consistency calculation module 150 calculates the current reference parameter state and the current candidate parameter state respectively. Assume the parameter update response set contains... If there are 1 valid response component, then the state of a certain parameter is... Data consistency residuals can be expressed as

[0108]

[0109] in, This represents the data consistency residual for the current window; Indicates the number of valid response components in the parameter update response set; superscript This indicates the transpose of a vector; the remaining symbols are consistent with the previous definitions.

[0110] Using the root mean square form can reduce the direct impact of changes in the number of responses on the residual scale, and using error covariance weighting can prevent high-noise channels from dominating conflict triggering due to large numerical amplitudes.

[0111] like Figure 3 As shown, independent physical consistency is not forced to be merged into a single weighted total score, but is instead retained as a component vector:

[0112]

[0113] in, Represents the independent physical consistency residual vector of the current window; The meaning is the first Normalized residuals of physical verification constraints; The meaning refers to the number of physical verification constraint types. Each component is divided by its own allowable boundary to form a comparable state without changing the physical meaning; in the normalized residual... When the value of is no more than 1, it indicates that the th is... For physical verification, within the corresponding allowed boundaries, the pure number 1 represents the normalized boundary and does not need to be treated as a specialized symbol.

[0114] like Figure 3 As shown, in the implementation method that uses reserved monitoring response, the first... The physical consistency component is determined by the noise-weighted residual of the reserved response; in the implementation method using cross-window state continuity, the first... The class component is determined by the difference between the internal state of the current window starting point and the result of the previous valid state transmission.

[0115] In embodiments employing constitutive state acceptability, the first The class component is determined by the degree of violation of plastic dissipation, damage evolution, yield state, or other constitutive constraints. In the implementation using an independent reference model, the first... Class components are determined by the difference between the predicted results of candidate parameters in the reference model and the validation conditions. The inputs, calculations, and allowable boundaries of different components are recorded separately to avoid general weighting without engineering basis.

[0116] like Figure 3 As shown, the consistency conflict triggering module 160 calculates the data consistency improvement and the degradation of each physical component:

[0117]

[0118]

[0119] in, The meaning is the current reference parameter status; Indicates the current state of the candidate parameters; The meaning of "positive timing" is that data consistency has been improved. When positive, it indicates the first Physical consistency deteriorates. To eliminate redundant calculation fluctuations and monitoring noise, System 100 only... Greater than the data fluctuation threshold and at least one Greater than the corresponding physical fluctuation threshold Furthermore, when the physical residual component of the candidate state exceeds the normalization allowable boundary, it is determined that there is a consistency conflict.

[0120] like Figure 3 As shown, data fluctuation threshold The physical fluctuation threshold can be determined by the fluctuation of residuals from repeated calculations under the same input conditions, monitoring repeatability errors, or the distribution of residuals within a stable window; The threshold can be determined by the error of repeated calculations of the reference model, the numerical error of constitutive integration, the error of state transfer, or the error of reserved monitoring response. The engineering role of the threshold is to separate repeatable numerical fluctuations from the real consistency changes that require triggering parameter attribution, rather than artificially raising the trigger threshold.

[0121] like Figure 3 As shown, the dual consistency changes are divided into four categories. The first category is where both data consistency and independent physical consistency improve, in which case the current candidate parameter state enters a full review. The second category is where data consistency improves but at least one independent physical consistency deteriorates, in which case a counterfactual rollback is initiated. The third category is where data consistency deteriorates but independent physical consistency improves, in which case the parameter incremental update module 140 reduces the step size, changes the update parameter set, or rejects the current update. The fourth category is where both types of consistency deteriorate, in which case the exception handling module 210 restores the current reference parameter state as a whole. This four-category state diversion ensures that conflict attribution only handles local parameter compensation issues and is not mixed with ordinary update failures.

[0122] Figure 3 The graph uses data consistency changes as the horizontal axis and independent physical consistency changes as the vertical axis. In the accompanying figures, physical consistency used for compressed text refers to independent physical consistency. The "double excellent," "number excellent but object poor," "number poor but object excellent," and "double poor" states in the graph correspond one-to-one with the aforementioned four states. The area containing "number excellent but object poor" is the only normal trigger area for initiating counterfactual conflict attribution. Other areas proceed to candidate review, parameter update adjustment, or overall rollback, respectively, thus avoiding the erroneous inclusion of ordinary update failures into conflict parameter attribution.

[0123] like Figure 3 As shown, in an exemplary operating condition, assuming the parameter update response set fits the displacement response well, but the reserved strain response or the continuity of the state within a certain window exceeds the allowable range, system 100 enters the second type of state. At this time, even if the data update target decreases, these candidate parameters cannot be directly submitted to the candidate parameter state. By adding independent physical verification, the system can discover compensatory parameter changes that are difficult to expose by data fitting targets alone.

[0124] like Figure 3 As shown, in one implementation, the physical verification constraint set does not use reserved monitoring channels, but instead uses a separate reference model. Since the incremental parameter update task is completed by the computationally less computationally intensive main model, the candidate states and counterfactual backoff states are verified by the reference model, which has higher accuracy or a different discretization method. Because it does not participate in the current update objective, the approximation error of the main model is less likely to simultaneously mask the parameter update. This alternative approach is suitable for scenarios where the number of monitoring channels is too small to distinguish between the update set and the reserved set.

[0125] like Figure 2 and Figure 4 As shown, in step S104, the counterfactual rollback state construction module 170 reads the set of updated parameter indices for this round. Assume the complete parameter vector contains... There are [number] parameters, and the parameter index set has been updated in this round. The set of indices corresponding to a certain tested parameter unit is Then the first A counterfactual backoff state can be constructed using the following formula:

[0126]

[0127] in, Indicates the first The first counterfactual regression state Each parameter value; Indicates the first in the current reference parameter state Each parameter value; Indicates the first candidate parameter in the current candidate parameter state. Each parameter value; This indicates that the parameter set has actually been updated in this round; Indicates the first There are 10 test parameter units. This segmentation ensures that if the test parameter is revoked from its update, other updated parameters retain their candidate values, and parameters not updated in this round retain their reference values.

[0128] like Figure 4 As shown, in the single-parameter backoff implementation, each tested parameter unit It contains only one parameter index. The counterfactual fallback state construction module 170 sequentially copies the current candidate parameter state, replaces one parameter, and generates an independent task identifier. The task identifier records the index of the parameter being tested, the reference state version, the candidate state version, the current engineering state version, and the model version. The mechanics forward solution module 130 executes the task only when these versions are consistent, thereby avoiding the mixing and comparison of solution results from different engineering states or different model configurations.

[0129] like Figure 4 As shown, for the implementation using path-dependent constitutive relations, the forward mechanics solution module 130 cannot simply combine the internal variables of the candidate states with the backoff parameters directly. The system 100 starts from the same historical state snapshot at the beginning of the current window and replays the forward process of the current window with the current reference parameter state, the current candidate parameter state, and each counterfactual backoff state. If the window contains multiple load increments, the same increment sequence is used for all three types of states. This process ensures that the counterfactual comparison reflects the effect of parameter cancellation itself, rather than differences in internal variable initialization.

[0130] like Figure 4As shown, the forward mechanical solution module 130 recalculates the predicted quantities corresponding to the parameter update response set and physical verification constraint set for each counterfactual backoff state. If a finite element method is used, the mesh, boundary index, and parameter-independent matrix structure can be reused, and only the constitutive terms or stiffness terms affected by parameter changes need to be reassembled. If a validated reduced-order model is used, the state basis can be reused, but the parameter correlation coefficients can be recalculated. The above reuse method only reduces the overhead of repeated solutions and does not change the requirement that the mechanical response must be regenerated for each counterfactual state.

