Shield tunneling parameter variable complexity grouping sampling multi-process prediction method
By using a multi-process prediction method based on the complexity of shield tunneling parameter variables through group sampling, the problem of resource allocation deviation caused by short-term high-frequency fluctuations is solved, resulting in more stable and efficient prediction results, and improving the utilization efficiency of computing resources and the coordination of the prediction process.
Patent Information
- Application Number
- CN202611236424.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies are overly sensitive to instantaneous amplitude fluctuations when characterizing the complexity of shield tunneling parameters, especially during short-term, high-frequency changes. This causes resource allocation to deviate from the actual trend, affecting overall prediction efficiency and result stability.
By analyzing the number of amplitude fluctuations, the difference between peak intervals, and the duration of changes, the activity range of variables is divided, the disturbance decay characteristics of short-term increases and rapid declines are identified, the intensity results of complex states are constructed, and the predicted attention level is adjusted according to the amplitude diffusion range and the fluctuation coverage span to determine the dynamic resource tilting strategy and realize the orderly transfer of resources between the decline and expansion areas.
It improves the stability and continuity of shield tunneling parameter prediction results, reduces misjudgments caused by short-term fluctuations, enhances the efficiency of computing resource utilization, and maintains the coordinated operation of multi-process prediction.
Smart Images

Figure CN123068089A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, specifically to a multi-process prediction method for the complexity of shield tunneling parameter variables through group sampling. Background Technology
[0002] Multi-process prediction based on the complexity of shield tunneling parameters is an efficient data prediction strategy for shield tunneling construction. Its core lies in the structured, hierarchical processing and parallel computation of multi-source operating parameters. Specifically, firstly, parameters collected during shield tunneling, such as propulsion speed, cutterhead torque, earth pressure, and synchronous grouting volume, are used to construct complexity profiles from the perspectives of fluctuation intensity, change frequency, and coupling correlation. Then, based on the differences in complexity, all parameters are divided into multiple hierarchical groups, and representative sampling is performed within each group to reduce redundant data interference and maintain key feature representation. On this basis, different groups are mapped to independent computation processes for synchronous prediction calculations, allowing high-complexity parameters to receive more refined processing paths while maintaining a rapid deduction pace for low-complexity parameters. Finally, by unifying and integrating the output results of each process, an overall prediction result reflecting the changing trends of the shield tunneling state is formed, thereby achieving a synergistic improvement in computational efficiency and prediction accuracy.
[0003] The existing technology has the following shortcomings: In the process of characterizing the complexity of shield tunneling parameters, the complexity assessment mechanism is overly sensitive to instantaneous amplitude fluctuations when faced with short-term, high-frequency changes. It judges transient disturbances that only occur within a local time frame as persistent high-complexity states. When outputting the complexity assessment results, the group sampling strategy continuously classifies the corresponding parameters into high-complexity groups based on the assessment results and triggers high-intensity prediction paths. As time progresses, the transient disturbance dissipates, but the group status is not synchronously reverted, and computational resources are still concentrated on this parameter. This causes the resource allocation direction to deviate from the actual change trend, while weakening the ability to process other continuously changing parameters, ultimately affecting the overall prediction efficiency and result stability.
[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0005] The purpose of this invention is to provide a multi-process prediction method for the complexity of shield tunneling parameter variables through group sampling, in order to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-process prediction method for the complexity of shield tunneling parameter variables by group sampling, comprising the following steps: The parameters and variables obtained during the tunnel boring machine are analyzed over a continuous time range. The number of amplitude fluctuations, the difference between peak intervals, and the duration of change are extracted. Based on the corresponding change relationship between the number of amplitude fluctuations and the duration of change, the activity range of variables is divided to obtain the basis for grouping variables by complexity. The variable complexity is grouped according to the corresponding amplitude change process. The fallback span and fallback time values in the amplitude change process are extracted. The segments with short-term rise and rapid fall are identified and the disturbance attenuation characteristics are output. Based on the time range corresponding to the disturbance decay characteristics, the fluctuation dwell time value is extracted, and the fluctuation dwell time value is correlated with the fallback time value to identify the change segment with insufficient continuous expansion capability and generate the complex state retention strength result. For complex states, retain the change clustering area corresponding to the intensity result, statistically analyze the amplitude diffusion range and fluctuation coverage span, adjust the predicted attention level according to the change rhythm of the amplitude diffusion range value, and determine the dynamic resource allocation strategy. The trajectory of attention change is extracted according to the dynamic resource tilt strategy. The migration rate and dwell ratio values in the trajectory of attention change are dynamically adjusted. The computational input corresponding to the fallback area is gradually transferred to the expansion area, and the dynamic allocation of resources is completed in the multi-process prediction process of shield tunneling parameter variable complexity grouping sampling.
