A method for intelligent optimization of near existing line non-uniform deep foundation fat trench backfill

Abstract

The application discloses a kind of near existing line unequal deep foundation fat groove backfill intelligent optimization method, belong to underground engineering construction safety control technical field.The technical scheme is: obtaining super large cover body foundation, fat groove and existing line spatial relationship data, and dividing backfill control subarea and demarcating sensitive grade;Collect construction state, foundation, compaction machine and existing line monitoring data, and construct the backfill influence characteristic set containing foundation burial depth difference, stiffness mutation, compaction disturbance, spatial influence and monitoring lag response;Using backfill influence prediction model obtains settlement, displacement, differential settlement and compaction degree prediction result;Generate candidate working condition, calculate risk index, after risk determination and hard constraint screening, output target backfill scheme, and according to measured deviation dynamically correct subsequent scheme.The application can quantitatively evaluate backfill working condition before construction, reduce existing line deformation risk and improve backfill scheme adjustment efficiency.

Description

Technical Field

[0001] This invention relates to the field of underground engineering construction safety control technology, and in particular to an intelligent optimization method for backfilling trenches of unequal depth foundations near existing power lines. Background Technology

[0002] In projects such as subway depots, large industrial plants, and transportation hubs, it is becoming increasingly common to see super-large cover structures located close to existing operating lines. These projects typically feature large plan dimensions, significant differences in foundation depth, narrow trench space, and proximity to existing operating lines, making safety control during backfilling construction far more challenging than in conventional projects.

[0003] Existing trench backfilling technologies mainly employ graded sand and gravel, fluidized solidified soil, and other backfill materials, combined with construction techniques such as layered spreading, layered compaction, and small-scale tamping. Backfill quality is ensured by controlling layer thickness, number of compaction passes, and moisture content. However, these methods rely heavily on construction experience, specification limits, or test section results. They cannot quantitatively compare the impact of different backfilling times, sequences, layer thicknesses, backfill materials, and compaction equipment parameter combinations on the safety of existing railway lines, making it difficult to determine the most advantageous backfilling scheme for protecting existing lines among various candidate conditions.

[0004] Existing protection technologies for existing railway lines mainly rely on settlement monitoring, horizontal displacement monitoring, track deformation monitoring, and alarm thresholds. When monitoring data exceeds the alarm threshold, measures such as slowing down the construction rate, adjusting the construction sequence, or suspending construction are taken. However, these methods are passive control measures implemented after construction. The monitoring data reflects deformations that have already occurred and cannot be used for prediction, evaluation, or proactive screening before backfilling. This results in a lag in adjustment response and makes it difficult to avoid the accumulation of risks during the construction process.

[0005] Existing machine learning settlement prediction methods, such as random forests, gradient boosting trees, and LSTM, have been applied to underground engineering settlement prediction, using historical monitoring data to train models to predict future settlement trends. However, existing methods often focus on single settlement indicators or overall trend prediction, typically treating the construction process as an already occurred external background quantity. They fail to uniformly construct a backfill influence feature set that can be updated with candidate working conditions, taking into account factors such as the zoning attributes of foundations of varying depths, abrupt changes in stiffness between shallow and deep sections, physical parameters of compaction equipment, backfill materials, zoning backfilling sequence, construction intervals, existing railway protection levels, and monitoring lag responses. Therefore, existing prediction results struggle to reflect the differences in the impact of different combinations of candidate backfilling parameters on existing railway deformation and backfill compaction, and are also difficult to directly use for pre-construction condition selection.

[0006] Existing risk assessment or optimization methods often employ single-item limit control, experience-based scoring, or post-event alarm adjustments, such as controlling settlement to not exceed a certain limit or horizontal displacement to not exceed a certain limit. However, these methods typically fail to incorporate predicted response results, compaction disturbance evaluation values, backfill compaction degree, and backfill height differences between adjacent zones into the same candidate condition screening chain, nor do they distinguish the boundaries of action between the initial screening of comprehensive risks and the verification of engineering hard constraints. When the comprehensive risk of a candidate condition is low but the individual engineering control conditions are not met, existing methods are prone to misselection or require repeated manual verification.

[0007] How to solve the above-mentioned technical problems is the challenge facing this invention. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent optimization method for backfilling of uneven-depth foundation trenches near existing railway lines. This method addresses the technical problem that empirical methods and single settlement prediction methods are insufficient for quantitatively comparing backfilling time, sequence, layer thickness, materials, and compaction equipment parameter combinations before construction, and are also insufficient to simultaneously meet the requirements of existing railway line protection, backfill compaction degree, and adjacent zone elevation difference control. This invention transforms the spatial relationship between the large-scale uneven-depth foundation and the adjacent existing railway line into an engineering feature that can be incorporated into the model. It also transforms the backfilling material, backfilling sequence, layer thickness, compaction equipment parameters, construction interval time, and adjacent zone backfill elevation difference in candidate backfilling conditions into updatable construction condition parameter features. Through a backfilling impact prediction model, multi-index predicted responses corresponding to each candidate condition are obtained. Then, a comprehensive risk screening is performed using the backfilling condition risk index, and existing railway line protection and backfilling quality control conditions are checked item by item through hard constraint screening. Finally, the subsequent conditions, model parameters, or training sample library are dynamically corrected based on the actual measured deviations on site. This shifts construction decision-making from experience-based adjustments following post-construction monitoring to pre-construction prediction, evaluation of candidate operating conditions, and screening of hard constraints, thereby reducing the likelihood of risk accumulation and shortening the adjustment response lag time.

[0009] The inventive concept of this invention is as follows: This invention provides an intelligent optimization method for backfilling of super-large covered structures with unequal depths and thicknesses near existing railway lines. It is mainly applied to super-large covered structure construction projects near existing operational subway lines, railway lines, or other underground structures. In these scenarios, this technology achieves quantitative prediction and evaluation and proactive optimization of backfilling construction parameters by constructing a backfilling impact feature set that can be updated with candidate working conditions, a backfilling impact prediction model, a mapping relationship between candidate backfilling working conditions and predicted response results, calculation of backfilling working condition risk index, hard constraint screening, and a dynamic correction mechanism. In the actual construction environment of backfilling super-large covered structures with unequal depths and thicknesses, this method can dynamically adjust the backfilling layers, layer thicknesses, backfilling materials, zoned backfilling sequence, compaction equipment parameters, and construction interval time based on the basic attributes, sensitivity levels, construction stage correction coefficients, and on-site monitoring deviations of each backfilling control zone, thereby achieving coordinated control of existing line protection and backfilling quality.

[0010] To achieve the aforementioned objectives, the present invention employs the following technical solution: an intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines. This method is based on a super-large cover structure with a planar projected area of ​​not less than 10,000 square meters. Unequal depth foundations refer to foundations within the same super-large cover project where the difference in foundation bottom elevation between at least two foundation zones is not less than 2.0 meters. "Near existing power lines" means that the minimum horizontal net distance from the edge of the trench in at least one backfill control zone of the super-large cover project to the outer edge of the existing power line structure is not greater than 50 meters. The method includes the following steps:

[0011] S1. Obtain spatial relationship data of the super-large cover foundation, trench, and adjacent existing line;

[0012] S2. Based on spatial relationship data, divide the backfill area of ​​the fertilizer trench into at least two backfill control zones, determine the basic attributes of each backfill control zone, and mark the sensitivity level and construction stage correction coefficient.

[0013] S3. Collect construction status data, foundation parameters, compaction equipment parameters and existing line monitoring data. Perform outlier identification, missing value completion, timestamp alignment, unit unification and normalization on the collected data, and establish the correlation between the above data and backfill control zones and construction time.

[0014] S4. Based on spatial relationship data, basic attributes, construction status data, foundation parameters, compaction equipment parameters and existing line monitoring data, construct a backfill influence feature set including foundation depth difference characteristics, deep-shallow transition stiffness abrupt change characteristics, compaction disturbance characteristics, existing line spatial influence characteristics, construction condition parameter characteristics and monitoring lag response characteristics.

[0015] S5. Input the backfill impact feature set into the trained backfill impact prediction model to obtain the prediction response results. The prediction response results include the predicted values ​​and corresponding prediction intervals of existing line settlement, existing line settlement rate, existing line horizontal displacement, deep and shallow foundation differential settlement and backfill compaction degree.

[0016] S6. Based on the basic attributes, sensitivity level, construction stage correction coefficient, and preset construction parameter range of each backfill control zone, generate a candidate backfill condition parameter set. The candidate backfill condition parameter set includes at least two types of variable condition parameters from backfill material parameters, layer thickness, compaction equipment type and parameters, backfill time information, zone backfill sequence, construction interval time, and backfill height difference between adjacent zones. Substitute the variable condition parameters from the candidate backfill condition parameter set into the backfill influence feature set of the corresponding backfill control zone, update the construction condition parameter features, compaction disturbance features, and monitoring lag response features, and form the backfill influence feature set corresponding to each candidate backfill condition. Input the updated backfill influence feature sets into the backfill influence prediction model to obtain the predicted response results of each candidate backfill condition, and establish a mapping relationship between the candidate backfill conditions and the predicted response results.

[0017] S7. Based on the predicted response results in the mapping relationship between candidate backfill conditions and predicted response results, and the construction parameters in the candidate backfill condition parameter set, calculate the backfill condition risk index and make a risk assessment.