[0131] like Figure 4 As shown, in an exemplary operating condition, the current candidate state updates multiple parameters, while the consistency of independent physical states deteriorates. System 100 sequentially constructs counterfactual states that recover only the first parameter, only the second parameter, and only all subsequent parameters. If physical consistency is restored after recovering a certain parameter, but recovering other parameters does not produce the same effect, then the current update of this parameter has conflicting implications. Compared to directly sorting based on gradient magnitude, resolving the problem can reveal the combined effects of nonlinear constitutive relations, parameter coupling, and boundary states on the actual response.

[0132] Figure 4 The parameters A, B, and C, as well as the rollback parameters A, B, and C, are used only to illustrate the state replacement relationship of single-parameter counterfactual rollback and do not limit the specific names, number, or order of the candidate parameters. Taking rollback A as an example, only parameter A is restored to the reference value, while parameters B, C, and other parameters updated in this round retain their candidate values; parameters not updated in this round continue to retain their reference values. Each rollback state is submitted to the forward mechanics solution module 130, thus enabling comparable re-solutions under the same engineering state and the same model version.

[0133] like Figure 4 As shown, if many parameters have been updated in this round, the counterfactual rollback state construction module 170 can first construct a single-parameter task, and then decide whether to construct a parameter group based on the results of the single parameter. For parameters with clear constitutive relationships, parameter groups can be pre-constructed, such as parameter groups in the same strength criterion, parameter groups in the same creep branch, or the same spatial parameter block. Parameter group rollback still uses the same state construction relationship, only the parameter unit to be tested is changed. This has been extended to multiple indexes, and this alternative approach is often used when there are multiple parameters that compensate together, and physical consistency cannot be restored by rolling back any one parameter.

[0134] like Figure 4As shown, when a counterfactual backoff state causes the forward mechanical solution to fail to converge, the anomaly handling module 210 does not classify this task as evaluable, but directly treats it as a conflict of the tested parameters. The counterfactual backoff state construction module 170 can narrow down the parameter set obtained after the backoff, use a more conservative step size, or switch to an alternative solution configuration for recalculation; if a suitable solution result still cannot be obtained, the tested parameter unit will not be included in the current conflict parameter set, and other tested parameter units will continue to be evaluated. This degradation logic prevents numerical solution failures from being mistakenly identified as a failure of parameter attribution.

[0135] like Figure 2 and Figure 5 As shown, in step S105, the conflict parameter set determination module 180 compares the current candidate parameter state with the double consistency results of each counterfactual backoff state. For the first... For each tested parameter unit, the conflict parameter set determination module 180 checks at least two conditions: First, after revoking the update of the parameter unit, the deteriorated independent physical consistency components recover to within the allowable boundaries, or achieve an improvement exceeding the physical fluctuation threshold compared to the candidate state; second, after revoking the update of the parameter unit, the data consistency residual still does not exceed the allowable data boundary. Only when both conditions are met simultaneously is the tested parameter unit included in the conflict parameter candidate.

[0136] like Figure 5 As shown, the data allowable boundary can be determined by monitoring error, stable residuals of the parameter update response set, or engineering allowable fitting error; the physical allowable boundary is determined by reserved response error, state continuity tolerance, constitutive state allowable range, or independent model error, respectively. System 100 does not require the data residuals of the counterfactual backoff state to be better than the current candidate state, because backoff conflicting parameters may sacrifice some data fitting improvement; however, the data residuals must still be within the acceptable boundary to ensure that selective backoff does not cause the parameter state to lose its basic interpretability of the monitoring response.

[0137] like Figure 5 As shown, when multiple single-parameter rollback states all meet the conflict conditions, the conflict parameter set determination module 180 can adopt a one-time inclusion method, adding all parameters that meet the conditions to the initial conflict parameter set; or it can adopt a greedy sequential method, first selecting the parameter that maximizes the improvement of the main physical residual, forming a temporary mixed state, and then re-evaluating the remaining parameters. The former method has fewer calculation steps and is suitable for situations where the conflict points between parameters are relatively independent; the latter method can handle situations where the conflict states of other parameters change after rolling back one parameter.

[0138] like Figure 5 and Figure 10As shown, when no single parameter rollback can restore physical consistency, the conflict parameter set determination module 180 enters the parameter group search. The parameter group search proceeds from smallest to largest parameter group size, first evaluating combinations containing two parameters, and then increasing the combination size as needed. For multiple parameter groups of the same size, an ordered selection rule is used: first, select the parameter group that can restore all target physical residuals to the allowable boundary; if multiple parameter groups satisfy this rule, select the parameter group with smaller data residuals; if they are still the same, select the combination with lower overlap with conflict records in the previous stable window or with fewer parameters. This ordered rule avoids arbitrarily weighting multiple indicators into a single comprehensive score.

[0139] like Figure 5 As shown, the number of parameter combinations can increase with the number of updated parameters. To avoid excessive computation time for a single window, which could cause the system to malfunction, the exception handling module 210 limits the maximum number of combinations based on the remaining tasks after each window's computation is completed. Before the specified time limit is reached, if at least one parameter combination meets the criteria, the smallest qualified parameter combination found is selected; if no qualified parameter combination is found, a complete rollback is initiated. Therefore, before the window has to be closed, the process terminates as long as a path that is still usable when resources are insufficient is encountered.

[0140] Combination Figure 10 It is known that the search for the minimum conflict parameter group will start with a candidate group containing two parameters, with the failure of a single parameter as the entry condition. The counterfactual rollback state construction module 170 generates candidate groups at the current scale, and the mechanics forward solution module 130 resolves each group; parameter groups that meet the physical recovery and data acceptability conditions are recorded as qualified groups. After the evaluation of candidate groups at the same scale is completed, if a qualified group exists, the smallest qualified group at the current scale is selected and handed over to the selective rollback and verification module 190 for processing; if no qualified group exists, the group is added. If the maximum scale is reached or the calculation time limit is reached and no qualified group is found, the anomaly handling module 210 performs an overall rollback.

[0141] like Figure 5 and Figure 6 As shown, the selective rollback and verification module 190 performs a rollback based on the final set of conflict parameters. Forming a mixed parameter state:

[0142]

[0143] in, Represents the first state in the mixed parameter state Each parameter value; This represents the set of conflicting parameters; the remaining symbols are consistent with the previous definition. This relationship indicates the fate of the three parameters: parameters that have been conflicted are revoked from updating; non-conflicting parameters that have already been updated retain their candidate values, while parameters that have not been updated in this round continue to be used as references.

[0144] like Figure 6 As shown, after the mixed parameter state is formed, the mechanics forward solution module 130 re-executes the complete forward calculation from the window starting point and engineering state consistent with the aforementioned state. The double consistency calculation module 150 recalculates all parameter update responses and all physical verification constraints, instead of only checking the single residual component that has conflicted. The selective rollback and verification module 190 also checks whether the parameter boundaries, solution validity, and path-dependent internal states are valid. Only when all conditions are met can the mixed parameter state enter the current valid parameter state.

[0145] Figure 6 The conflict parameter reference, non-conflict candidate protection, and unupdated reference protection at the top correspond to the value sources of conflicting parameters, non-conflicting updated parameters, and parameters not updated in this round, respectively. The data verification, physical verification, range verification, and solution verification in the diagram correspond to data consistency verification, independent physical consistency verification, parameter boundary verification, and the validity verification of the forward mechanical solution, respectively. The current valid parameter state can only be saved after all four conditions are met; if any one condition fails, the mixed parameter state cannot be directly submitted.