[0007] Preferably, the variable complexity grouping is based on the following steps: The parameters of advance speed, cutterhead torque, earth pressure and synchronous grouting volume are arranged in the order of collection time. The changes in values at adjacent time nodes are compared, the rising and falling segments are marked, the turning points are counted to form the amplitude fluctuation number, the peak value is recorded and the time interval is calculated to obtain the peak interval difference, and the interval time difference is accumulated to obtain the change duration value. The numerical values of the number of amplitude fluctuations and the numerical values of the duration of change are matched and arranged, and the segments corresponding to different change relationships are classified to obtain the expression form of activity level; By introducing the peak interval difference to compare the classified segments, the time segments are grouped according to the characteristics of the change rhythm, and the variable activity interval is constructed. The activity level intervals corresponding to each time segment are marked and summarized. The activity level intervals are then mapped to parameter variables to complete the classification and grouping, and the grouping basis for variable complexity is obtained.
[0008] Preferably, for each time segment corresponding to the variable activity level interval, the distribution relationship of the amplitude fluctuation number, peak interval difference, and change duration value in the time sequence is recorded. Time segments with continuous changes in amplitude fluctuation number and synchronous changes in change duration are marked, and time segments with consistent trends in peak interval difference are centrally collected. The grouping of variable complexity is refined based on the time distribution location.
[0009] Preferably, the perturbation attenuation feature output steps are as follows: Based on the grouping of variable complexity, time segments are defined, the trajectory of parameter variable amplitude changes is unfolded point by point, the rising phase and peak position are marked and the time nodes are recorded, the falling path is tracked along the time axis, the difference between the peak and the falling end position is calculated to obtain the falling span value, and the time difference from the peak time node to the stable interval is calculated to obtain the falling time value. The corresponding values of the fallout span and fallout time are organized, and the fallout behaviors in each time segment are arranged in chronological order. The interval between adjacent fallout behaviors is recorded and associated. By referring to the correspondence between the pullback span value and the pullback time value, the connection between the rising stage and the pullback stage in the amplitude change trajectory is identified, and the change segments that rise briefly and fall immediately are extracted and arranged. The extracted change segments are integrated, and the start time, peak time, and fall-off end time of each segment are recorded and summarized in chronological order to output the disturbance decay characteristics.
[0010] Preferably, for a short-term rise and rapid fall change segment, the time node corresponding to the peak position and the time node of the fall end are continuously recorded, the time interval between adjacent change segments is associated with the amplitude change path, and the fall span value and fall time value are summarized and organized, and the disturbance attenuation characteristics are output in chronological order.
[0011] Preferably, the steps for generating the strength results for complex states are as follows: Mark time segments according to disturbance attenuation characteristics, unfold the amplitude change trajectory point by point, identify the position of entering the fluctuation range and record the start and end times, calculate the time difference to obtain the fluctuation dwell time value. By combining the fluctuation dwell time value and the fall time value, segment by segment matching is performed, and the corresponding time nodes are aligned and arranged to establish the correspondence between the fluctuation dwell interval and the fall interval. By introducing the correspondence between the fluctuation dwell time value and the fall time value, the change segment is identified, and the fluctuation dwell interval and fall interval are combined into a complete change segment and arranged in chronological order. The identified change segments are summarized, and the time nodes of each segment, the corresponding fluctuation dwell time value and the fallback time value are integrated to generate the complex state retention strength result.
[0012] Preferably, the corresponding markings are made by combining the continuous distribution interval of the fluctuation dwell time value and the concentrated change segment of the fall time value. The change segment that immediately follows the end of the fluctuation dwell interval is screened and arranged in chronological order. The start time, dwell end time and fall end time of each change segment are recorded uniformly.
[0013] Preferably, the steps for determining the dynamic resource allocation strategy are as follows: According to the intensity results of the complex state, the change segments are marked, the time interval and amplitude connection of adjacent change segments are read, the segments that meet the continuous change relationship are merged and the time range is recorded to obtain the change cluster area; By combining the amplitude change trajectory within the change cluster area, the amplitude data is read point by point and the maximum and minimum amplitude values are filtered out. The difference is calculated to obtain the amplitude diffusion range value. By introducing the time distribution within the change cluster area, arranging the time intervals of each change segment and statistically analyzing the coverage time range, the value of the fluctuation coverage span is obtained. The amplitude diffusion range is recorded in segments to identify increasing, decreasing, and alternating segments, and to determine the rhythm of change. By combining the numerical change rhythm of amplitude diffusion range with the numerical value of fluctuation coverage span, the predicted attention level corresponding to each change cluster area is adjusted to obtain a dynamic resource allocation strategy.