[0018] S8. Perform hard constraint screening on candidate backfill conditions that have passed risk assessment; when there are candidate backfill conditions that have passed both risk assessment and hard constraint screening, sort them in ascending order of risk index and output the sorted target backfill schemes, taking the first-ranked scheme as the execution scheme; when there are no candidate backfill conditions that have passed both risk assessment and hard constraint screening, reduce the layer thickness, reduce the excitation force, reduce the excitation frequency, reduce the number of compaction passes or tamping passes, use low-disturbance compaction equipment, extend the construction interval time and / or reduce the backfill height difference between adjacent zones according to the preset conservative adjustment rules, and regenerate candidate backfills. The working condition parameter set is returned and S6 to S8 are executed. If no candidate backfill working condition that has passed both risk assessment and hard constraint screening is found after M consecutive regenerations, a pause backfilling and manual review prompt is output, where M is an integer from 2 to 5. The pause backfilling and manual review prompt includes the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations. The target backfilling scheme includes the target backfilling control zone, backfilling layer, backfilling time information, zone backfilling order, layer thickness, backfilling material, compaction equipment parameters, construction interval time, and control requirements for the backfilling height difference between adjacent zones.

[0019] S9. Based on the deviation between the field measurement results and the predicted response results, dynamically correct the construction parameters, model parameters and / or training sample library of the subsequent backfill layer and / or the subsequent backfill control zone, and update the subsequent scheme based on the correction results; after completing the dynamic correction, based on the corrected construction parameters, model parameters and / or training sample library, re-execute S4 to S8 to update the mapping relationship between the candidate backfill conditions and the predicted response results, the risk assessment results and the target backfill scheme of the subsequent backfill layer and / or the subsequent backfill control zone.

[0020] Further, in S2, the backfill control zones include at least two of the following: a sensitive zone for deep foundation trenches near the existing line, a sensitive zone for transitional deep and shallow foundation trenches, a general control zone for shallow foundation trenches, and a general backfill zone far from the existing line; the sensitivity level is determined based on the minimum horizontal clearance between each backfill control zone and the outer edge of the existing line structure, the foundation depth, and the protection level of the existing line; the construction stage correction coefficient is determined based on the sensitivity level and the current construction stage, which includes the initial backfilling stage, the middle backfilling stage, and the later backfilling stage; wherein, when the minimum horizontal clearance is not greater than 20 meters, the foundation... When the burial depth is not less than 8 meters and the existing line protection level is strict protection, the sensitivity level is high sensitivity, and the correction factor during the construction stage is 1.2 to 1.5; when the minimum horizontal clearance is greater than 20 meters and not greater than 50 meters, or when the backfill control zone is located in the deep-shallow transition range and there is a difference in foundation burial depth, the sensitivity level is sensitive, and the correction factor during the construction stage is 1.0 to 1.2; when the minimum horizontal clearance is greater than 50 meters, or when the minimum horizontal clearance from the shallow foundation area to the outer edge of the existing line structure is greater than 3 times the foundation burial depth, the sensitivity level is general control, and the correction factor during the construction stage is 0.9 to 1.0.

[0021] Furthermore, in S3, the construction status data includes backfilling time information, backfilling sequence, layer thickness, backfilling material, and compaction equipment type; the foundation parameters include foundation soil type, foundation bearing capacity characteristic value, and groundwater level; the compaction equipment parameters include at least one of excitation force, excitation frequency, equipment travel speed, equipment self-weight, and number of compaction passes and number of tamping passes; the existing line monitoring data includes existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, and track deformation.

[0022] Further, in S4, the backfill influence feature set includes foundation depth difference features, abrupt change in stiffness during the transition between shallow and deep sections, compaction disturbance features, existing line spatial influence features, construction condition parameter features, and monitoring lag response features; the existing line spatial influence features are composed of the normalized value of the minimum horizontal clearance from the backfill control zone to the outer edge of the existing line structure, the normalized value of the vertical clearance, the sensitivity level code value, the existing line protection level code value, and the construction stage correction coefficient; the construction condition parameter features are composed of at least two of the following: backfill material code value, material compression modulus or strength grade, backfill planned start time code value, construction interval time, zone backfill sequence code value, and compaction equipment type code value.

[0023] Furthermore, in S4, the monitoring lag response feature is not directly expanded to include all the original monitoring values ​​for each monitoring cycle, but rather a statistical feature calculated based on the previous N monitoring cycles, where N is an integer not less than 2. The statistical features include the existing line settlement increment, settlement rate change, horizontal displacement change, and differential settlement increment between shallow and deep foundations. When the current backfill layer is the first backfill and there is no historical monitoring data, the first backfill initialization strategy is adopted, the statistical feature is set to 0, and the unfavorable direction boundary value of the prediction interval is used as the corresponding prediction value in the risk index calculation or hard constraint screening.

[0024] Furthermore, in step S4, the compaction disturbance characteristics are quantitatively characterized by compaction disturbance evaluation values, the calculation formula of which is:

[0025] ;

[0026] In the formula, ,and to All are not less than 0; To provide vibration force for compaction equipment; This is a reference value for the excitation force; The excitation frequency; This is the reference value for the excitation frequency; This refers to the number of compaction passes or tamping passes; This is a reference value for the number of passes; For layer thickness; This is a reference value for the layer thickness; The speed at which the machine travels; Correct the reference value for speed; The minimum horizontal clearance between the backfill control zone and the outer edge of the existing line structure; This is the reference value for net distance correction; where , , , , and All are positive values; The zoning construction correction factor is determined jointly by the sensitivity level of the backfill control zone and the current construction stage. This is a correction factor for the protection level of existing lines; This is a dimensionless evaluation value for compaction disturbance. To and Dimensionless compaction disturbance limits using the same normalized benchmark. It is determined based on the existing line protection level, minimum horizontal clearance, type of compaction equipment, construction stage, and statistical results of similar projects.

[0027] Further, in S5, the backfill impact prediction model includes a feature input unit, a numerical encoding and normalization unit, a model training unit, a multi-output prediction unit, and a prediction interval generation unit. The backfill impact prediction model is trained using the backfill impact feature set of historical construction samples as input and the corresponding historical monitoring responses and backfill compaction test values ​​as outputs. The multi-output prediction unit uses a multi-output regression model, or sets five single-output regression models corresponding to existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between shallow and deep foundations, and backfill compaction degree, respectively. When the backfill material, compaction equipment type, sensitivity level, construction stage, and zonal backfilling sequence in the input are discrete quantities, they are encoded using one-hot encoding. The sequence number or target code is converted into numerical features, and continuous quantities use the same normalization benchmark as the training samples. The training process includes forming an input matrix X and an output matrix Y based on the sample numbers, and dividing the samples into training, validation, and test sets according to construction time sequence or partitioning, or using K-fold cross-validation. When using a random forest regression model, the number of trees is 100 to 500, the maximum depth is 3 to 20, and the minimum number of samples per leaf node is 1 to 10. When using a gradient boosting tree regression model, the number of trees is 100 to 800, the maximum depth is 2 to 10, the learning rate is 0.01 to 0.30, and the subsample ratio is 0.5 to 1.0. Model validation metrics include mean absolute error, root mean square error, and coefficient of determination. Coefficient of determination on the test set If the value is below 0.85, adjust the feature combination or increase the number of training samples and retrain.

[0028] Furthermore, in step S7, the calculation of the backfilling condition risk index includes seven individual risk ratios, the formulas of which are as follows:

[0029]

[0030] In the formula, The settlement risk ratio of the existing line; Risk ratio of existing line settlement rate; The risk ratio of horizontal displacement of existing lines; The risk ratio of differential settlement between shallow and deep foundations; Risk ratio for backfill compaction; To solidify the disturbance risk ratio; Risk ratio of backfill elevation difference between adjacent zones; These are the settlement control values ​​for existing railway lines; This is the control value for the settlement rate of the existing railway line; These are the horizontal displacement control values ​​for existing railway lines; This represents the control value for differential settlement between shallow and deep foundations; This refers to the required compaction degree for backfill. Set the elevation difference limit for backfilling adjacent zones; The elevation difference between adjacent zones is backfilled. This is the predicted settlement value for the existing railway line; This is the predicted settlement rate of the existing railway line. The predicted horizontal displacement of the existing line; This represents the predicted value for differential settlement between shallow and deep foundations. This is the predicted value for backfill compaction. , , , , C max , and All are positive values; To preset the upper limit of compaction and Not less than ;when Less than the preset lower limit correction value At that time, with As The compaction degree value used in the calculation It is a constant greater than 0; when Greater than At that time, with As The compaction degree is calculated, and over-compaction is checked in the hard constraint screening. For settlement, settlement rate, horizontal displacement, and differential settlement, if the monitoring control values ​​are given as absolute limits, the corresponding predicted values ​​are substituted into the corresponding individual risk ratios according to the absolute value or the value in the unfavorable direction.

[0031] Furthermore, in step S7, the formula for calculating the backfilling condition risk index R is:

[0032]

[0033]

[0034] In the formula, The maximum single-item risk ratio; The multi-indicator comprehensive correction coefficient ranges from 0 to 1.0; to They are respectively to The corresponding weighting coefficients, to All are not less than 0, and Weighting coefficient to The weights are determined based on the existing line protection level, the sensitivity of monitoring indicators, the contribution rate of historical deformation, the analytic hierarchy process (AHP) or the entropy weight method. When prior information is lacking, each weight is equal.

[0035] The risk assessment rule is: when Greater than or equal to the preset risk threshold When the corresponding candidate backfilling condition is identified as a high-risk prohibited condition and eliminated; when Less than When this happens, the corresponding candidate backfilling condition will be considered as a candidate condition for passing the risk assessment. according to Sure, The risk reserve coefficient is set between 0.10 and 0.50, and is selected within this range based on the existing line control values, monitoring alarm thresholds, and the project safety reserve level; the higher the existing line protection level or the more stringent the alarm threshold, The smaller the value.