[0146] like Figure 6 As shown, in one operating condition, only a portion of multiple parameters are selected for updating, resulting in a mismatch between reserved responses or constitutive states. Therefore, the selective rollback and verification module 190 revoks the updates of these parameters, but retains the updated values ​​of other parameters formed in the current window. Compared to a full rollback, this process reduces the range of revoked parameters; and compared to a full acceptance, this process blocks updates that are determined to be conflicting by independent physical verification. Thus, the granularity of parameter state restoration is reduced from the entire set of parameters to attributable parameters or parameter groups.

[0147] like Figure 6 As shown, in one implementation, independent physical consistency is checked using a multi-layered component approach. System 100 can first check the constitutive state component, then the state continuity component, and finally the reserved response component; alternatively, the check order can be determined according to engineering safety relevance. Each component uses its own permissible boundary, and the mixed-state check fails if any required item is not met. This layered check method preserves the engineering meaning of each physical constraint, preventing improvements in one component from offsetting severe deterioration in another.

[0148] like Figure 6As shown, if the mixed parameter state fails the review, the exception handling module 210 first expands the conflicting parameter set, adding parameters that significantly improve the physical residual to the rollback set and recalculating them; if the single parameter expansion still fails, then parameter group rollback is initiated; if all updated parameters are included in the conflicting parameter set, the complete current reference parameter state is restored. System 100 does not save unreviewed partial states, thereby preventing the propagation of partial rollback failure results.

[0149] like Figure 2 and Figure 6 As shown, in step S106, the parameter state storage and feedback module 200 generates a new state version for the reviewed mixed parameter states. This version stores at least the current window identifier, current project state version, model version, current reference parameter state version, list of updated parameters for this round, set of conflicting parameters, final parameter values, data consistency residuals, individual physical consistency residuals, solution validity identifier, and submission time. By preserving the complete source chain, the parameter state formation process for this round can be reconstructed in subsequent windows or during review and reproduction.

[0150] like Figure 6 As shown, a write-then-activate approach is used when submitting the status. Specifically, the parameter status storage and feedback module 200 first writes the new status to the inactive version area, then verifies whether the record is complete and consistent with the associated version; after successful verification, the active status index is switched to the new version. If a disconnection, failure, or process exception occurs during this write process, the active status index cannot become the original current reference parameter status. In other words, the system will not use an incomplete parameter record as the reference status for the next window due to a storage failure.

[0151] like Figure 7 As shown, when the current valid parameter status enters the next monitoring window, only the complete parameter status and necessary internal status snapshots are transmitted, without directly transmitting the conflicting parameter set from the previous window as a fixed judgment result. After the next window arrives, the monitoring window construction module 110 re-forms the window, the parameter incremental update module 140 regenerates the candidate status, and the double consistency calculation module 150 re-judges conflicts. In this way, if a parameter is judged to be in conflict in the previous window, it will not be permanently prohibited from updating in the next window; after changes in the engineering status, monitoring information, and parameter coupling relationship, the parameter can still participate in the update again.

[0152] Figure 7The cross-marked conflict sets do not inherit, meaning that only the verified complete parameter state and necessary internal state snapshots are fed back to the next monitoring window, without fixing the conflict parameter set of the previous window into the preset prohibited update set of the next window. This diagram relationship, together with the parameter state feedback arrows, defines the cross-window processing rules for state inheritance and conflict conclusion re-determination.

[0153] like Figure 7 As shown, for path-dependent constitutive models, the parameter state storage and feedback module 200 simultaneously stores internal state snapshots or reconstructable indexes corresponding to the currently valid parameter states. At the start of the next window, the mechanics forward solution module 130 continues calculation from this valid snapshot, rather than from the candidate state or the failed counterfactual state. This state isolation prevents internal variable contamination caused by abnormal solutions from affecting subsequent windows.

[0154] like Figure 7 As shown, when a sudden change in engineering state occurs between the next window and the current window, system 100 first loads the new boundary and load configuration from the engineering state data source 20, and then uses the current valid parameter state as the parameters. Because the physical verification components brought about by the switch in engineering state are not comparable, the observation mapping and response partitioning module 120 re-establishes the verification constraints before the new window begins; for the internal states that still have continuity, they are passed according to the model rules, so that not only can the parameter values ​​be fed back across windows, but the physical verification conditions can also be updated, so as to prevent the old window verification boundaries from being applied to the new engineering state.

[0155] like Figure 7 As shown, in one implementation, system 100 employs event-triggered feedback. When a new monitoring window fails to meet data integrity conditions, the parameter state storage and feedback module 200 maintains the current valid parameter state without performing the next round of inversion; when certain windows meet relevant requirements or significant changes occur in the engineering state, S101 to S106 can be reopened. This reduces unnecessary window calculations while ensuring that each changed state has a set of input and verification conditions.

[0156] like Figure 7 As shown, if the parameter status writing fails, there is a status version conflict, or the activity status index update times out, the exception handling module 210 locks the currently valid parameter status, prohibits the next window from reading the incomplete status, and prevents retrying the status submission. If continuous retries still fail, the system 100 continues to use the previous activity version and records the reason for the next failure to submit. That is, this path is set to have a clear acknowledgment between the parameter calculation result and the effective status, which can avoid the situation where a status without acknowledgment is mistakenly considered to have taken effect.

[0157] like Figure 3 , Figure 5 and Figure 9As shown, in the first parallel embodiment, the observation mapping and response partitioning module 120 partitions the current window response into a parameter update response set and a reserved monitoring response set according to the monitoring channel or spatial location before the parameter update begins. The parameter update response set is used for incremental parameter updates, while the reserved monitoring response set is only used for independent physical verification. The two sets of responses can have the same physical quantity but come from different locations, or they can have different physical quantities; for example, parameter updates use one set of displacement responses, while independent verification uses the displacement or strain responses from another location. The partitioning results are saved together with the current window, allowing subsequent counterfactual backoff to compare the same reserved set.

[0158] Figure 9 Further illustrating the usage isolation relationship in the reserved monitoring response implementation: the observation mapping and response partitioning module 120 completes the partitioning before entering the parameter incremental update module 140 in the current window, and writes the partitioning identifier into the window record; subsequent candidate states, counterfactual fallback states, and mixed parameter states all use the same partitioning result. In this way, changes in the reserved response can participate in the double consistency calculation module 150 as independent verification results, without losing comparability due to repeated partitioning adjustments within the same round. Figure 3 As shown, the dual consistency calculation module 150 uses the noise-weighted residual of the parameter update response set as the data consistency residual and the noise-weighted residual of the reserved monitoring response as an independent physical consistency component. A counterfactual backoff is triggered only when the residual of the update set improves while the residual of the reserved set deteriorates and crosses the allowable boundary. Since the reserved response is not included in the parameter update target, candidate parameters cannot mask parameter compensation by directly minimizing the reserved residual, thus enabling the reserved response to serve as independent verification information.

[0159] like Figure 5 As shown, the counterfactual rollback state construction module 170 generates tasks using a single-parameter method. The conflict parameter set determination module 180 uses a greedy order: first, it selects the parameter that maximizes the recovery magnitude of the reserved response residual and ensures that the updated response residual is still within the allowable boundary as the rollback parameter; after a temporary mixture is formed, the single-parameter test is repeated for the remaining updated parameters. This embodiment does not require exhaustively listing all parameter sets and is suitable for situations where there are many monitoring channels and a single conflict parameter has a clear impact on the reserved response.

[0160] like Figure 5 As shown, if the rollback of a certain parameter restores the reserved response, but the updated response exceeds the allowable boundary, then that parameter will not be included separately in the conflict parameter set, and the system will continue to evaluate other parameters or parameter combinations. Therefore, this double boundary condition ensures that the system only wants to restore the updated response that occurred under the given conditions, without completely losing its ability to interpret the updated response. The conflict parameters determined in this way can simultaneously satisfy both the conditions of physical verification recovery after reversal and the preservation of basic data fitting.