[0014] Preferably, the incremental change segments in the amplitude diffusion range numerical change rhythm and the continuous change segments in the fluctuation coverage span numerical change are marked accordingly, and the predicted attention level within the corresponding time interval is continuously adjusted. At the same time, the decreasing change segments and the coverage interruption segments are separated and marked. By arranging each change cluster area in chronological order, the continuous change processing of the predicted attention level between adjacent areas is completed.
[0015] Preferably, the trajectory of attention changes corresponding to the dynamic resource allocation strategy is processed, and dynamic resource allocation is completed by extracting migration rate and dwell ratio values. The specific steps are as follows: Read the corresponding change clustering areas of the dynamic resource tilt strategy in chronological order, extract and arrange the attention level values of each time node, record the change nodes and connect the time intervals to obtain the trajectory of attention level change. By referring to the trajectory of changes in attention, the time intervals and magnitudes of changes of adjacent change nodes are extracted. The time intervals and magnitudes of changes are paired and organized to obtain migration rate values, which are then arranged in chronological order. The time intervals are divided based on the migration rate value, the intervals with stable attention levels are identified and the time lengths are counted, and the time lengths are matched with the overall time range to obtain the dwell ratio value. The migration rate and the dwell ratio are combined for classification, and the expansion area and the fallback area are divided and the input changes are adjusted accordingly. Based on the time distribution of the expansion and fallback regions, the computing resources are transferred and allocated, and the changes at each time node are recorded to complete the dynamic allocation of resources.
[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention jointly characterizes the correspondence between the number of amplitude fluctuations and the duration of the change, and introduces the values of the fallback span and fallback time to identify short-term rise and rapid fall segments. This allows short-term disturbances to be distinguished and processed during the complexity assessment stage, thereby avoiding the continuous judgment of instantaneous fluctuations as high-complexity states. This processing method makes the complexity grouping results more consistent with the actual change process of parameter variables, reduces misjudgments caused by short-term fluctuations, and thus improves the stability and continuity of the overall prediction results.
[0017] This invention characterizes the change clustering region by constructing a complex state retention intensity result and combining the amplitude diffusion range value and the fluctuation coverage span value. It further forms a dynamic resource tilting strategy and adjusts the migration rate value and the dwell ratio value based on the change trajectory of attention, so that the computing input can be transferred in an orderly manner between the fallback region and the expansion region. This method enables the resource allocation process to be adjusted synchronously with the parameter change trend, thereby improving the efficiency of computing resource utilization and maintaining the coordinated operation of the multi-process prediction process. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0019] Figure 1 This is a flowchart of the method for predicting the complexity of shield tunneling parameter variables through grouped sampling and multi-process prediction according to the present invention. Detailed Implementation
[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.
[0021] This invention provides, for example Figure 1 The shield tunneling parameter variable complexity grouping sampling multi-process prediction method shown includes the following steps: The parameters and variables obtained during the tunnel boring machine are analyzed over a continuous time range. The number of amplitude fluctuations, the difference between peak intervals, and the duration of change are extracted. Based on the corresponding change relationship between the number of amplitude fluctuations and the duration of change, the activity range of variables is divided to obtain the basis for grouping variables by complexity. During tunnel boring machine (TBM) excavation, the collected parameters change continuously over time. Different parameters exhibit differences in their frequency of change, duration, and fluctuation rhythm. By refining and breaking down these differences, a grouping basis for characterizing the complexity of the variables can be constructed. The specific steps are as follows: The parameters of propulsion speed, cutterhead torque, earth pressure, and synchronous grouting volume collected during the shield tunneling process are arranged in chronological order. A trajectory of amplitude variation within a continuous time range is established for each parameter variable. The numerical changes between adjacent time nodes are compared point by point. When the value of a later time node is greater than that of a previous time node, it is marked as an upward segment; when the value of a later time node is less than that of a previous time node, it is marked as a downward segment. The turning point between consecutive upward and downward segments is used as the boundary position of amplitude fluctuation. All turning points are statistically analyzed to form the number of amplitude fluctuations. Simultaneously, the peak position corresponding to each turning point is recorded. The time interval between adjacent peaks is calculated in chronological order to form a set of peak interval difference values. Furthermore, the time is accumulated for each consecutive upward or downward interval, and the time difference between the start and end time of the interval is used as the duration of change. This ensures that the number of amplitude fluctuations, peak interval difference, and duration of change values correspond to each other within the same time dimension.