[0036] Further, in S7 and S8, the risk assessment is used to conduct a comprehensive risk screening of candidate backfill conditions based on the backfill condition risk index, and the hard constraint screening is used to verify the engineering control conditions item by item for candidate backfill conditions that have passed the risk assessment. The two are performed sequentially and the verification objects are different. The verification content of the hard constraint screening includes the existing line settlement prediction value, the existing line settlement rate prediction value, the existing line horizontal displacement prediction value, the deep and shallow foundation differential settlement prediction value, the backfill compaction degree prediction value, the compaction disturbance evaluation value, and the backfill height difference between adjacent zones. Among them, the existing line settlement prediction value is not greater than the existing line settlement control value. The predicted settlement rate of the existing railway line shall not exceed the control value of the settlement rate of the existing railway line. The predicted horizontal displacement of the existing line shall not exceed the control value of the horizontal displacement of the existing line. The predicted value of differential settlement between shallow and deep foundations shall not exceed the control value of differential settlement between shallow and deep foundations. The predicted value of backfill compaction should not be less than the required value of backfill compaction. And not greater than the preset upper limit of compaction degree The compaction disturbance evaluation value shall not exceed the compaction disturbance limit. The backfill elevation difference between adjacent zones shall not exceed the limit value for backfill elevation difference between adjacent zones. Only when all the above verification criteria are met will the candidate backfill conditions proceed to the sorting step, and will be ranked according to the backfill condition risk index. The target backfill schemes are formed in ascending order, and the first-ranked scheme is selected as the execution scheme. If no candidate backfilling condition passes both risk assessment and hard constraint screening, the "no feasible candidate" processing is performed, a new set of candidate backfilling condition parameters is generated, and execution returns to steps S6 to S8. If consecutive... If no candidate backfill conditions are found after the second regeneration, a message indicating that backfilling should be paused and manual review should be provided will be output. The value is an integer from 2 to 5; the pause backfilling and manual review prompts include the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations.

[0037] Furthermore, in S9, the dynamic correction includes first-level correction, second-level correction, and third-level correction;

[0038] When the relative deviation δ between the actual measured settlement and the predicted settlement is s If the value exceeds a first preset threshold, or if the measured value exceeds the prediction range, a first-level correction is triggered; the first preset threshold is 10% to 20%. , To measure the settlement on site, To predict settlement, Settlement control value 5%; the first-level correction is used to adjust the layer thickness, number of compaction passes, type of compaction equipment, or construction interval of subsequent backfill layers;

[0039] When the settlement rate of the existing railway line shows an increasing trend for three consecutive monitoring cycles, and the difference between the settlement rate of the third monitoring cycle and the first monitoring cycle exceeds a second preset threshold, a secondary correction is triggered; where the second preset threshold is the control value for the settlement rate of the existing railway line. 10% to 30%; the secondary correction is used to correct the correction coefficients during the construction stage, the weights of compaction disturbance characteristics, or the parameters of the backfill impact prediction model.

[0040] A third-level correction is triggered when the differential settlement increment between adjacent zones exceeds the third preset threshold, or when the deviation between the measured and predicted results of at least one of the following indicators—existing line settlement, existing line settlement rate, existing line horizontal displacement, and differential settlement between shallow and deep foundations—exceeds the fourth preset threshold for three consecutive monitoring periods; wherein the third preset threshold is the control value for differential settlement between shallow and deep foundations. The third-level correction is used to add the newly added on-site monitoring samples after anomaly verification to the training sample library, retrain or incrementally update the impact prediction model, and compare the updated model with the model before the update on the validation set or reserved test set.

[0041] The model parameters include at least one of the following: number of trees, maximum depth, minimum number of samples per leaf node, learning rate, subsample ratio, and column sampling ratio for the random forest regression model or gradient boosting tree regression model.

[0042] When the trigger conditions for Level 1, Level 2, or Level 3 corrections are met individually, the corresponding level of correction action is executed. When the trigger conditions for at least two levels are met simultaneously, the highest corresponding level of correction action is executed. If the simultaneously met trigger conditions include Level 1 correction trigger conditions, the construction parameter adjustments in Level 1 correction are executed synchronously. When the mean absolute error or root mean square error of the updated model resulting from Level 3 correction increases by more than a preset allowable range relative to the unupdated model, the preset allowable range is 5% to 15%, or a determination coefficient. If the model is lower than the model before the update, retain the model before the update and only perform construction parameter adjustments; otherwise, use the updated model. After completing the first-level correction, second-level correction, or third-level correction, based on the corrected construction parameters, model parameters, and / or training sample library, re-execute S4 to S8 to update the mapping relationship between candidate backfill conditions and predicted response results, risk assessment results, and target backfill schemes for subsequent backfill layers and / or subsequent backfill control zones.

[0043] Meanwhile, the present invention also provides an intelligent optimization system for backfilling unequal depth trenches with ultra-large covers near existing power lines. The system, applying the method described in this invention, includes the following steps:

[0044] The module includes: spatial relationship data acquisition module, backfill control zoning module, construction and monitoring data association module, backfill impact feature construction module, backfill impact prediction module, candidate backfill working condition generation module, risk index calculation module, hard constraint screening module, dynamic correction module, and target backfill scheme output module.

[0045] The spatial relationship data acquisition module is used to acquire spatial relationship data of the super-large cover foundation, trench, and adjacent existing line;

[0046] The backfill control zoning module is used to divide backfill control zones and mark sensitivity levels and construction stage correction factors;

[0047] The construction and monitoring data association module is used to establish the association between construction status data, foundation parameters, compaction equipment parameters, existing line monitoring data and backfill control zones and construction time;

[0048] The backfill impact feature construction module is used to construct a backfill impact feature set including foundation depth difference features, shallow-deep transition stiffness abrupt change features, compaction disturbance features, existing line spatial impact features, construction condition parameter features, and monitoring lag response features; the backfill impact prediction module is used to output the predicted values ​​and prediction ranges of existing line settlement, settlement rate, horizontal displacement, shallow-deep foundation differential settlement, and backfill compaction degree.

[0049] The candidate backfill condition generation module is used to generate a set of candidate condition parameters, update the construction condition parameter characteristics, compaction disturbance characteristics and monitoring lag response characteristics, and establish a mapping relationship between candidate conditions and predicted response results; the risk index calculation module is used to calculate the risk index of backfill conditions.

[0050] The hard constraint screening module is used for engineering control condition verification and sorting; the dynamic correction module is used to correct subsequent schemes based on on-site measured deviations; the target backfill scheme output module is used to output target backfill schemes containing target zones, layers, time, sequence, thickness, materials, compaction equipment parameters, construction intervals, and zone elevation differences.

[0051] The construction and monitoring data association module includes a data preprocessing unit, which is used to identify outliers, fill in missing values, align timestamps, unify units, and normalize the collected data. The preprocessed data is then associated and stored according to the backfill control zone number, backfill level, and construction time. The data preprocessing unit is also used to convert backfill materials, compaction equipment type, sensitivity level, construction stage, and zone backfill sequence into numerical coding features, and to make the training samples and candidate backfill conditions use the same continuous quantity normalization benchmark.

[0052] The risk index calculation module and the hard constraint screening module are connected sequentially. The risk index calculation module is used to perform a comprehensive risk screening of candidate backfill conditions based on the backfill condition risk index. The hard constraint screening module is used to perform item-by-item verification of the candidate backfill conditions that have passed the comprehensive risk screening, including existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, backfill compaction degree, compaction disturbance evaluation value, and backfill height difference between adjacent zones. The candidate backfill condition generation module is also used to regenerate the candidate backfill condition parameter set according to preset conservative adjustment rules when there are no candidate backfill conditions that have passed both the comprehensive risk screening and item-by-item verification. The preset conservative adjustment rules include at least two of the following: reducing the excitation force, reducing the excitation frequency, reducing the number of compaction passes or tamping passes, using low-disturbance compaction equipment, reducing the layer thickness, extending the construction interval time, and reducing the backfill height difference between adjacent zones. The target backfill scheme output module is also used to continuously... If, after regeneration, no candidate backfill conditions that have passed both the initial comprehensive risk screening and item-by-item verification still exist, a prompt to pause backfilling and request manual review will be output. The value is an integer from 2 to 5; the pause backfilling and manual review prompts include the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations.

[0053] The dynamic correction module includes a first-level correction unit, a second-level correction unit, and a third-level correction unit. The first-level correction unit is used to adjust the construction parameters of subsequent backfill layers. The second-level correction unit is used to correct the correction coefficients, compaction disturbance feature weights, or backfill impact prediction model parameters during the construction stage. The model parameters include at least one of the following: number of trees, maximum depth, minimum number of leaf nodes, learning rate, subsample ratio, and column sampling ratio. The third-level correction unit is used to add newly added field monitoring samples after anomaly verification to the training sample library and update the backfill impact prediction model. It is also used to compare the performance of the model before and after the update on the validation set or the reserved test set. When the mean absolute error or root mean square error of the updated model increases by more than a preset allowable range relative to the model before the update (the preset allowable range is 5% to 15%), or when the coefficient of determination R² is lower than that of the model before the update, the model before the update is retained. After the dynamic correction module completes the correction, it triggers the backfill impact feature construction module, the backfill impact prediction module, the risk index calculation module, and the hard constraint screening module to run again.

[0054] Meanwhile, the present invention proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed, it implements the steps of the method described in the present invention.

[0055] Furthermore, the present invention proposes a computer-readable storage medium having a computer program stored thereon, the computer program being configured to implement the steps of the method described in the present invention when invoked by a processor.

[0056] Finally, the present invention provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method described in the present invention.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] 1. This invention divides backfill control zones and calibrates sensitivity levels and construction stage correction coefficients, transforming the spatial attributes of foundations of unequal depths, spatial relationships with existing lines, and construction stage information into engineering features that can be identified by the prediction model, enabling the prediction model to distinguish structural response differences under different zones, different foundation depths, and different existing line protection levels.

[0059] 2. This invention uses the compaction disturbance evaluation value to normalize and weight the physical parameters of the compaction equipment, such as excitation force, excitation frequency, number of compaction passes, layer thickness, equipment travel speed, and minimum clearance to the existing line, and limits the weights to be non-negative and the reference values ​​to be positive, so as to achieve quantitative coupling between the physical parameters of the compaction equipment and the protection level, construction stage, and spatial clearance of the existing line.