[0161] like Figure 6 As shown, in this parallel embodiment, the mixed parameter state after selective rollback is re-verified against the updated response set and all reserved responses. If the verification passes, the parameter state storage and feedback module 200 saves the current valid state; if the verification fails, the exception handling module 210 expands the rollback set or restores the overall reference state. This embodiment constructs independent verification using monitoring data space or channel redundancy, and is suitable for operating environments that can reserve some monitoring responses and do not require an additional reference model.

[0162] like Figure 3 and Figure 7 As shown, in the second parallel embodiment, when the number of available monitoring responses in the current window is insufficient to divide independent reserved channels, the observation mapping and response partitioning module 120 uses all valid monitoring responses for parameter updates. Furthermore, cross-window state continuity and constitutive acceptability are used as physical verification constraints. This approach is particularly suitable for path-dependent elastoplastic, creep, or damage models, where parameter updates affect not only the current response but also the transmission of internal states between adjacent windows.

[0163] like Figure 7 As shown, the parameter state storage and feedback module 200 retains a snapshot of the internal state at the end of the previous valid window. At the beginning of the current window, the mechanics forward solution module 130 calculates the internal state evolution of the reference state based on the current engineering state and this snapshot; for candidate states and each counterfactual backoff state, it re-enacts the state starting from the snapshot at the same window start point; the double consistency calculation module 150 calculates the state continuity residuals at the window start and window end points, for example, whether the same state variable is a jump variable under conditions without physical abrupt changes, or the difference between the state at the end of the previous window and the initial reconstructed state of the current window. The allowable boundary of state continuity is determined by the model integral error, the engineering state switching type, and historical stable window fluctuations.

[0164] like Figure 3 As shown, for the implementation using a plastic constitutive model, physical verification constraints may also include yield state, plastic dissipation, or acceptable plastic multiplier conditions; for the implementation using a damage model, they may include damage variable value boundaries and irreversible evolution conditions; for the implementation using a creep model, they may include the continuity of creep internal variables and the direction of temporal evolution. These constraints each generate an independent physical consistency component and do not mask the failure of a necessary component through ordinary weighted total scores.

[0165] like Figure 5As shown, when data consistency improves while the continuity or constitutive acceptability of a certain internal state deteriorates, the counterfactual backoff state construction module 170 forms tested parameter units according to the constitutive submodule to which the parameter belongs. For example, multiple parameters in the same creep branch can be backoffed as a parameter group, and parameters in the same strength criterion can be evaluated at two levels: single parameter and parameter group. The conflict parameter set determination module 180 records the recovery effect of each tested parameter unit on different physical components, thereby distinguishing between state-transfer conflict parameters and constitutive state conflict parameters.

[0166] like Figure 6 As shown, the selective rollback and verification module 190 first recovers the conflicting parameters directly related to the failed physical component, and then verifies all physical components. If state continuity is recovered, but the constitutive model can be passed, the set of these conflicting parameters is expanded; if all conflicting parameters pass and are consistent within the allowable range, the mixed parameter state is saved. This embodiment does not rely on additional monitoring channels and constructs independent physical verification based on the internal state of the path-dependent model, which can detect candidate parameters that cause the model evolution mechanism to be disrupted when monitoring data is limited.

[0167] like Figure 1 , Figure 5 and Figure 7 As shown, in the third parallel embodiment, the mechanical model configuration data source 30 simultaneously stores both the parameter update model and the independent reference model. The parameter update model is used to quickly generate the current candidate parameter state, while the independent reference model is used to evaluate the reference state, candidate state, counterfactual backoff state, and mixed parameter state. The two models may employ different discretization accuracies, different solution methods, or different model degradation levels, but they should accept the same current engineering state and parameter state, and output comparable verification responses.

[0168] like Figure 3 As shown, the data consistency residual is still determined by the fitting result of the parameter update response set in the parameter update model; the independent physical consistency residual is determined by the difference between the verification response and the physical verification constraint in the independent reference model for the same parameter state. Since the independent reference model does not participate in the parameter update objective, the approximation error in the parameter update model is less likely to reduce the same objective function together with parameter changes and thus mask the physical mismatch.

[0169] like Figure 5 and Figure 10As shown, when a consistency conflict occurs, in the counterfactual rollback state construction module 170, a single-parameter rollback state is first formed and resolved by the independent reference model. If the rollback of all individual parameters fails to restore independent physical consistency, the search for the minimum conflict parameter set begins. Starting with combinations containing two parameters, the search increments according to the size of the parameter set; each time, the reference value is restored, the independent reference model is recalculated, and the double consistency and boundary are determined. Once a combination that meets the conditions is found, larger combinations are no longer evaluated; therefore, the parameter set with fewer conflicts is prioritized.

[0170] Figure 10 When the parameter group search works in conjunction with the independent reference model, each candidate parameter group is recalculated using the same current project state, the same reference state version, and the same reference model version. Recording a qualified group only indicates that the parameter group that meets the physical recovery and data preservation conditions is retained, and does not mean that the parameter group has taken effect. Only after the selective rollback and verification module 190 completes a second verification of the mixed parameter state formed by the selected minimum qualified group is the group of parameters formally included in the conflict parameter set.

[0171] like Figure 7 As shown, to control the overhead of repetitive solving of the reference model, the forward mechanics solver module 130 can cache parameter-independent meshes, boundary indices, and observation maps, and reuse them in counterfactual tasks. The system can also pre-form candidate groups based on the constitutive module to which the parameters belong, spatial parameter blocks, or historical correlations, thereby reducing the number of combinations without physical correlations. Caching and pre-screening of candidate groups only change the computation order; they do not change the requirement that each final candidate conflict group must be re-solved using an independent reference model.

[0172] like Figure 6 As shown, after the minimum conflict parameter set is determined, the selective rollback and verification module 190 performs verification in a mixed parameter state, namely the parameter update model and the independent reference model. If both models pass, the parameter state storage and feedback module 200 submits the current valid state; if either model fails, the exception handling module 210 expands the parameter set or performs a complete rollback. This embodiment uses model independence and searching for the minimum parameter set to constrain the rollback range, matching a computing environment with a fast-updating model and a highly reliable reference model.

[0173] like Figure 11As shown, this implementation divides global anomaly handling into three levels: anomaly monitoring, local repair, and security degradation. When anomalies such as communication interruption, data loss, solution failure, parameter out-of-bounds errors, deterioration of double consistency, and rollback timeout enter the anomaly handling module 210, it first determines whether local repair can be completed through the counterfactual rollback state construction module 170, selective rollback and verification module 190, and resolving. If local repair fails or the conditions for local repair are not met, the current update is frozen, and the most recently valid parameter state is maintained. The following anomaly embodiments are described separately. Figure 11 Different abnormal entry points, degradation actions, and recovery conditions.

[0174] like Figure 1 and Figure 7 As shown, when system 100 receives data from external monitoring data source 10 or engineering status data source 20 via network interface, each batch of data includes at least window identifier, data version, number of records, and integrity verification information. The monitoring window construction module 110 returns a receipt after receiving the data; if all data cannot be obtained within the time limit, and there are discrepancies between records and declarations, and integrity verification has been performed but the true version of the data cannot be found, the data will be displayed as abnormal; at this time, the exception handling module 210 will mark the input batch as unavailable and prevent step S102 from starting.

[0175] like Figure 7 As shown, the reception time limit is determined by the window arrival period, normal network delay distribution, and data batch size, rather than using a fixed value without a source. During communication interruption, the parameter status storage and feedback module 200 locks the currently valid parameter status and does not splice partially received data into a new window. After the network is restored, the monitoring window construction module 110 re-requests the missing batch according to the window identifier; if the missing window has exceeded the allowed retransmission period, it is skipped and restored from the next complete window. This process ensures that data disconnection does not result in incomplete parameter status from the source, nor does it destroy the previous valid status.