[0022] The amplitude fluctuation count and duration values extracted within the same time range are matched segment by segment. The amplitude fluctuation count and duration values in each time segment are arranged in a one-to-one correspondence and formed into multiple sets of combined data in chronological order. The amplitude fluctuation count and duration values in each set of combined data are compared and analyzed. Segments with continuously increasing amplitude fluctuation count and synchronously increasing duration values are grouped into one category, segments with frequently changing amplitude fluctuation count but dispersed duration values are grouped into another category, and segments with fewer amplitude fluctuation counts and continuously distributed duration values are classified separately. By classifying and organizing different sets of combined data, the changing behavior of parameter variables in each time segment is transformed into a distinguishable expression of activity level, providing a clear data foundation for subsequent activity level interval division.
[0023] The combined data results from the previous stage are fused with the set of peak interval difference values. The peak interval difference in each time segment is read, and the peak interval difference is compared with the amplitude fluctuation number and duration of change in the corresponding time segment. By identifying the segment where the peak interval difference is continuously shortened, the segment with accelerated change rhythm is identified. By identifying the segment where the peak interval difference is gradually extended, the segment with slowed change rhythm is identified. The correspondence between the change rhythm information, amplitude fluctuation number, and duration of change is comprehensively organized.
[0024] Time segments with the same rhythmic characteristics and consistent combinations of amplitude fluctuations and duration of change are grouped together, so that the changing behavior of different time segments can be uniformly divided according to rhythmic distribution characteristics, thereby forming multiple levels of variable activity intervals.
[0025] The already defined variable activity intervals are integrated and processed as a whole. Each time segment corresponding to each parameter variable within a continuous time range is marked according to its corresponding activity interval. The marking results of all time segments are summarized in chronological order to form an activity distribution map covering the entire time range. The different activity intervals in this distribution map are mapped to specific parameter variables. The activity level of each parameter variable in different time intervals is uniformly labeled. Based on the time proportion and distribution location of each activity interval, the parameter variables are classified and grouped, thereby constructing a grouping basis for describing the complexity of shield tunneling parameter variables. This provides a clear data partitioning foundation for subsequent complexity grouping sampling and multi-process prediction.
[0026] The variable complexity is grouped according to the corresponding amplitude change process. The fallback span and fallback time values in the amplitude change process are extracted. The segments with short-term rise and rapid fall are identified and the disturbance attenuation characteristics are output. Given a clear grouping criterion for variable complexity, a quantitative description of the decline behavior can be achieved by meticulously breaking down the amplitude changes within the corresponding time range. This allows for the identification of short-term rises followed by rapid falls, and the formation of a continuous disturbance decay characteristic expression. The specific steps are as follows: Around the time segments defined in the variable complexity grouping criteria, the amplitude change trajectory of each parameter variable within that time segment is unfolded point by point. The continuously recorded values are arranged in chronological order, and the numerical changes between adjacent time nodes are compared item by item. When the value enters a continuous rising phase from the stable interval, the starting position is recorded, and the numerical change is continuously tracked until the peak position is reached. The peak position is time-marked and the value is recorded. Then, subsequent values are read along the time axis. When the value begins to change downward from the peak, the falling path is continuously recorded. The difference between the value corresponding to the peak and the value at the end of the fall is calculated, and this difference is used as the fall span value. At the same time, the time difference between the time node of the peak occurrence and the time node of the value recovering to the stable interval is accumulated segment by segment, and the accumulated result is recorded as the fall time value. By performing the above processing on the falling segment corresponding to each peak, the fall behavior in the amplitude change process is formed into a set of corresponding data consisting of the fall span value and the fall time value.
[0027] The drop span values and drop duration values formed within each time segment are systematically matched and organized. Each set of drop span values and drop duration values is arranged in chronological order to form a continuous record. The interval time between adjacent drop behaviors is marked, and the interval time is associated with the corresponding drop span value and drop duration value. Segments where the drop span value appears continuously and the corresponding drop duration value changes in a concentrated manner are centrally identified. Segments where the drop span value is scattered and the drop duration value changes continuously are segmented and recorded. By carefully distinguishing the distribution of different drop behavior combinations, the amplitude drop behavior is given a structured expression with a one-to-one correspondence in both the amplitude change and time change dimensions.