[0060] 3. This invention constructs a mapping relationship between candidate backfill conditions and predicted response results, and substitutes variable condition parameters such as backfill material, zonal backfill sequence, construction interval time, layer thickness and compaction equipment parameters into the same backfill influence feature set for updating, so that different candidate backfill schemes can be quantitatively compared under a unified model input and a unified risk index.

[0061] 4. This invention employs a two-stage screening mechanism of risk assessment and hard constraint screening. First, it performs rapid initial screening based on the maximum single-item risk ratio and weighted comprehensive risk. Then, it conducts item-by-item verification of existing line settlement, settlement rate, horizontal displacement, differential settlement between deep and shallow foundations, backfill compaction degree, compaction disturbance, and backfill height difference between adjacent zones. This reduces the possibility of candidate conditions with low comprehensive risk but unmet individual control conditions being mistakenly selected.

[0062] 5. This invention uses a first-level, second-level, and third-level dynamic correction mechanism to adjust subsequent construction parameters, model parameters, or training sample libraries according to the actual measured deviations on site. After correction, feature updates, predictions, risk assessments, and hard constraint screenings are re-executed, enabling subsequent backfill layers and subsequent backfill control zones to be updated in a closed loop with monitoring feedback.

[0063] 6. This invention limits the backfill influence feature set to six core features, which not only covers the key mechanical, geometric and construction condition influencing factors of backfilling trenches of unequal depth, but also avoids directly expanding all original monitoring sequences and irrelevant construction records into the model, thereby reducing feature redundancy and improving the correspondence between candidate condition prediction results and actual construction parameters.

[0064] 7. This invention can complete the prediction, evaluation, risk ranking, and constraint screening of candidate backfilling conditions before construction, shifting the construction decision-making process from experience-based adjustments after monitoring and exceeding limits after construction to the proactive comparison and execution plan output stage before construction; when no candidate conditions meet the risk judgment and hard constraint screening, it triggers the regeneration of conservative parameters or suspends backfilling and manual review, avoiding the forced output of execution plans when safety reserves are insufficient, thereby reducing the possibility of accumulated deformation risks of existing lines and delayed construction adjustments. Attached Figure Description

[0065] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0066] Figure 1 This is a flowchart of the method of the present invention.

[0067] Figure 2 This is a diagram showing the spatial relationship between the basic fertilizer tank and the existing line in this invention.

[0068] Figure 3 This is a diagram showing the relationship between feature prediction and risk assessment in this invention.

[0069] Figure 4 This is a diagram illustrating the candidate operating condition screening and dynamic feedback in this invention.

[0070] Figure 5 This is a flowchart of the dynamic correction hierarchical triggering process in this invention.

[0071] Figure 6 This is a comparison diagram of the effects of the method of the present invention and the traditional layered filling and compaction method.

[0072] Figure 7 This is a ranking chart of candidate operating condition risk indices. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0074] Example 1: As Figure 1 As shown, the technical solution provided in this embodiment is: an intelligent optimization method for backfilling of super-large cover trenches of unequal depths near existing lines, which is applied to the backfilling construction of super-large cover trenches of unequal depths near existing operating lines.

[0075] Specifically, the project object in this embodiment is a subway vehicle depot operation project. The super-large cover has a planar dimension of 240 meters × 80 meters, with a planar projected area of ​​19,200 square meters. The foundation type is an integral raft foundation, with the foundation bottom elevation in the deep foundation zone at -8.5 meters and the foundation bottom elevation in the shallow foundation zone at -4.5 meters, and the maximum foundation depth difference is 4.0 meters. The trench width ranges from 1.2 meters to 3.0 meters, and the depth varies with the foundation depth. The existing line is an underground shield tunnel with an outer diameter of 6.0 meters and a segment thickness of 0.35 meters. The minimum horizontal clearance from the edge of the trench to the outer edge of the existing line structure is 18 meters, and the vertical clearance is 9 meters. The track type is ballastless track, and the protection level is strict protection level.

[0076] Specifically, S1 acquires spatial relationship data of the super-large cover foundation, the trench, and the adjacent existing railway line. In this embodiment, the spatial relationship data includes the planar dimensions and vertical elevation of the super-large cover, the distribution range of the deep and shallow foundation areas and the foundation bottom elevation, the planar position and depth variation of the trench, the axial position and outer contour dimensions of the existing railway tunnel, the position of the existing railway track centerline, and the minimum horizontal and vertical clearance between the edge of the trench and the outer edge of the existing railway structure. The above spatial relationship data is obtained through analysis of construction drawings and on-site measurements, and is used to determine the spatial influence relationship between the trench backfill area and the existing railway line.

[0077] Specifically, S2 divides the backfill area of ​​the trench into at least two backfill control zones based on spatial relationship data, determines the basic attributes of each backfill control zone, and labels the sensitivity level and construction stage correction coefficient. In this embodiment, the backfill area of ​​the trench is divided into four backfill control zones: Zone A, Zone B, Zone C, and Zone D.

[0078] Area A is a sensitive area with deep foundation trenches near the existing line. The foundation bottom elevation is -8.5 meters, the trench depth is about 8.7 meters, the minimum horizontal clearance is 18 meters, and the horizontal distance to the track centerline is about 21 meters. The sensitivity level is high sensitivity, and the correction factor during the construction stage is 1.3.

[0079] Area B is a sensitive area for the transition between shallow and deep trenches. The foundation bottom elevation varies between -8.5 meters and -4.5 meters, resulting in differences in foundation depth and abrupt changes in stiffness during the transition between shallow and deep trenches. The sensitivity level is sensitive, and the correction factor during the construction stage is taken as 1.1.

[0080] Zone C is a general control zone for shallow foundation trenches. The foundation bottom elevation is -4.5 meters, the trench depth is about 4.7 meters, it is far from the existing line, the sensitivity level is general control, and the correction factor during the construction stage is 1.0.

[0081] Zone D is a general backfill area far from the existing line, located at the long end of the cover body, with a horizontal distance of more than 50 meters from the existing line. The sensitivity level is general control, and the correction factor during the construction stage is 1.0.

[0082] In this embodiment, the sensitivity level is determined comprehensively based on the minimum horizontal clearance, foundation depth, whether it is located in the transition zone between shallow and deep foundations, and the existing line protection level: when the minimum horizontal clearance is no more than 20 meters, the foundation depth is no less than 8 meters, and the existing line protection level is strict protection, it is judged as highly sensitive; when the minimum horizontal clearance is greater than 20 meters but no more than 50 meters, or when it is located in the transition zone between shallow and deep foundations and there is a foundation bottom elevation difference of no less than 2.0 meters, it is judged as sensitive; when the minimum horizontal clearance is greater than 50 meters, or the minimum horizontal clearance from the shallow foundation area to the outer edge of the existing line structure is greater than 3 times the foundation depth, it is judged as generally controlled. The current construction stage includes the initial backfilling stage, the middle backfilling stage, and the later backfilling stage, which can be divided according to the construction organization plan and the on-site backfilling progress. The correction factor for the construction stage is determined based on the sensitivity level and the current construction stage. It is 1.2 to 1.5 for highly sensitive areas, 1.0 to 1.2 for sensitive areas, and 0.9 to 1.0 for general control areas. In this embodiment, it is 1.3 for area A, 1.1 for area B, and 1.0 for areas C and D.

[0083] Specifically, S3 collects construction status data, foundation parameters, compaction equipment parameters, and existing line monitoring data, and establishes the correlation between the above data and backfill control zones and construction time. In this embodiment, construction status data includes backfill time information, backfill sequence, layer thickness, backfill material, and compaction equipment type; foundation parameters include foundation soil type, foundation bearing capacity characteristic value, and groundwater level; compaction equipment parameters include excitation force, excitation frequency, equipment travel speed, equipment self-weight, and at least one of compaction passes and tamping passes; existing line monitoring data includes existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, and track deformation. The above data are correlated according to backfill control zones and construction time to form a construction and monitoring dataset associated with spatial location and time series. Among them, track deformation is used as an auxiliary monitoring indicator for anomaly verification and dynamic correction triggering, but is not a mandatory single-item risk ratio in the backfill condition risk index; backfill time information includes the planned start time, duration, and construction period.

[0084] Specifically, the collected data undergoes preprocessing, including outlier identification, missing value completion, timestamp alignment, unit unification, and normalization. Continuous data uses a minimum-maximum normalization benchmark or an engineering reference value normalization benchmark determined by the training samples. Training samples and candidate backfill conditions within the same backfill control zone use the same normalization benchmark. When the normalization denominator is 0, the corresponding feature is set to 0 and marked as a low-variability feature. When existing line monitoring data exhibits a sudden change exceeding a preset anomaly threshold, the corresponding monitoring period is marked as an anomaly verification period and is not directly used as a model training sample. It is added to the training sample library after on-site verification or rule verification. When multiple monitoring points correspond to the same construction time, data records are formed according to the monitoring point location, backfill control zone, and monitoring time.

[0085] Specifically, S4 constructs a backfill impact feature set based on spatial relationship data, basic attributes, construction status data, foundation parameters, compaction equipment parameters, and existing line monitoring data. In this embodiment, the backfill impact feature set includes six core features: foundation depth difference features, abrupt change in stiffness during the transition between shallow and deep sections, compaction disturbance features, existing line spatial impact features, construction condition parameter features, and monitoring lag response features.

[0086] The foundation depth difference feature is one-dimensional, representing the normalized value of the absolute value of the difference between the foundation bottom elevation of each backfill control zone and the foundation bottom elevation of the shallow foundation zone relative to the maximum foundation depth difference.