[0176] like Figure 1 As shown, in one implementation using database reading, data integrity is not determined by the data frame checksum, but by the transaction version, total number of records, and commit identifier. The monitoring window construction module 110 only loads data when both the monitoring record and the project status record in the same window are in the committed version. This alternative approach maintains the core mechanism of parameter updates only after the input is complete, and is suitable for local databases or shared storage environments.

[0177] like Figure 1 and Figure 2 As shown, if a monitoring channel continuously outputs values ​​exceeding its physical range, the quality identifier is invalid, the timestamp is in reverse order, the duplicate recording ratio exceeds the duplicate limit, or the missing ratio is also considered. Exceeding the missing limit If so, the monitoring window construction module 110 will remove the channel from the current valid responses. (Missing upper limit) The decision is based on the historical complete window resampling, the current minimum coverage requirement of the monitoring channel, or the identifiable conditions of the monitoring channel response.

[0178] like Figure 2 As shown, after removing an abnormal channel, the observation mapping and response partitioning module 120 uses the remaining channels to check whether a parameter update response set and a physical verification constraint set can be formed simultaneously. If both sets can be formed, the system 100 continues to run with dimensionality-reduced input and records the removed channel and the reason in the status log; if there is only one parameter update response and no independent physical verification can be generated, the current window does not perform consistency conflict attribution, nor does it submit a new valid parameter state, or if there is no channel for parameter update in the current window, the current window is skipped. In this way, even if an accurate conclusion cannot be obtained, the system avoids writing a candidate parameter as a valid state in the absence of separate verification of the candidate parameter by relying solely on hierarchical judgment to determine whether an independent physical verification problem exists.

[0179] like Figure 1 As shown, when data exists in a monitoring channel but cannot be mapped to a position, orientation, or physical quantity in the current model version, the observation mapping and response partitioning module 120 first verifies whether the mapping version is consistent with the engineering status version. If it is only an index change caused by model mesh updates, the pre-saved spatial mapping relationship can be called for repositioning; if a definite mapping cannot be established, the channel needs to be removed. This avoids directly comparing model responses at different locations or in different orientations with monitoring values.

[0180] like Figure 1 and Figure 7 As shown, the engineering status data source 20 generates a monotonically increasing status version based on changes in each construction stage, load, or boundary. In step S101, the monitoring window construction module 110 checks the engineering status version corresponding to all monitoring records in the current window. If the same window contains multiple versions that are not split according to the switching time, the exception handling module 210 rejects the window and re-executes the window division. If the engineering status version is rolled back or missing, the system 100 maintains the current valid parameter status and waits for the status data to be repaired.

[0181] like Figure 7As shown, if the same window is triggered multiple times due to task retries, process resumption, or duplicate messages, the parameter state storage and feedback module 200 constructs a unique task key based on the window identifier and the current reference parameter state version. If the same task key is already running or has been submitted, a new parameter update task will not be started again. If an existing task has failed, retry is only allowed if the failure reason has been resolved and the reference has not changed. Idempotency control ensures that the same window cannot be repeatedly updated and that multiple competing parameter states cannot be generated.

[0182] like Figure 2 and Figure 7 As shown, when the forward mechanics solution module 130 returns a non-convergence flag, the residual cannot be reduced to the solution tolerance, internal variables are in a non-numerical state, or the calculation process does not receive a response within a single task time limit, the exception handling module 210 will take corresponding actions according to the task type. For example, if the reference state fails, the system 100 will close the entire window. When the candidate state solution fails, the parameter incremental update module 140 reduces the update step size, increases damping, or restores the reference state. If a counterfactual state fails, the tested parameter unit is evaluated to achieve the purpose of not being evaluated, and is not directly included in the conflict parameter set.

[0183] like Figure 7 As shown, the single-task time limit is jointly determined by the current model's average solution time, model size, and total window computation time. For tasks without feedback, the exception handling module 210 suspends the corresponding computation process, releases temporary resources, and then retryes according to the backup solution configuration. If the retry still fails, other parameter units are evaluated. If the number of unevaluable parameter units is so large that the set of conflicting parameters is difficult to determine, then a global rollback is performed. In this way, even if a counterfactual task failure occurs, the current window will not be blocked indefinitely, nor will it be mistaken for evidence of parameter conflict.

[0184] like Figure 7 As shown, when the parameter group search approaches the window computation time limit, the system 100 stops expanding the combination size and checks whether a conflicting parameter group that has passed the double boundary conditions has been obtained. If a qualified group already exists, the qualified group with the smallest current size is used to form a mixed state; if no qualified group exists, the current reference parameter state is restored. The system does not save intermediate states that have not yet completed a full review, thus ensuring that the output after the computation timeout still has a clear safety boundary.

[0185] like Figure 2 and Figure 6As shown, the parameter incremental update module 140 checks the lower and upper bounds of the parameter after each trial update. If a slight out-of-bounds situation occurs, it can project the parameter to the nearest boundary and re-execute the forward calculation. The process can only continue after data consistency and independent physical consistency are verified. If the parameter remains outside the boundary and the update direction still points outside the boundary, the system 100 determines that the parameter is restricted in this round of updates, reduces the step size, or restores it to the reference value.

[0186] like Figure 6 As shown, for implementations employing plastic, creep, or damage constitutive relations, although parameter values ​​are within numerical upper and lower bounds, candidate states may lead to invalid constitutive states, such as internal states exceeding the model domain, damage variables violating irreversible evolution conditions, and plastic dissipation not meeting acceptable conditions. The dual consistency calculation module 150 converts these states into corresponding physical consistency components. Candidate states trigger conflict attribution. If all individual parameter rollbacks fail to recover, the exception handling module 210 enters parameter group rollback or overall rollback. That is, the validity of parameters is constrained by both numerical boundaries and constitutive states.

[0187] like Figure 5 and Figure 7 As shown, when a consistency conflict has been triggered, but all single-parameter counterfactual rollback states fail to restore physical consistency, the conflict parameter set determination module 180 does not output an empty set and continues to accept candidate states; instead, it initiates a parameter group search. The parameter group search prioritizes combining parameters that have a common effect on the same failed physical component, and then increases the combination size incrementally. If the maximum combination size is reached or the calculation time limit is exceeded without finding a suitable parameter group, the exception handling module 210 restores the current reference parameter state as a whole.

[0188] like Figure 6 As shown, when all updated parameters meet the conflict conditions, the selective rollback and verification module 190 restores all updated parameters to their reference values, and no new current valid parameter state is generated in this round. If only some parameters conflict, but the resulting mixed state still fails the complete verification, the selective rollback and verification module 190 first expands the conflict set and solves again; if it still fails after expansion, it rolls back completely. This process avoids the system skipping the final verification because some conflicting parameters have already been located.

[0189] like Figure 6 and Figure 7 As shown, the parameter status storage and feedback module 200 employs versioned writing and active index switching. When writing mixed parameter statuses, the inactive version is written first, followed by the residual, model version, project status version, and submission verification information. Finally, the active index is switched. If any one of these steps fails, the active index remains unchanged, and the exception handling module 210 deletes or isolates the incomplete version.

[0190] like Figure 7 As shown, when the electronic device is powered on again or the process restarts, the parameter status storage and feedback module 200 reads the most recent valid parameter status with a complete submission identifier and that has passed verification. If the record pointed to by the active index is missing or fails, it rolls back to the previous version; if no valid version exists, it reads the initial parameter status from the initial parameter and parameter boundary data source 40, and puts the system 100 into an initialization state, waiting to complete the forward solution of the reference state and the model version verification before accepting new windows. This avoids incomplete status taking effect due to power outages or write interruptions.