[0028] Based on the data corresponding to the already compiled pullback behavior, the connection between the rising and falling segments in the amplitude change process is identified. Segments where the amplitude enters a continuous rising phase from the stable range and reaches a peak in a short period of time are extracted. The pullback behavior that immediately follows this segment is associated with the corresponding pullback span value and pullback time value and are bound and recorded with the rising segment. Segments with a short rising phase duration followed by a pullback behavior are marked separately. Segments with a very small interval between the rising and falling phases are centrally organized and arranged in chronological order to form a continuous distribution, so that the segments with short-term rise and rapid pullback can form a clear distribution structure within the overall time range.
[0029] The marked short-term rise and rapid fall segments are uniformly integrated and processed. The fall span and fall time values corresponding to each segment are summarized. The start time, peak time and fall end time of each segment are uniformly recorded and arranged in chronological order. The time interval and amplitude change relationship between adjacent segments are supplemented and labeled. All segments form a complete disturbance change description path. The amplitude change trajectory of each segment in the path is expressed as a whole, so that the correspondence between short-term rise and rapid fall forms a continuous mapping in the time dimension. Thus, the disturbance attenuation characteristics that can reflect the disturbance dissipation process during amplitude change are output.
[0030] Based on the time range corresponding to the disturbance decay characteristics, the fluctuation dwell time value is extracted, and the fluctuation dwell time value is correlated with the fallback time value to identify the change segment with insufficient continuous expansion capability and generate the complex state retention strength result. Given that the disturbance attenuation characteristics have been clearly defined, by further refining the dwell behavior of amplitude changes within the corresponding time range, the sustained performance of the change segments can be decomposed and expressed, thereby identifying change segments with insufficient sustained expansion capability and forming the results of complex state retention strength. The specific steps are as follows: For each time segment marked in the disturbance attenuation feature, the amplitude change trajectory within the segment is processed point by point. All sampling points within the time segment are arranged in chronological order. The starting position of the amplitude entering the fluctuation state is identified point by point. When the amplitude enters the continuous fluctuation interval from the relatively stable interval, the entry time point is recorded, and the amplitude change at each subsequent time point is continuously tracked. When the amplitude fluctuates repeatedly within a certain range without continuous unidirectional change, the change segment is defined as the fluctuation dwell interval. The start time and end time of the fluctuation dwell interval are recorded, and the time difference between the two is accumulated segment by segment to form the fluctuation dwell duration value. At the same time, multiple fluctuation dwell intervals existing in the same time segment are independently marked, and the fluctuation dwell duration values of each interval are arranged in chronological order, so that a complete set of fluctuation dwell duration values is formed within the time range corresponding to each disturbance attenuation feature.
[0031] The extracted fluctuation dwell time value within each time segment is matched segment by segment with the recorded fallback time value within that time segment. Using time nodes as alignment benchmarks, the end time of the fluctuation dwell interval is matched with the start time of the adjacent fallback interval. The fluctuation dwell time value and fallback time value within the same time segment are bound one by one. Each pair of values is arranged in chronological order and its positional relationship within the entire time segment is recorded. Segments with continuous distribution of fluctuation dwell time values and concentrated occurrence of corresponding fallback time values are marked. Segments with large intervals in the distribution of fluctuation dwell time values and segmented changes in fallback time values are grouped and recorded, so that fluctuation dwell behavior and fallback behavior form a clear matching structure in the time dimension.
[0032] By combining the established matching relationship between fluctuation dwell time and fallback time, the change segments within each time interval are identified in detail. Each fluctuation dwell interval and its corresponding fallback interval are treated as a complete change segment. The fluctuation dwell time and fallback time are compared, and segments with concentrated fluctuation dwell time and continuous fallback time are classified. Segments with scattered fluctuation dwell time and discontinuous fallback time change paths are individually identified. Segments where a fallback interval immediately follows the end of a fluctuation dwell interval are extracted. These change segments are arranged in chronological order, and the start time, dwell end time, and fallback end time of each segment are fully recorded. This allows change segments with insufficient continuous expansion capability to form an identifiable distribution set within the overall time range.
[0033] The identified change segments are integrated and processed as a whole. The fluctuation dwell time and fallback time values corresponding to each change segment are summarized and organized. The segments are arranged in a unified manner according to time order. The time intervals and amplitude change connections between adjacent segments are supplemented and marked to form a continuous time distribution structure of all change segments. The dwell behavior and fallback behavior of each change segment are uniformly calibrated. By comprehensively expressing the distribution of each segment over the entire time range, a complex state retention strength result that can reflect the persistence of the change state is generated.