[0087] The abrupt change in stiffness between light and dark transitions is 1-dimensional, using... This indicates that when the backfill control zone is located within the transition range between light and dark areas, ,in and These are the foundation bottom elevations of adjacent deep foundation zones and shallow foundation zones, respectively, L. g The horizontal length of the transition zone between light and dark areas in the backfill direction. The preset reference burial depth gradient is greater than 0; when the backfill control zone is not located in the depth transition range, Take 0.

[0088] After normalization, the compaction disturbance characteristics are 6-dimensional, including the ratio of the excitation force of the compaction equipment to the reference value of the excitation force, the ratio of the excitation frequency to the reference value of the excitation frequency, the ratio of the number of compaction passes to the reference value of the number of passes, the ratio of the layer thickness to the reference value of the layer thickness, the ratio of the speed correction reference value to the sum of the equipment travel speed and the speed correction reference value, and the ratio of the clearance correction reference value to the sum of the minimum horizontal clearance from the backfill control zone to the outer edge of the existing line structure and the clearance correction reference value.

[0089] The spatial impact characteristics of the existing line are coded into 5 dimensions, including the normalized value of the minimum horizontal clearance from the backfill control zone to the outer edge of the existing line structure, the normalized value of the vertical clearance, the sensitivity level code value, the existing line protection level code value, and the construction stage correction coefficient.

[0090] The construction condition parameters are 6-dimensional, including the backfill material code, material compression modulus or strength grade, backfill start time code, construction interval, zonal backfill sequence code, and compaction equipment type code. When the candidate backfill condition changes the backfill material, compaction equipment type, backfill start time, construction interval, or zonal backfill sequence, the corresponding construction condition parameters are updated synchronously with the candidate backfill condition. The monitoring lag response characteristics are 4-dimensional. Instead of directly expanding all the original monitoring values ​​for each monitoring cycle, they are statistical characteristics calculated based on the previous N monitoring cycles. N is an integer not less than 2. The statistical characteristics include the existing line settlement increment, settlement rate change, horizontal displacement change, and differential settlement increment between deep and shallow foundations.

[0091] Specifically, when the current backfill layer is the first backfill and there is no historical monitoring data, the initialization strategy for the first backfill is adopted, and the existing line settlement increment, settlement rate change, horizontal displacement change and deep-shallow foundation differential settlement increment in the monitoring lag response characteristics are all set to 0; at the same time, in the risk index calculation or hard constraint screening, the boundary value that is more unfavorable to the protection of the existing line in the upper or lower limit of the prediction interval is used as the corresponding prediction value, so that the evaluation result of the first backfill is biased towards safety.

[0092] Specifically, the compaction disturbance characteristics are quantitatively characterized by compaction disturbance evaluation values. The calculation formula is as follows:

[0093]

[0094] In the formula, ,and to All are not less than 0; To provide vibration force for compaction equipment; This is a reference value for the excitation force; The excitation frequency; This is the reference value for the excitation frequency; This refers to the number of compaction passes or tamping passes; This is a reference value for the number of passes; For layer thickness; This is a reference value for the layer thickness; The speed at which the machine travels; Correct the reference value for speed; The minimum horizontal clearance between the backfill control zone and the outer edge of the existing line structure; This is the reference value for net distance correction; where , , , , and All are positive values; The zoning construction correction factor is determined jointly by the sensitivity level of the backfill control zone and the current construction stage. This is a correction factor for the protection level of existing lines; This is a dimensionless evaluation value for compaction disturbance. To and Dimensionless compaction disturbance limits using the same normalized benchmark. It is determined based on the existing line protection level, minimum horizontal clearance, type of compaction equipment, construction stage, and statistical results of similar projects.

[0095] Specifically, in this embodiment, the backfill impact feature set is a 23-dimensional feature set. Among them, the foundation depth difference feature is 1-dimensional, the abrupt change in stiffness during the transition between shallow and deep is 1-dimensional, the compaction disturbance feature is 6-dimensional after normalization, the existing line spatial impact feature is 5-dimensional after encoding, the construction condition parameter feature is 6-dimensional, and the monitoring lag response feature is 4-dimensional.

[0096] Specifically, S5 inputs the backfill impact feature set into the trained backfill impact prediction model to obtain the predicted response results. Insufficient historical samples in the early stages of the project refers to a situation where the number of effective samples is less than five times the dimension of the backfill impact features, or it is impossible to form training, validation, and test sets according to construction time or zoning. In this case, an initial training sample library is constructed using historical data from similar existing railway backfill projects in the same region, backfill data from test sections, and early monitoring data from this project. As the number of measured samples from this project increases, the model is updated using samples from this project first. In this embodiment, the backfill impact prediction model takes the backfill impact feature set of historical construction samples as input and outputs the corresponding historical monitoring response and backfill compaction degree detection value; the model uses at least one of the random forest regression model and gradient boosting tree regression model, with a 23-dimensional backfill impact feature set as input and an output of... , , , and During training, an input matrix X and an output matrix Y are formed. The training, validation, and test sets are partitioned according to construction time sequence or by partitioning and layering, or K-fold cross-validation is used. For the random forest regression model, the number of trees is 100 to 500, the maximum depth is 3 to 20, and the minimum number of samples per leaf node is 1 to 10. For the gradient boosting tree regression model, the number of trees is 100 to 800, the maximum depth is 2 to 10, the learning rate is 0.01 to 0.30, and the subsample ratio is 0.5 to 1.0. The predicted response results include five types of predicted values ​​and prediction intervals. Model validation metrics include MAE, RMSE, and... When the test set If the value is below 0.85, adjust the feature combination or increase the number of training samples and retrain. c,pre It is not the output of the backfilling impact prediction model, but the compaction disturbance evaluation value obtained by normalizing and weighting the compaction disturbance characteristics.

[0097] Specifically, the backfill impact prediction model includes a feature input unit, a numerical encoding and normalization unit, a model training unit, a multi-output prediction unit, and a prediction interval generation unit. The feature input unit receives a 23-dimensional backfill impact feature set; the numerical encoding and normalization unit processes discrete and continuous quantities into the model; the model training unit trains a random forest regression model or a gradient boosting tree regression model based on the input matrix X and the output matrix Y; the multi-output prediction unit uses a single multi-output regression model, or sets five single-output regression models corresponding to existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, and backfill compaction degree, respectively; the prediction interval generation unit generates prediction intervals according to each output index.

[0098] Specifically, each historical construction sample in the training sample library includes at least the sample number, backfill control zone number, backfill layer, construction time, foundation depth difference, minimum horizontal clearance, vertical clearance, backfill material, material compression modulus or strength grade, layer thickness, compaction equipment type, equipment weight, excitation force, excitation frequency, number of compaction passes, equipment travel speed, zone backfill sequence, construction interval time, backfill height difference between adjacent zones, statistical characteristics of the previous N monitoring cycles, corresponding monitoring response, and backfill compaction degree detection value. These fields ensure a clear correspondence between the input data, output data, and construction condition parameters of the backfill impact prediction model; discrete fields are converted to numerical codes before being entered into the model, while continuous fields are normalized according to the same benchmark.

[0099] Specifically, the prediction interval is determined based on the model ensemble results of the backfill impact prediction model, historical residual distribution, test set error distribution, or a preset confidence level. In this embodiment, the prediction interval is the 95% prediction interval; when using a random forest regression model, the prediction interval can be the 2.5% and 97.5% quantiles of the output results of each decision tree, or the sum of the predicted point values ​​and the 2.5% and 97.5% quantiles of the residuals of the training or test set; when using a gradient boosting tree regression model, the prediction interval is the sum of the predicted point values ​​and the 2.5% and 97.5% quantiles of the residuals of the validation or test set. The residuals are calculated separately for each output index, and the prediction interval is used to determine whether the measured values ​​on site exceed the reasonable fluctuation range of the predicted response results, and to provide unfavorable direction boundary values ​​in the initial backfill initialization and risk assessment.

[0100] Specifically, S6 generates a candidate backfill condition parameter set based on the basic attributes, sensitivity level, construction stage correction coefficient, and preset construction parameter range of each backfill control zone. In this embodiment, the preset construction parameter range is determined according to engineering specifications, equipment capabilities, and construction period requirements, including the selectable types of backfill materials, the range of layer thickness, the selectable types of compaction equipment, the range of excitation frequency, the range of compaction passes, and the range of backfill elevation differences between adjacent zones. The candidate backfill condition parameter set is composed of combinations of the above parameters within the preset range. Variable condition parameters include at least one of layer thickness, backfill material, compaction equipment type, excitation frequency, backfill elevation differences between adjacent zones, and the number of compaction passes and the number of tamping passes.

[0101] Specifically, the variable working condition parameters from the candidate backfill working condition parameter set are substituted into the backfill influence feature set of the corresponding backfill control zone. This updates the construction working condition parameter characteristics, compaction disturbance characteristics, and monitoring lag response characteristics that vary with the construction working conditions, forming the backfill influence feature set corresponding to each candidate backfill working condition. The updated backfill influence feature sets are then input into the backfill influence prediction model to obtain the predicted response results for each candidate backfill working condition, establishing a mapping relationship between the candidate backfill working conditions and the predicted response results. This mapping establishes a correspondence between the candidate backfill working condition parameter combinations and the corresponding predicted response results.

[0102] Specifically, S7 calculates the risk index of the backfill condition based on the predicted response results in the mapping relationship between the candidate backfill condition and the predicted response results, and the construction parameters in the parameter set of the candidate backfill condition, and makes a risk assessment.