[0191] like Figure 7 As shown, the main function of the exception handling module 210 is to save the cause code and recovery conditions for different exceptions. Data communication exceptions are reset after complete data retransmission and version consistency; observation mapping exceptions are reset after updating the mapping configuration and performing a channel check; solution exceptions are reset after successful re-solution in the reference state; parameter out-of-bounds exceptions are reset after updating the step size adjustment or parameter recovery; and storage exceptions are reset after correct state write and read verification. Exception resets only clear the corresponding exception state and cannot accept previously failed candidate parameter states. All tasks executed after recovery restart from the most recently valid parameter state. This rule prevents the continued use of old failed intermediate values ​​after the exception is resolved.

[0192] like Figure 11 As shown, when system 100 is in normal operation, it continuously performs anomaly monitoring. If no anomaly is detected, it continues normal operation. When an anomaly is detected, the anomaly handling module 210 identifies the anomaly type and determines whether partial repair is possible. If partial repair is possible, the counterfactual rollback state construction module 170 constructs a rollback state, and the selective rollback and verification module 190 performs the rollback and re-solution. After verification, the parameter state storage and feedback module 200 maintains or submits the most recently valid parameter state and enters the next monitoring window. If partial repair is not possible or partial verification fails, the system enters a security degradation phase, freezing the current update, rejecting the current candidate parameter state, maintaining the most recently valid parameter state, and recording the cause of the anomaly. After entering the next monitoring window, the anomaly handling module 210 checks the recovery conditions. If the recovery conditions are met, it returns to normal operation; if the recovery conditions are still not met, it performs a full rollback and continues to maintain the most recently valid parameter state. This state machine ensures that anomaly resolution only restores processing capability and does not automatically accept candidate states that have not yet passed verification before the anomaly occurred.

[0193] like Figure 1As shown, this embodiment also provides a virtual device for online inversion of surrounding rock mechanical parameters. This virtual device can be composed of software modules of system 100 and uniformly scheduled by a single entity for calculating surrounding rock mechanical parameters inversion. Each module in the virtual device corresponds one-to-one with the aforementioned method steps, but they do not exist merely as abstract names; rather, they are implemented by executable programs, data structures, and state interfaces, respectively.

[0194] like Figure 1 As shown, the monitoring window construction module 110 can be implemented at the bottom layer as a combination of a data access program, a time indexing program, a window partitioning program, and a validity verification program. This layer connects the external monitoring data source 10 and the engineering status data source 20 through a data interface. After receiving the monitoring records and engineering status records, it performs time sorting, version matching, window splitting, missing ratio calculation, and quality identifier verification. Then, it outputs a valid monitoring window with a window identifier and engineering status version. In case of incompleteness or conflict with status version, it outputs an invalid window identifier to the exception handling module 210 and blocks the start of the parameter incremental update module 140. This layer is one of the most basic, making the window formation process repeatable and traceable.

[0195] like Figure 1 As shown, the observation mapping and response partitioning module 120 can implement a mapping configuration reading program, spatial location index, coordinate transformation function, and response purpose labeling program at the underlying level. It connects to the monitoring window construction module 110, the mechanical model configuration data source 30, the mechanical forward solution module 130, and the double consistency calculation module 150. The inputs are the valid monitoring window and model observation configuration; the actions are mapping the model output to monitored physical quantities, recording the response before parameter updates as the update purpose or verification purpose; and the outputs are the updated response set and the physical verification constraint set. When a mapping is missing, the observation mapping and response partitioning module 120 will remove the corresponding channel or trigger window degradation to prevent mismatched data from entering the residual calculation.

[0196] like Figure 1 As shown, the forward mechanics solution module 130 can be implemented at the underlying level as a numerical solution program encapsulation layer, a model state management program, and a solution receipt interface. It connects to the mechanics model configuration data source 30, the parameter incremental update module 140, the counterfactual rollback state construction module 170, the selective rollback and verification module 190, and the exception handling module 210. Its inputs are: the current engineering state, the complete parameter state, and the model configuration; its actions are to perform forward mechanics solution and generate a predicted response through observation mapping, and output the predicted response, internal state, and solution validity indicator. When the forward mechanics solution fails, it will return the failure node, residual state, and retry indicator, and the exception handling module 210 will determine whether to shrink the step size, use a backup configuration, or roll back the state.

[0197] like Figure 1 As shown, the parameter incremental update module 140 can be implemented at the underlying level as an objective function program, a parameter vector management program, a boundary projection program, an iterator, and a step size controller. It receives the current reference parameter state, the parameter update response set, and the predicted response; calculates and generates the target construction, update direction calculation, parameter boundary projection, and candidate states; and outputs the current candidate parameter state and the updated parameter indices for this round. If the solution goes out of bounds or the forward direction is invalid, the candidate state is not submitted; the update step size is adjusted, or the reference state is restored. This implementation translates the formed candidate parameter states into executable data structure transformations.

[0198] like Figure 1 As shown, the dual consistency calculation module 150 can be implemented at the underlying level as a data residual calculation program, a physical residual component program, an error scale reading program, and a residual change comparison program. This module reads the model outputs from the reference state, candidate state, and counterfactual state respectively, calculates the data residuals of the parameter update response set and the independent physical residuals of the physical verification constraint set, and outputs each residual component, the allowable boundary state, and the direction of change. The dual consistency calculation module 150 does not weight all residuals into a single score by default, but retains the physical components that must pass, thereby preventing severe physical mismatches from being offset by improvements from other components.

[0199] like Figure 1 As shown, the consistency conflict triggering module 160 can be implemented as a state machine and a condition comparison program at the underlying level. This module receives the post-update residuals, data fluctuation threshold, physical fluctuation threshold, and physical allowable boundary from the double consistency calculation module 150, and performs four state classifications. When effective improvement is achieved and at least one physical consistency deteriorates and goes out of bounds, this module sends an attribution task to the counterfactual rollback state construction module 170. If the above requirements are not met, the processing flow is transferred to candidate state review, parameter updates are recalculated, or the entire process rolls back to the previous branch.

[0200] like Figure 1 As shown, the counterfactual fallback state construction module 170 can be a parameter index queue, a state replicator, a reference value replacer, and a task scheduler at the underlying level. It receives the current reference parameter state, the current candidate parameter state, and the updated parameter list for this round. For one or a group of parameters, it generates a counterfactual state and binds each state with the engineering state version, model version, and task identifier before inputting it into the mechanics forward solution module 130. If a task times out or fails to solve, it is marked as unevaluable in the counterfactual fallback state construction module 170, and other tasks are scheduled accordingly.

[0201] like Figure 1As shown, the conflict parameter set determination module 180 can be implemented at the underlying level as a condition filtering program, a parameter group search program, and an ordered selection program. It reads the double consistency results of each counterfactual state, first checks the physical recovery conditions, then checks the data preservation conditions; if a single parameter is insufficient, it filters according to the size of each parameter; if multiple parameter groups are satisfied, it selects the parameter set sequentially according to the number of parameters, physical recovery status, and data residuals, avoiding the use of opaque comprehensive contribution scoring.

[0202] like Figure 1 As shown, the selective rollback and verification module 190 can be implemented at the underlying level as a mixed-state synthesis program, a complete verification program, and an expanded rollback control program. It restores conflicting parameters to reference values, uses non-conflicting updated parameters as candidate values, and obtains a mixed state for re-solving using the mechanics forward solution module 130. When the verification passes, a submission request is sent to the parameter state storage and feedback module 200; when the verification fails, the expanded conflict set or request exception handling module 210 is used to perform a global rollback operation, transforming the parameter attribution result into a state change.