[0034] For complex states, retain the change clustering area corresponding to the intensity result, statistically analyze the amplitude diffusion range and fluctuation coverage span, adjust the predicted attention level according to the change rhythm of the amplitude diffusion range value, and determine the dynamic resource allocation strategy. Given that the intensity results for complex states have already been established, by further refining and decomposing the regions of concentrated change within the corresponding time range, the spatial expansion and temporal coverage of amplitude changes can be quantified. Based on this, the level of attention for prediction can be dynamically adjusted, thus forming a clear dynamic resource allocation strategy. The specific steps are as follows: Around the marked change segments in the intensity results of complex states, the intervals between adjacent change segments on the time axis are read one by one. The time interval between two change segments is compared with the connection of the corresponding amplitude changes. When the time interval between adjacent change segments is less than a preset time threshold and there is a continuous change relationship between the ending amplitude of the previous change segment and the starting amplitude of the next change segment, the two change segments are merged. The overall time range after merging is remarked. Multiple consecutive change segments that meet the conditions are gradually merged into a complete interval, thus forming a change cluster region. The start time, end time, and arrangement order of the change segments within each change cluster region are completely recorded. At the same time, all amplitude change trajectories within the region are connected in chronological order, so that the change cluster region has a continuous amplitude change path expression.
[0035] For each established change cluster, the amplitude variation range is statistically analyzed point by point. The amplitude data corresponding to all time nodes in the region is read one by one, and the amplitude of each time node is recorded. The node with the largest amplitude and the node with the smallest amplitude are selected in the entire region. The difference between the largest amplitude and the smallest amplitude is taken as the amplitude diffusion range value of the change cluster. At the same time, the path of amplitude change from the minimum value to the maximum value is completely recorded, and each intermediate node passed through by the amplitude change is marked, so that the amplitude diffusion range value not only reflects the numerical difference, but also corresponds to the specific change path.
[0036] The time distribution within each change cluster area is refined and organized. The start and end times of each change segment within the area are extracted one by one. These time intervals are arranged in chronological order. The intervals between adjacent change segments are recorded. The time intervals of all change segments are superimposed. It is determined whether each time node is covered by a change segment. Time periods covered by at least one change segment are continuously spliced together. The total length of the spliced time is used as the fluctuation coverage span value. At the same time, the uncovered time intervals are individually marked so that the fluctuation coverage span value can accurately reflect the distribution density of change segments in the time dimension.
[0037] Based on the changes in amplitude diffusion range values over time, the amplitude diffusion range values at different time points within each change cluster area are recorded in segments. The time axis is divided into several continuous intervals, and the amplitude diffusion range values within each interval are compared point by point. Segments where the amplitude diffusion range values increase across multiple consecutive time points are uniformly marked, while segments where the amplitude diffusion range values decrease across multiple consecutive time points are recorded independently. Segments where the amplitude diffusion range values alternate between adjacent time points are categorized and organized. By dividing segments with different change rhythms, the amplitude diffusion behavior within the change cluster area is transformed into a change rhythm expression with a clear temporal order, and different levels of predictive attention are set according to different change rhythms.
[0038] By combining the amplitude diffusion range value and the fluctuation coverage span value, the predicted attention level of each change cluster region is uniformly calibrated. The change rhythm of the amplitude diffusion range value is correlated with the corresponding fluctuation coverage span value. Regions with continuously expanding amplitude diffusion range values and continuously increasing fluctuation coverage span values are assigned a higher level of predicted attention level, while regions with declining amplitude diffusion range values and interrupted fluctuation coverage span values are assigned a lower level of predicted attention level. Furthermore, the change cluster regions are arranged in chronological order, and the relationship between the predicted attention levels of adjacent regions is continuously expressed. This determines a dynamic resource allocation strategy, enabling computing resources to be allocated in an orderly manner among different change cluster regions according to the changes in amplitude diffusion range values and fluctuation coverage span values.
[0039] The trajectory of attention change is extracted according to the dynamic resource tilting strategy. The migration rate and dwell ratio values in the trajectory of attention change are dynamically adjusted. The computational input corresponding to the fallback area is gradually transferred to the expansion area to complete the dynamic allocation of resources in the multi-process prediction process of shield tunneling parameter variable complexity grouping sampling. Given a clearly defined dynamic resource allocation strategy, by refining the trajectory of predicted attention levels over time and adjusting key parameters along the trajectory segment by segment, the orderly migration of computational input across different change zones can be achieved, thus completing the dynamic resource allocation process. The specific steps are as follows: Around the marked change clusters in the dynamic resource allocation strategy, the predicted attention level of each region is extracted point by point within a continuous time range. The attention level values corresponding to each time node are arranged in chronological order, and the changes in attention level between adjacent time nodes are compared one by one. When the attention level changes, the time position of the change and the difference in values before and after the change are recorded. All change nodes are uniformly numbered and arranged in chronological order. The time intervals between adjacent change nodes are connected to construct an attention level change trajectory covering the entire time range, so that the attention level change path of each parameter variable has a continuous expression form throughout the entire time range.