[0103] In this embodiment, the seven individual risk ratios are first calculated, and their calculation formulas are as follows:

[0104]

[0105] In the formula, The existing line settlement risk ratio, The risk ratio of the existing line's settlement rate. The risk ratio of horizontal displacement of existing railway lines. The differential settlement risk ratio of shallow and deep foundations. For the backfill compaction risk ratio, To solidify the disturbance risk ratio, Risk ratio of backfill elevation difference between adjacent zones; , , , , C max E lim and H lim All are positive values. The preset upper limit of compaction degree is not less than ;when Less than the preset lower limit correction value At that time, with As The compaction degree value in the calculation; when Greater than At that time, with This value is used as the compaction degree in the ρ5 calculation and is also used to check for over-compaction during hard constraint screening; in this embodiment... Take 0.50, Take 0.98.

[0106] Specifically, the formula for calculating the backfilling risk index R is:

[0107]

[0108]

[0109] In the formula, The maximum single-item risk ratio; The multi-indicator comprehensive correction coefficient ranges from 0 to 1.0; w1 to w7 are the weighting coefficients corresponding to ρ1 to ρ7, respectively. to All are not less than 0, and Weighting coefficient to The weights are determined based on the existing line protection level, the sensitivity of monitoring indicators, the contribution rate of historical deformation, the analytic hierarchy process (AHP) or the entropy weight method. When prior information is lacking, each weight is equal.

[0110] The risk assessment rule is: when Greater than or equal to the preset risk threshold When the corresponding candidate backfilling condition is identified as a high-risk prohibited condition and eliminated; when Less than When this happens, the corresponding candidate backfilling condition will be considered as a candidate condition for passing the risk assessment. according to Sure, The risk reserve coefficient is set between 0.10 and 0.50, and is selected within this range based on the existing line control values, monitoring alarm thresholds, and the project safety reserve level; the higher the existing line protection level or the more stringent the alarm threshold, The smaller the value.

[0111] Specifically, S8 performs hard constraint screening on candidate backfilling conditions that have passed risk assessment. When there are candidate backfilling conditions that have passed both risk assessment and hard constraint screening, they are sorted in ascending order of risk index, and the sorted target backfilling schemes are output, with the first-ranked scheme being the execution scheme. When there are no candidate backfilling conditions that have passed both risk assessment and hard constraint screening, no feasible candidate is processed, and the candidate backfilling condition parameter set is regenerated according to the preset conservative adjustment rules, and the process returns to S6 to S8. If no candidate backfilling condition meets the conditions after M consecutive regenerations, a pause backfilling and manual review prompt is output, where M is an integer from 2 to 5. The pause backfilling and manual review prompt includes the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations.

[0112] Hard constraint screening is used to verify the engineering control conditions of candidate backfill conditions that have passed risk assessment. The verification includes the predicted settlement value of the existing railway line, the predicted settlement rate of the existing railway line, the predicted horizontal displacement of the existing railway line, the predicted differential settlement value between shallow and deep foundations, the predicted backfill compaction degree, the evaluation value of compaction disturbance, and the difference in backfill elevation between adjacent zones. Among these, the predicted settlement value of the existing railway line must not exceed the existing railway line settlement control value. The predicted settlement rate of the existing railway line shall not exceed the control value V of the settlement rate of the existing railway line. lim The predicted horizontal displacement of the existing line shall not exceed the control value of the horizontal displacement of the existing line. The predicted value of differential settlement between shallow and deep foundations shall not exceed the control value of differential settlement between shallow and deep foundations. The predicted value of backfill compaction should not be less than the required value of backfill compaction. And not greater than the preset upper limit of compaction degree The compaction disturbance evaluation value shall not exceed the compaction disturbance limit. The backfill elevation difference between adjacent zones shall not exceed the limit value for backfill elevation difference between adjacent zones. Only when all of the above verification criteria are met will the candidate backfilling condition pass the hard constraint screening.

[0113] Specifically, risk assessment and hard constraint screening constitute a two-level screening mechanism. Risk assessment is used to determine the risk index of the backfilling condition. A comprehensive risk screening is performed on candidate backfilling conditions, with a focus on excluding those with a comprehensive risk level exceeding a preset threshold. Hard constraint screening is used to verify the engineering control conditions of candidate backfilling conditions that have passed the risk assessment, ensuring that each individual control indicator meets the requirements for existing line protection and backfilling quality. These two processes are performed sequentially, with different verification objects and functions. If no candidate condition passes both screenings simultaneously, no execution plan is output. Instead, candidate conditions are regenerated by reducing the excitation force, reducing the excitation frequency, reducing the number of compaction passes or tamping passes, using low-disturbance compaction equipment, reducing the layer thickness, extending the construction interval, or reducing the backfill height difference between adjacent zones. If no feasible candidate is found after a preset number of regenerations, a prompt to pause backfilling and provide manual verification is output. This prompt includes the unmet individual control indicator, its corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations.

[0114] The target backfilling plan includes target backfilling control zones, backfilling layers, backfilling time information, zone backfilling sequence, layer thickness, backfilling materials, compaction equipment parameters, construction interval time, and control requirements for backfilling elevation differences between adjacent zones. Among them, the backfilling layer refers to the sequence number or elevation position of the current backfilling layer in the overall backfilling operation.

[0115] Specifically, in this embodiment, taking the first layer of backfill in area A as an example, the control value is set as follows: = 6.0 mm, = 0.5 mm per day, = 4.0 mm, = 8.0 mm, = 0.94, = 0.98, =0.8 meters. α = 0.30, = 1.25, that is = 0.25, to All are set to 1 / 7. In this embodiment, when reliable prior weight difference information is lacking, equal weights are used as the initial weight settings.

[0116] Candidate working condition K1: Layer thickness 0.35 meters, backfill material is graded sand and gravel, compaction equipment is a small plate compactor, excitation frequency 35 Hz, compaction passes 4 times, and the backfill height difference between adjacent zones is 0.4 meters. The predicted response results are as follows: = 3.2 mm, = 0.24 mm per day = 1.6 mm, = 4.3 mm, = 0.955, corresponding It is 0.420. The calculation yields... = 0.533, = 0.480, = 0.400, = 0.538, = 0.984, =0.420, = 0.500, = 0.984, = 1.149. K1 passes the risk assessment and enters the hard constraint screening.

[0117] Candidate working condition K2: Layer thickness 0.40 meters, backfill material is graded sand and gravel, compaction equipment is a medium-sized vibratory roller, excitation frequency 40 Hz, number of compaction passes 5, and the backfill height difference between adjacent zones is 0.5 meters. The predicted response results are as follows: = 4.1 mm, = 0.31 mm per day = 2.2 mm, = 5.6 mm, = 0.962, corresponding It is 0.580. The calculation yields... = 0.683, = 0.620, = 0.550, = 0.700, = 0.977, =0.580, = 0.625, = 0.977, = 1.180. K2 passes the risk assessment and enters the hard constraint screening.

[0118] Candidate working condition K3: Layer thickness 0.38 meters, backfill material is graded sand and gravel, compaction equipment is a small plate compactor, excitation frequency 35 Hz, compaction passes 4 times, and the backfill height difference between adjacent zones is 0.82 meters. The predicted response results are as follows: = 3.5 mm, = 0.26 mm per day = 1.7 mm, = 4.5 mm, = 0.956, corresponding It is 0.430. The calculation yields... = 0.583, = 0.520, = 0.425, = 0.563, = 0.983, = 0.430, = 1.025, = 1.025, = 1.219. Since R = 1.219 is less than R lim = 1.25, K3 passes the risk assessment and enters the hard constraint screening; but = 1.025 is greater than 1, indicating that the backfill elevation difference between adjacent zones exceeds 1. Therefore, K3 was eliminated during the hard constraint screening stage.

[0119] Candidate working condition K4: Layer thickness 0.42 meters, backfill material is graded sand and gravel, compaction equipment is a medium-sized vibratory roller, excitation frequency 42 Hz, number of compaction passes 5, and the backfill height difference between adjacent zones is 0.76 meters. The predicted response results are as follows: = 5.7 mm, = 0.47 mm per day, = 3.6 mm, = 7.7 mm, = 0.958, corresponding It is 0.920. The calculation yields... = 0.950, = 0.940, = 0.900, = 0.963, = 0.981, =0.920, = 0.950, = 0.981, = 1.264. Although the risk ratios of each individual item in K4 are less than 1, and there are no individual items exceeding the limit if only checked by hard constraints, multiple indicators are close to the control limits, resulting in insufficient overall safety reserves. Therefore, because =1.264 greater than = 1.25 was eliminated during the risk assessment stage and will not be included in the hard constraint screening.

[0120] Both K1 and K2 passed the risk assessment and entered the hard constraint screening. After verification of engineering control conditions, both K1 and K2 meet the control requirements for existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, upper and lower limits of backfill compaction, compaction disturbance evaluation value, and backfill elevation difference between adjacent zones. K3 is used to illustrate the situation where the risk assessment is passed but a single hard constraint is not met, and K4 is used to illustrate the situation where a single indicator does not exceed the limit but the overall risk reserve is insufficient. Since the backfilling risk index of K1 is less than that of K2, K1 is determined as the target backfilling scheme for the first layer of Area A.

[0121] Specifically, after completing the screening of the first-level candidate working conditions in area A, the same method was used to calculate the schemes for areas B, C and D, and the examples of multi-zone target backfill schemes are shown in Table 1.

[0122] Table 1. Examples of calculation results for the target backfill schemes in areas A, B, C, and D.

[0123] The risk indices for each zone in Table 1 are obtained using the aforementioned formulas for the seven individual risk ratios, weighting coefficients, and the backfilling condition risk index R. The same control value determination principles and hard constraint verification rules as those used in the example of Zone A are employed. Although Zone B is located in the sensitive area of ​​the deep-shallow transition trench and is affected by abrupt changes in foundation stiffness, its minimum horizontal clearance to the outer edge of the existing line structure is greater than that of Zone A. Furthermore, the target scheme employs a lower disturbance compaction condition; therefore, its comprehensive risk index is lower than that of Zone A.