[0203] like Figure 1 As shown, the parameter status storage and feedback module 200 can be implemented at the underlying level as a versioned parameter database, a status serialization program, an integrity verification program, and an active version index. This module receives mixed states that have passed review. It saves inactive versions, performs integrity checks on them, and switches the active index; the next window only reads the active version. If the write operation fails, the active index is not changed, and a failure receipt is returned to the exception handling module 210. This ensures that only complete and verified states take effect.

[0204] like Figure 1 As shown, the exception handling module 210 can be implemented at the underlying level as an exception cause code table, a state transition program, a retry counter, and a degradation action scheduler. This module receives exception identifiers from each core module and performs actions such as window discarding, data retransmission, step size shrinking, alternative solutions, parameter group search, overall rollback, state locking, or initialization recovery. Each exception has a corresponding recovery condition. After recovery, the system restarts from the most recently valid parameter state without continuing to use the failed candidate state.

[0205] like Figure 8As shown, this embodiment also provides an electronic device 300. The electronic device 300 includes a processor 310, a memory 320, a communication bus 330, a network interface 340, a non-volatile memory unit 350, and a volatile memory unit 360. The processor 310, memory 320, network interface 340, non-volatile memory unit 350, and volatile memory unit 360 exchange data via the communication bus 330. The electronic device 300 can be a server, workstation, edge computing node, or computing node controlled by a unified task; its specific shape and mechanical structure are not limited to this embodiment.

[0206] Figure 8 In this context, memory 320 represents the upper-level storage resource or storage management unit of electronic device 300. Non-volatile storage unit 350 and volatile storage unit 360 are used to further distinguish between persistent storage area and temporary working area. In some embodiments, non-volatile storage unit 350 and volatile storage unit 360 can constitute different physical media or logical regions of memory 320; in other embodiments, they can also be independent storage units connected in parallel with memory 320 via communication bus 330. Regardless of the organization method, submitted parameter states and unsubmitted intermediate states should be stored in distinguishable storage areas.

[0207] like Figure 8 As shown, network interface 340 is used to receive input from external monitoring data source 10, engineering status data source 20, and mechanical model configuration data source 30, and write the received results to the current window workspace in volatile storage unit 360. Network interface 340 can support wired network, wireless network, or local data exchange interface; when electronic device 300 reads data from local database, network interface 340 can be replaced by local input / output interface. Regardless of the interface used, the input must include window identifier, data version, and integrity status.

[0208] like Figure 8 As shown, non-volatile storage unit 350 is used to permanently store computer programs, mechanical model configurations, parameter boundaries, parameter state versions, and exception logs. Non-volatile storage unit 350 can be solid-state memory, flash memory, read-only memory, erasable non-volatile memory, or magnetic storage media. Volatile storage unit 360 is used to load current window data, model intermediate states, candidate parameter states, counterfactual task queues, and mixed parameter states; it can be random access memory. By storing active parameter states in non-volatile storage unit 350 and temporary states in volatile storage unit 360, committed and uncommitted states can be distinguished after computation is interrupted.

[0209] like Figure 8As shown, processor 310 may include a general-purpose central processing unit, a graphics processing unit, a vector computing unit, or a heterogeneous computing unit. Processor 310 reads the computer program from non-volatile memory 350 and loads it into volatile memory 360, and executes window construction, parameter updating, double consistency calculation, counterfactual rollback state construction, repeated mechanics solving, conflict parameter determination, selective rollback, and state feedback according to steps S101 to S106. For multiple counterfactual rollback states, processor 310 can execute them sequentially or in parallel while maintaining the same project state and model version; parallel execution only changes the computation schedule and does not change the requirement to re-solve each state separately.

[0210] like Figure 8 As shown, when processor 310 executes parameter state commit, it first writes the mixed parameter state and corresponding log to the inactive version area of ​​non-volatile storage unit 350, then reads the write result for integrity verification, and finally updates the active version index. If electronic device 300 loses power before the index update, it will still read the original active version after power-on; if the index has been updated but the new version verification fails, it will revert to the previous complete version. This hardware and program coordination ensures that a power outage will not cause the half-written state to become the current valid parameter state.

[0211] like Figure 8 As shown, in an alternative implementation of the electronic device, the forward mechanics solution and counterfactual tasks are executed by one or more computing nodes, while parameter state storage and feedback are handled by a unified state node. All computing nodes read the tasks from the same reference state version and engineering state version, and return the solution results with task identifiers; all state nodes only accept the results of the same version. Even with distributed computing, conflict triggering, task generation, conflict parameter determination, and state submission are still completed by the same logical control entity.

[0212] like Figure 8 As shown, this embodiment also provides a computer-readable storage medium storing a computer program. The storage medium can be flash memory, solid-state storage, read-only memory, erasable memory, magnetic storage, or optical storage in the non-volatile storage unit 350, or other tangible storage media readable by the electronic device 300. When the computer program is executed by the processor 310, the electronic device 300 performs steps S101 to S106.

[0213] like Figure 2 and Figure 8As shown, a computer program includes at least window processing instructions, model solving instructions, parameter update instructions, double consistency calculation instructions, conflict triggering instructions, counterfactual state construction instructions, conflict parameter determination instructions, selective rollback instructions, state verification instructions, and state submission instructions. These instructions exchange data through a unified window identifier, project state version, and parameter state version to prevent the cross-use of intermediate results from different windows or different model versions.

[0214] like Figure 8 As shown, the computer program product may include the aforementioned computer program and its associated model configuration description, parameter dictionary, observation mapping configuration, and state record format. After these computer programs are loaded into the electronic device 300, the processor 310 reads the current input and forms the current candidate parameter state. If it finds that the data consistency and independent physical consistency are changing inversely, the program constructs a counterfactual rollback task to perform selective rollback. After verification, the current valid parameter state is written into the non-volatile storage unit 350, so that the computer program product is no longer just an abstract mathematical comparison, but enables the electronic device 300 to complete the solution of the surrounding rock mechanics model and the transformation of parameter states.

[0215] like Figure 8 As shown, during program upgrades or model configuration updates, the computer program product saves the program and model versions for each version. If the new program cannot read the old parameter status format, the parameter status storage and feedback module 200 will first perform status migration and verification. If the process fails, the electronic device 300 will continue to use the original program version and the original active parameter status. This prevents the loss or incorrect parsing of historical valid parameter statuses due to software updates.

[0216] 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 method for real-time inversion of surrounding rock mechanical parameters, characterized in that: include, Acquire monitoring responses, engineering status, and reference parameter status; generate candidate parameter status based on parameter update response set. Based on the response set and the physical verification constraint set generated independently of the candidate states, determine the data residuals and physical residuals between the two states; When the data residual of the candidate parameter state decreases and the physical residual increases beyond the limit, select the parameter unit from the updated parameters and restore it to the reference value. Keep the candidate values ​​of the remaining updated parameters, construct the counterfactual backoff state and perform the forward mechanical calculation. Parameter units whose physical residuals are recovered and whose data residuals are not out of bounds are included in the conflict parameter set. The parameters in this set are restored to the reference values, and the remaining updated values ​​are retained to form a mixed parameter state. When a conflicting parameter set is formed, submit the mixed parameter state; when no conflicting parameter set is formed, the data residual of the candidate parameter state is not greater than the data residual of the reference parameter state, and none of the physical residual components have crossed the corresponding physical allowable boundary, submit the candidate parameter state. When no conflicting parameter set is formed and the data residual of the candidate parameter state is greater than the data residual of the reference parameter state, or when any physical residual component crosses the corresponding physical allowable boundary, the reference parameter state is maintained; if the submitted state passes the review of both types of residuals, it is considered a valid parameter state, and if it fails, the reference parameter state is maintained.

2. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: Before generating candidate parameter states, the monitoring response is divided into a parameter update response set and a physical verification constraint set. The parameter update response set participates in the construction of the parameter update target, while the physical verification constraint set does not participate in the construction of the target. The partitioning results, window identifiers, and project status versions are saved together and remain unchanged during residual comparisons of candidate parameter status, counterfactual backoff status, and mixed parameter status.

3. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: Calculate the decrease in data residuals of candidate parameter states relative to reference parameter states and the increase in each physical residual component; when the decrease exceeds the data fluctuation threshold, at least one increase exceeds the corresponding physical fluctuation threshold, and the corresponding physical residual component crosses the physical allowable boundary, construct a counterfactual backoff state; When the data residual of the candidate parameter state is not greater than the data residual of the reference parameter state and none of the physical residual components have crossed the corresponding boundary, the candidate parameter state is submitted; otherwise, the reference parameter state is maintained.

4. The real-time inversion method for surrounding rock mechanical parameters according to claim 3, characterized in that: The status of reference parameters, candidate parameters, and each counterfactual rollback status are all associated with the same monitoring window identifier, engineering status version, model version, and reference parameter status version; For path-dependent mechanical models, each state starts from the internal state snapshot at the same window starting point and performs forward mechanical calculations according to the same load increment sequence; when versions are inconsistent or internal state snapshots are missing, the corresponding residuals are not compared and the reference parameter state is maintained.

5. The real-time inversion method for surrounding rock mechanical parameters according to claim 4, characterized in that: Each parameter unit contains only one updated parameter. When constructing the corresponding counterfactual fallback state, only that parameter is restored to the corresponding value in the reference parameter state. Other updated parameters retain the corresponding values ​​in the candidate parameter state, and unupdated parameters retain the corresponding values ​​in the reference parameter state. After re-performing the forward mechanical calculation, if the physical residual recovers to within the physical allowable boundary and the data residual does not exceed the data allowable boundary, this parameter will be included in the conflict parameter set; otherwise, it will not be included.

6. The real-time inversion method for surrounding rock mechanical parameters according to claim 5, characterized in that: When multiple single-parameter counterfactual fallback states all meet the conditions for inclusion in the conflict parameter set, the parameter that maximizes the reduction in the physical residual that has deteriorated is first selected as the reference value. The restored parameter state is used as the temporary mixed parameter state. Then, the single-parameter counterfactual fallback states are reconstructed for the remaining updated parameters in sequence. When there is no single parameter that meets the conditions, a parameter group counterfactual fallback state containing at least two updated parameters is constructed.

7. The real-time inversion method for surrounding rock mechanical parameters according to claim 6, characterized in that: When none of the single-parameter counterfactual rollback states simultaneously meet the physical residual recovery condition and the data residual preservation condition, the counterfactual rollback states of the parameter group are constructed sequentially, starting from two parameters and increasing level by level according to the number of parameters contained in the parameter group, and the forward mechanical calculation is re-executed. When a parameter set that satisfies both conditions is found, the qualified parameter set with the fewest parameters is selected as the conflict parameter set; if no qualified parameter set is found after reaching the maximum combination size or the calculation time limit, the reference parameter status is maintained.

8. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: The physical verification constraint set includes at least one of the following: reserved monitoring response, state continuity constraint within adjacent monitoring windows, constitutive state acceptability constraint, boundary response not involved in parameter update, and output of independent reference mechanical model. Each constraint forms a physical residual component and corresponds to its own physical allowable boundary. If any required component crosses the corresponding boundary, the physical residual is deemed to have failed. If all required components are within the corresponding boundary, the physical residual is deemed to have passed.

9. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: When data is missing, quality identifiers are invalid, or observation mappings fail, remove the corresponding monitoring channel and re-examine the remaining monitoring responses. When the remaining monitoring response can simultaneously form a parameter update response set and a physical verification constraint set, candidate parameter states are generated based on the remaining monitoring response. If only a parameter update response set can be formed or if no parameter update response set can be formed, no new valid parameter state is submitted and the reference parameter state is maintained.

10. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: A status record is generated for the valid parameter status. The status record includes the window identifier, project status version, model version, reference parameter status version, updated parameter list, conflicting parameter set, final parameter value, data residual, each physical residual, solution validity identifier, and submission time. The next monitoring window only uses the valid parameter status with complete status records as the reference parameter status and redetermines the conflicting parameter set.

11. The real-time inversion method for surrounding rock mechanical parameters according to claim 10, characterized in that: For each counterfactual rollback state, a counterfactual task record is generated. The counterfactual task record includes the task identifier, the index of the parameter being tested, the window identifier, the project state version, the model version, the reference parameter state version, and the candidate parameter state version. The results of the forward mechanical calculations carry the corresponding task identifier and the solution validity identifier. If the versions are consistent and the solution is valid, the result is accepted. If the versions are inconsistent, the result is discarded and the task is regenerated according to the consistent version. If the re-solution is still invalid, the reference parameter state is maintained.

12. The real-time inversion method for surrounding rock mechanical parameters according to claim 10, characterized in that: First, write the verified mixed parameter status and its status record to the inactive version area. Then, verify the integrity of the status record and the consistency of the window identifier, project status version, model version and reference parameter status version. If the verification is successful, switch the active status index to the inactive version area to make it a valid parameter status. If the writing or verification fails, the active status index will continue to point to the original valid parameter status.

13. The real-time inversion method for surrounding rock mechanical parameters according to claim 1, characterized in that: Each batch of monitoring data is associated with a window identifier, data version, number of declared records, and integrity verification information. A data receipt is generated after receiving a batch of monitoring data. When a complete batch is obtained within the receiving time limit and the declared number of records matches the actual number of records, candidate parameter status generation is initiated. If a complete batch is not obtained, the number of records is inconsistent, or the integrity check fails, the batch will be marked as unavailable, the generation of candidate parameter status will be prohibited, and the reference parameter status will be maintained.

14. A real-time inversion system for surrounding rock mechanical parameters, characterized in that, include: The parameter incremental update module receives monitoring responses, project status, and reference parameter status, and generates candidate parameter status based on the parameter update response set. The dual consistency calculation module determines the data residuals and physical residuals between the reference parameter state and the candidate parameter state based on the parameter update response set and the physical verification constraint set that did not participate in the generation of candidate parameter states. The consistency conflict triggering module triggers the counterfactual backoff state construction module when the data residual of the candidate parameter state decreases and the physical residual increases and exceeds the physical allowable boundary. The counterfactual rollback state construction module selects parameter units from the updated parameters, restores the parameter units to the reference values, and keeps the remaining updated parameters as candidate values ​​to form a counterfactual rollback state. The forward mechanics solution module performs forward mechanics calculations on counterfactual backoff states. The conflict parameter set determination module includes parameter units whose physical residuals are restored to within the physical allowable boundary and whose data residuals do not exceed the data allowable boundary into the conflict parameter set; The selective rollback and verification module restores the parameters in the conflict parameter set to the reference value and retains the remaining updated values ​​to form a mixed parameter state. When a conflict parameter set is formed, the mixed parameter state is set to the commit state. When the trigger condition is not met, the candidate parameter state is set to the commit state. When the trigger condition is met but a conflict parameter set is not formed, the reference parameter state is set to the commit state, and the data residual and physical residual verification is performed on the commit state. The parameter status storage and feedback module saves the submission status as a valid parameter status when the review is passed, and retains the reference parameter status when the review fails.

15. An electronic device, characterized in that, It includes a processor, a memory, a network interface, and a communication bus. The processor, memory, and network interface are connected through the communication bus. The network interface receives monitoring responses, engineering status, and mechanical model configurations. The memory stores reference parameter status and computer programs. When the computer program is executed by the processor, the electronic device implements the real-time inversion method for surrounding rock mechanical parameters as described in claim 1.

16. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the real-time inversion method for surrounding rock mechanical parameters as described in claim 1.

Images