[0040] For the established trajectory of attention change, the time intervals between adjacent change nodes are read segment by segment. Each time interval is recorded in relation to the corresponding magnitude of attention change. The time taken for attention to change from one value to another is statistically analyzed. This time length is paired with the corresponding magnitude of the magnitude of change. Each pairing result is organized to express the magnitude of attention change per unit time, thus forming a migration rate value. All migration rate values are arranged in chronological order, so that the trajectory of attention change has a continuous description of the rate of change in the time dimension.
[0041] The time intervals corresponding to the migration rate values are further divided, and time intervals where the level of attention remains stable are identified. The start and end times of each stable interval are recorded, the duration of the stable interval is statistically analyzed, and this duration is proportionally converted to the entire time range to form the dwell ratio value. At the same time, the migration rate values before and after each stable interval are marked accordingly, so that each time interval contains two types of parameters: migration rate value and dwell ratio value, thus forming a complete trajectory segmentation expression.
[0042] By combining migration rate and dwell rate values, different time segments in the trajectory of attention change are classified. Time segments with continuously increasing migration rate and gradually decreasing dwell rate are defined as expansion regions, while time segments with gradually decreasing migration rate and gradually increasing dwell rate are defined as decline regions. The computational input corresponding to the expansion region is increased segment by segment in chronological order, and the computational input corresponding to the decline region is decreased segment by segment in chronological order. Furthermore, the changes in computational input between adjacent time segments are continuously transitioned to ensure that the change path of computational input remains consistent across different regions.
[0043] For each time segment that has been classified and adjusted, the computational input corresponding to the fallback area is transferred to the expansion area segment by segment in chronological order. The change in computational input corresponding to each time node is recorded. The value of computational input that gradually decreases from the fallback area is allocated to the corresponding time node in the expansion area. The distribution of computational input at each time node is uniformly organized so that the computational input forms a continuous change path throughout the entire time range. This completes the dynamic allocation of resources in the multi-process prediction process of shield tunneling parameter variable complexity grouping sampling, so that the trajectory of computational resource allocation and attention change between different change areas remains consistent.
[0044] This invention jointly characterizes the correspondence between the number of amplitude fluctuations and the duration of the change, and introduces the values of the fallback span and fallback time to identify short-term rise and rapid fall segments. This allows short-term disturbances to be distinguished and processed during the complexity assessment stage, thereby avoiding the continuous judgment of instantaneous fluctuations as high-complexity states. This processing method makes the complexity grouping results more consistent with the actual change process of parameter variables, reduces misjudgments caused by short-term fluctuations, and thus improves the stability and continuity of the overall prediction results.
[0045] This invention characterizes the change clustering region by constructing a complex state retention intensity result and combining the amplitude diffusion range value and the fluctuation coverage span value. It further forms a dynamic resource tilting strategy and adjusts the migration rate value and the dwell ratio value based on the change trajectory of attention, so that the computing input can be transferred in an orderly manner between the fallback region and the expansion region. This method enables the resource allocation process to be adjusted synchronously with the parameter change trend, thereby improving the efficiency of computing resource utilization and maintaining the coordinated operation of the multi-process prediction process.
[0046] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A multi-process prediction method for the complexity of shield tunneling parameter variables through grouped sampling, characterized in that, Includes the following steps: The parameters and variables obtained during the tunnel boring machine are analyzed over a continuous time range. The number of amplitude fluctuations, the difference between peak intervals, and the duration of change are extracted. Based on the corresponding change relationship between the number of amplitude fluctuations and the duration of change, the activity range of variables is divided to obtain the basis for grouping variables by complexity. The variable complexity is grouped according to the corresponding amplitude change process. The fallback span and fallback time values in the amplitude change process are extracted. The segments with short-term rise and rapid fall are identified and the disturbance attenuation characteristics are output. Based on the time range corresponding to the disturbance decay characteristics, the fluctuation dwell time value is extracted, and the fluctuation dwell time value is correlated with the fallback time value to identify the change segment with insufficient continuous expansion capability and generate the complex state retention strength result. For complex states, retain the change clustering area corresponding to the intensity result, statistically analyze the amplitude diffusion range and fluctuation coverage span, adjust the predicted attention level according to the change rhythm of the amplitude diffusion range value, and determine the dynamic resource allocation strategy. The trajectory of attention change is extracted according to the dynamic resource tilt strategy. The migration rate and dwell ratio values in the trajectory of attention change are dynamically adjusted. The computational input corresponding to the fallback area is gradually transferred to the expansion area, and the dynamic allocation of resources is completed in the multi-process prediction process of shield tunneling parameter variable complexity grouping sampling.