[0124] Table 1

[0125]

[0126] Specifically, S9 dynamically corrects the construction parameters, model parameters, and / or training sample library of subsequent backfill layers and / or subsequent backfill control zones based on the deviation between the actual field measurement results and the predicted response results, and updates subsequent schemes based on the correction results. Among them, the model parameters include at least one of the following for random forest regression models or gradient boosting tree regression models: number of trees, maximum depth, minimum number of samples per leaf node, learning rate, subsample ratio, and column sampling ratio.

[0127] Dynamic correction is divided into first-level correction, second-level correction, and third-level correction.

[0128] Level 1 correction: When the relative deviation δ between the measured settlement and the predicted settlement is... s Triggered when the value exceeds the first preset threshold, or when the measured value exceeds the prediction range; whereby... , To measure the settlement on site, To predict settlement, Settlement control value The first preset threshold is 15%, and the prediction interval is the 95% prediction interval. The first-level correction is used to adjust the layer thickness, number of compaction passes, type of compaction equipment, or construction interval of subsequent backfill layers, without changing the backfill impact prediction model itself.

[0129] Secondary correction: This is triggered when the settlement rate of the existing line shows an increasing trend for three consecutive monitoring cycles, and the difference in settlement rate between the first and third monitoring cycles exceeds a second preset threshold. In this embodiment, the second preset threshold is set to... 20%, or 0.1 mm per day. Secondary corrections are used to adjust the correction coefficients for the construction phase, the weights of compaction disturbance characteristics, or the parameters of the backfill impact prediction model.

[0130] Level 3 correction: This is triggered when the differential settlement increment between adjacent zones exceeds a third preset threshold, or when the deviation between the measured and predicted results of at least one of the following indicators—existing line settlement, existing line settlement rate, existing line horizontal displacement, and differential settlement between deep and shallow foundations—exceeds a fourth preset threshold within three consecutive monitoring periods. In this embodiment, the third preset threshold is set to... The third preset threshold is 30%, or 2.4 mm; the fourth preset threshold is 20% of the corresponding monitoring indicator control value. The third-level correction is used to add the newly added field monitoring samples after anomaly verification to the training sample library, and retrain or incrementally update the backfilling impact prediction model; the updated model is compared with the unupdated model on the validation set or the reserved test set. When the mean absolute error (MAE) or root mean square error (RMSE) increases by more than 10% relative to the unupdated model, or the coefficient of determination (R²) is lower than that of the unupdated model, the unupdated model is retained and only the construction parameters are adjusted; otherwise, the updated model is used.

[0131] When the triggering conditions for Level 1, Level 2, or Level 3 corrections are met individually, the corresponding level of correction action is executed. When the triggering conditions for at least two levels are met simultaneously, the highest corresponding level of correction action is executed. If the simultaneously met triggering conditions include Level 1 correction triggering conditions, the construction parameter adjustment in Level 1 correction is executed synchronously. After completing Level 1, Level 2, or Level 3 corrections, based on the corrected construction parameters, model parameters, and / or training sample library, S4 to S8 are re-executed to update the mapping relationship between candidate backfill conditions and predicted response results, risk assessment results, and target backfill schemes for subsequent backfill layers and / or subsequent backfill control zones.

[0132] Dynamic correction only applies to subsequent backfill layers and / or subsequent backfill control zones, and does not change the evaluation results of backfill layers and / or backfill control zones that have already been constructed, nor does it constitute a retrospective change to the output target backfill scheme. After dynamic correction is completed, based on the corrected construction parameters, model parameters, and / or training sample library, S4 to S8 are re-executed to update the mapping relationship between candidate backfill conditions and predicted response results, risk assessment results, and target backfill schemes for subsequent backfill layers and / or subsequent backfill control zones.

[0133] Specifically, taking the dynamic correction after the second layer of backfill in Area A as an example, if the actual measured settlement after the completion of this layer is 3.9 mm, while the corresponding predicted settlement is 3.2 mm, the relative deviation is 21.9%, exceeding the first preset threshold of 15%, triggering a first-level correction. Under the premise of meeting the backfill compaction requirements, the system adjusts the thickness of subsequent backfill layers from 0.35 meters to 0.30 meters, the number of compaction passes from 4 to 3, and the construction interval from 12 hours to 24 hours. Then, based on the corrected construction parameters, S4 to S8 are re-executed to update the mapping relationship between candidate backfill conditions and predicted response results, risk assessment results, and target backfill scheme for subsequent backfill layers. Reducing the number of compaction passes is used to reduce compaction disturbance in areas near existing lines, while reducing the layer thickness and extending the construction interval are used to compensate for insufficient compaction that may result from the reduced number of compaction passes. The adjustment is then verified through subsequent compaction test results.

[0134] Example 2: This example also provides an intelligent optimization system for backfilling of large-scale trenches with varying depths near existing power lines, used to implement the above method. The system includes a spatial relationship data acquisition module, a backfill control zoning module, a construction and monitoring data association module, a backfill impact feature construction module, a backfill impact prediction module, a candidate backfill condition generation module, a risk index calculation module, a hard constraint screening module, a dynamic correction module, and a target backfill scheme output module.

[0135] The spatial relationship data acquisition module receives the analysis results of construction drawings, on-site measurement results, and existing line structural data, and generates spatial relationship data between the super-large cover foundation, trench, and adjacent existing line. The backfill control zoning module divides the backfill control zones based on the spatial relationship data and outputs the foundation attributes, sensitivity level, and construction stage correction coefficients for each zone.

[0136] The construction and monitoring data association module is used to collect construction status data, foundation parameters, compaction equipment parameters and existing line monitoring data. It completes outlier identification, missing value completion, timestamp alignment, unit unification and normalization through the data preprocessing unit. Then, it establishes associations according to backfill control zones, backfill layers and construction time. Discrete fields are converted into numerical coding features, and continuous fields use a normalization benchmark consistent with training samples and candidate backfill conditions.

[0137] The system can connect to settlement monitoring points, horizontal displacement monitoring points, track deformation monitoring points, compaction degree testing equipment, compaction equipment operation parameter acquisition terminals, and construction progress recording terminals. Various types of data can be transmitted to the data processing unit through wired communication, wireless communication, or manual input, and the data processing unit will uniformly complete data preprocessing, partition matching, and time series correlation.

[0138] The backfill impact feature construction module is used to construct a backfill impact feature set based on the preprocessed data. This backfill impact feature set includes features related to foundation depth difference, abrupt changes in stiffness between shallow and deep foundations, compaction disturbance, spatial impact of existing railway lines, construction condition parameter features, and monitoring lag response features. The backfill impact prediction module is used to input the backfill impact feature set into the backfill impact prediction model and output predicted values ​​and ranges for existing railway line settlement, existing railway line settlement rate, existing railway line horizontal displacement, differential settlement between shallow and deep foundations, and backfill compaction degree.

[0139] The candidate backfill condition generation module generates a set of candidate backfill condition parameters within a preset construction parameter range. It then substitutes the variable parameters from the candidate conditions into the backfill influence feature set of the corresponding backfill control zone, updating the construction condition parameter features, compaction disturbance features, and monitoring lag response features to form a mapping relationship between candidate backfill conditions and predicted response results. The risk index calculation module calculates the seven individual risk ratios and the backfill condition risk index for each candidate backfill condition. .

[0140] The hard constraint screening module is used to verify the engineering control conditions of candidate backfill conditions that have passed the risk assessment item by item, and output the sorting results in ascending order of risk index of backfill conditions. The target backfill scheme output module is used to output the target backfill scheme in at least one of the following forms: construction parameter table, risk ranking table, zonal backfill sequence diagram, and dynamic correction prompt information. When there are no candidate backfill conditions that have passed both the risk assessment and hard constraint screening, the candidate backfill condition generation module regenerates the candidate backfill condition parameter set according to preset conservative adjustment rules. The preset conservative adjustment rules include at least two of the following: reducing excitation force, reducing excitation frequency, reducing the number of compaction passes or tamping passes, using low-disturbance compaction equipment, reducing layer thickness, extending construction interval time, and reducing the backfill height difference between adjacent zones; if continuous If no feasible candidates are found after the second regeneration, the target backfill scheme output module will output a message indicating that backfilling should be paused and manual review is required. The value is an integer from 2 to 5; the pause backfilling and manual review prompts include the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations.

[0141] The dynamic correction module includes a first-level correction unit, a second-level correction unit, and a third-level correction unit. When the measured results on site deviate from the predicted response results and meet the corresponding triggering conditions, the dynamic correction module corrects the construction parameters, model parameters, and / or training sample library of subsequent backfill layers and / or subsequent backfill control zones. The second-level correction unit can adjust the number of trees, maximum depth, minimum number of leaf nodes, learning rate, subsample ratio, or column sampling ratio. The third-level correction unit adds the newly added on-site monitoring samples after anomaly verification to the training sample library and triggers model updates. It compares the model performance before and after the update on the validation set or the reserved test set. If the performance of the updated model is worse than that of the model before the update, the model before the update is retained. After the correction is completed, the backfill impact feature construction module, backfill impact prediction module, risk index calculation module, and hard constraint screening module are triggered to run again.

[0142] Implementable Case: In another implementable case, the present invention is applied to the full-cycle scheme calculation under the conditions of a subway vehicle depot operation project. Compared with the traditional empirical method, the method of the present invention reduces the maximum settlement of the existing line from 5.8 mm to 3.4 mm, the maximum settlement rate from 0.43 mm / day to 0.25 mm / day, and the differential settlement between shallow and deep foundations from 7.6 mm to 4.8 mm. While prioritizing the safety protection of the existing line, the backfilling construction period is shortened from 36 days to 32 days, and the cost ratio increases from 100% to 103%. The effects of this embodiment are used to illustrate the application effect of the method of the present invention under specific engineering conditions and do not constitute a limitation on the scope of protection of the present invention.

[0143] Example 3: This example proposes an electronic system, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method steps of the present invention.

[0144] Example 4: This example proposes a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method described in this invention, which will not be repeated here.

[0145] Example 5: This example proposes a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the steps of the method described in this invention, which will not be repeated here.

[0146] It should be noted that the processing flow of embodiments 2-5 corresponds to the specific steps of the method provided in embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in this embodiment can be found in the method provided in embodiment 1 of the present invention.

[0147] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0148] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart optimization method for backfilling trenches of unequal depth foundations near existing power lines, characterized in that, Includes the following steps Step S1: Obtain spatial relationship data of the super-large cover foundation, trench, and adjacent existing line; Step S2: Based on the spatial relationship data, divide the backfill area of ​​the trench into at least two backfill control zones, determine the basic attributes of each backfill control zone, and mark the sensitivity level and construction stage correction coefficient. Step S3: Collect construction status data, foundation parameters, compaction equipment parameters and existing line monitoring data. Perform outlier identification, missing value completion, timestamp alignment, unit unification and normalization on the collected data, and establish the correlation between the above data and backfill control zones and construction time. Step S4: Based on spatial relationship data, basic attributes, construction status data, foundation parameters, compaction equipment parameters and existing line monitoring data, construct a backfill influence feature set including foundation depth difference characteristics, deep-shallow transition stiffness abrupt change characteristics, compaction disturbance characteristics, existing line spatial influence characteristics, construction condition parameter characteristics and monitoring lag response characteristics; Step S5: Input the backfill impact feature set into the trained backfill impact prediction model to obtain the prediction response results. The prediction response results include the predicted values ​​and corresponding prediction intervals of existing line settlement, existing line settlement rate, existing line horizontal displacement, deep and shallow foundation differential settlement, and backfill compaction degree. Step S6: Generate a set of candidate backfill working condition parameters based on the basic attributes, sensitivity level, construction stage correction coefficient and preset construction parameter range of each backfill control zone; Step S7: Calculate the risk index of the backfill condition based on the predicted response results in the mapping relationship between the candidate backfill condition and the predicted response results, and the construction parameters in the parameter set of the candidate backfill condition, and make a risk assessment. Step S8: Perform hard constraint screening on the candidate backfill conditions that have passed the risk assessment; Step S9: Based on the deviation between the on-site measured results and the predicted response results, dynamically correct the construction parameters, model parameters, and / or training sample library of the subsequent backfill layer and / or subsequent backfill control zone, and update the subsequent scheme based on the correction results; after completing the dynamic correction, based on the corrected construction parameters, model parameters, and / or training sample library, re-execute steps S4 to S8 to update the mapping relationship between the candidate backfill conditions and the predicted response results, the risk assessment results, and the target backfill scheme of the subsequent backfill layer and / or subsequent backfill control zone.

2. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 1, characterized in that, In step S2, the backfill control zones include at least two of the following: a sensitive zone for deep foundation trenches near the existing line, a sensitive zone for transitional deep and shallow foundation trenches, a general control zone for shallow foundation trenches, and a general backfill zone far from the existing line. The sensitivity level is determined based on the minimum horizontal clearance between each backfill control zone and the outer edge of the existing line structure, the foundation depth, and the protection level of the existing line. The construction stage correction coefficient is determined based on the sensitivity level and the current construction stage, which includes the initial backfilling stage, the middle backfilling stage, and the later backfilling stage.

3. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 1, characterized in that, In step S3, the construction status data includes backfilling time information, backfilling sequence, layer thickness, backfilling material, and compaction equipment type; the foundation parameters include foundation soil type, foundation bearing capacity characteristic value, and groundwater level; the compaction equipment parameters include excitation force, excitation frequency, equipment travel speed, equipment self-weight, and at least one of compaction passes and tamping passes; the existing line monitoring data includes existing line settlement, existing line settlement rate, existing line horizontal displacement, differential settlement between deep and shallow foundations, and track deformation.

4. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 1, characterized in that, In step S4, the backfill influence feature set includes foundation depth difference features, shallow-deep transition stiffness abrupt change features, compaction disturbance features, existing line spatial influence features, construction condition parameter features, and monitoring lag response features. The spatial impact characteristics of the existing line are composed of the normalized value of the minimum horizontal clearance from the backfill control zone to the outer edge of the existing line structure, the normalized value of the vertical clearance, the sensitivity level code value, the existing line protection level code value, and the construction stage correction coefficient; the construction condition parameter characteristics are composed of at least two of the following: backfill material code value, material compression modulus or strength grade, backfill planned start time code value, construction interval time, zone backfill sequence code value, and compaction equipment type code value.

5. The intelligent optimization method for backfilling trenches of unequal depth foundations near existing power lines according to claim 4, characterized in that, In step S4, the monitoring lag response characteristics are statistical characteristics calculated based on the previous N monitoring cycles, where N is an integer not less than 2. The statistical characteristics include the settlement increment of the existing line, the change in settlement rate, the change in horizontal displacement, and the differential settlement increment between shallow and deep foundations. When the current backfill layer is the first backfill and there is no historical monitoring data, the first backfill initialization strategy is adopted, the statistical characteristics are set to 0, and the unfavorable direction boundary value of the prediction interval is used as the corresponding prediction value in the risk index calculation or hard constraint screening.

6. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 5, characterized in that, In step S4, the compaction disturbance characteristics are quantitatively characterized by the compaction disturbance evaluation value, the calculation formula of which is: ； In the formula, ,and to All are not less than 0; To provide vibration force for compaction equipment; This is a reference value for the excitation force; The excitation frequency; This is the reference value for the excitation frequency; This refers to the number of compaction passes or tamping passes; This is a reference value for the number of passes; For layer thickness; This is a reference value for the layer thickness; The speed at which the machine travels; Correct the reference value for speed; The minimum horizontal clearance between the backfill control zone and the outer edge of the existing line structure; This is a reference value for net distance correction; in , , , , and All are positive values; The zoning construction correction factor is determined jointly by the sensitivity level of the backfill control zone and the current construction stage. This is a correction factor for the protection level of existing lines; This is a dimensionless evaluation value for compaction disturbance. To and Dimensionless compaction disturbance limits using the same normalized benchmark. It is determined based on the existing line protection level, minimum horizontal clearance, type of compaction equipment, construction stage, and statistical results of similar projects.

7. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 5, characterized in that, In step S7, the calculation of the backfilling condition risk index includes seven individual risk ratios, the formulas of which are as follows: ； In the formula, The settlement risk ratio of the existing line; Risk ratio of existing line settlement rate; The risk ratio of horizontal displacement of existing lines; The risk ratio of differential settlement between shallow and deep foundations; Risk ratio for backfill compaction; To solidify the disturbance risk ratio; Risk ratio of backfill elevation difference between adjacent zones; These are the settlement control values ​​for existing railway lines; This is the control value for the settlement rate of the existing railway line; These are the horizontal displacement control values ​​for existing railway lines; This represents the control value for differential settlement between shallow and deep foundations; This refers to the required compaction degree for backfill. Set the elevation difference limit for backfilling adjacent zones; Backfill the elevation difference between adjacent zones; This is the predicted settlement value for the existing railway line; This is the predicted settlement rate of the existing railway line. The predicted horizontal displacement of the existing line; This represents the predicted value for differential settlement between shallow and deep foundations. This is the predicted value for backfill compaction. , , , , C max , and All are positive values; To preset the upper limit of compaction and Not less than .

8. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 7, characterized in that, In step S7, the formula for calculating the backfilling risk index R is: ； ； In the formula, The maximum single-item risk ratio; The multi-indicator comprehensive correction coefficient ranges from 0 to 1.0; to They are respectively to The corresponding weighting coefficients, to All are not less than 0, and .

9. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 1, characterized in that, In step S6, the candidate backfill condition parameter set includes at least two types of variable condition parameters among backfill material parameters, layer thickness, compaction equipment type and parameters, backfill time information, zonal backfill sequence, construction interval time, and backfill height difference between adjacent zonals. The variable condition parameters in the candidate backfill condition parameter set are substituted into the backfill influence feature set of the corresponding backfill control zone to update the construction condition parameter features, compaction disturbance features, and monitoring lag response features, forming the backfill influence feature set corresponding to each candidate backfill condition. The updated backfill influence feature sets are then input into the backfill influence prediction model to obtain the predicted response results of each candidate backfill condition, establishing a mapping relationship between the candidate backfill conditions and the predicted response results.

10. The intelligent optimization method for backfilling trenches of foundations with unequal depths near existing power lines according to claim 1, characterized in that, In step S8, candidate backfill conditions that pass the risk assessment are subjected to hard constraint screening. When there are candidate backfill conditions that pass both the risk assessment and hard constraint screening, they are sorted in ascending order of risk index, and the sorted target backfill schemes are output, with the first-ranked scheme being the execution scheme. When there are no candidate backfill conditions that pass both the risk assessment and hard constraint screening, the layer thickness is reduced, the excitation force is reduced, the excitation frequency is reduced, the number of compaction passes or tamping passes is reduced, low-disturbance compaction equipment is used, the construction interval time is extended, and / or the backfill height difference between adjacent zones is reduced according to preset conservative adjustment rules, and candidate backfill conditions are regenerated. Fill in the working condition parameter set and return to execute steps S6 to S8. If no candidate backfill working condition passes both risk assessment and hard constraint screening after M consecutive regenerations, output a prompt to pause backfilling and perform manual review, where M is an integer from 2 to 5. The prompt to pause backfilling and perform manual review includes the unmet single control indicator, the corresponding predicted value and limit, the most recent conservative adjustment parameter, and the number of regenerations. The target backfilling scheme includes the target backfilling control zone, backfilling layer, backfilling time information, zone backfilling order, layer thickness, backfilling material, compaction equipment parameters, construction interval time, and control requirements for the backfilling height difference between adjacent zones.

Images