2. The method for predicting the complexity of shield tunneling parameter variables through grouped sampling and multi-process prediction according to claim 1, characterized in that, The steps for grouping variables by complexity are as follows: The parameters of advance speed, cutterhead torque, earth pressure and synchronous grouting volume are arranged in the order of collection time. The changes in values at adjacent time nodes are compared, the rising and falling segments are marked, the turning points are counted to form the amplitude fluctuation number, the peak value is recorded and the time interval is calculated to obtain the peak interval difference, and the interval time difference is accumulated to obtain the change duration value. The numerical values of the number of amplitude fluctuations and the numerical values of the duration of change are matched and arranged, and the segments corresponding to different change relationships are classified to obtain the expression form of activity level; By introducing the peak interval difference to compare the classified segments, the time segments are grouped according to the characteristics of the change rhythm, and the range of variable activity is constructed. The activity level intervals corresponding to each time segment are marked and summarized. The activity level intervals are then mapped to parameter variables to complete the classification and grouping, and the grouping basis for variable complexity is obtained.
3. The method for predicting the complexity of shield tunneling parameter variables through grouped sampling and multi-process analysis according to claim 2, characterized in that, For each time segment corresponding to the activity level range of variables, the distribution relationship of the number of amplitude fluctuations, the peak interval difference, and the duration of change in the time sequence is recorded. Time segments with continuous changes in the number of amplitude fluctuations and synchronous changes in the duration of change are marked, and time segments with consistent trends in the peak interval difference are grouped together. The grouping of variable complexity is refined based on the time distribution location.
4. The method for predicting the complexity of shield tunneling parameter variables through grouped sampling and multi-process prediction according to claim 2, characterized in that, The steps for outputting the disturbance attenuation feature are as follows: Based on the grouping of variable complexity, time segments are defined, the trajectory of parameter variable amplitude changes is unfolded point by point, the rising phase and peak position are marked and the time nodes are recorded, the falling path is tracked along the time axis, the difference between the peak and the falling end position is calculated to obtain the falling span value, and the time difference from the peak time node to the stable interval is calculated to obtain the falling time value. The corresponding values of the fallout span and fallout time are organized, and the fallout behaviors in each time segment are arranged in chronological order. The interval between adjacent fallout behaviors is recorded and associated. By referring to the correspondence between the pullback span value and the pullback time value, the connection between the rising stage and the pullback stage in the amplitude change trajectory is identified, and the change segments that rise briefly and then fall are extracted and arranged. The extracted change segments are integrated, and the start time, peak time, and fall-off end time of each segment are recorded and summarized in chronological order to output the disturbance attenuation characteristics.
5. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 4, characterized in that, For short-term rises and rapid falls, the time nodes corresponding to the peak positions and the end time nodes of the fall are continuously recorded. The time intervals between adjacent change segments are associated with the amplitude change paths. The fall span values and fall time values are summarized and organized, and the disturbance attenuation characteristics are output in chronological order.
6. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 4, characterized in that, The steps for generating the strength results for complex states are as follows: Mark time segments according to disturbance attenuation characteristics, unfold the amplitude change trajectory point by point, identify the position of entering the fluctuation range and record the start and end times, calculate the time difference to obtain the fluctuation dwell time value. By combining the fluctuation dwell time value and the fall time value, segment by segment matching is performed, and the corresponding time nodes are aligned and arranged to establish the correspondence between the fluctuation dwell interval and the fall interval. By introducing the correspondence between the fluctuation dwell time value and the fallback time value, the change segment is identified, and the fluctuation dwell interval and fallback interval are combined into a complete change segment and arranged in chronological order. The identified change segments are summarized, and the time nodes of each segment, the corresponding fluctuation dwell time values and the fallback time values are integrated to generate the complex state retention strength results.
7. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 6, characterized in that, By combining the continuous distribution interval of fluctuation dwell time value with the concentrated change segment of fall time value, the change segment immediately following the end of the fluctuation dwell interval is screened and arranged in chronological order. The start time, dwell end time and fall end time of each change segment are recorded uniformly.
8. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 6, is characterized in that, The steps for determining a dynamic resource allocation strategy are as follows: According to the intensity results of the complex state, the change segments are marked, the time interval and amplitude connection of adjacent change segments are read, the segments that meet the continuous change relationship are merged and the time range is recorded to obtain the change cluster area; By combining the amplitude change trajectory within the change cluster area, the amplitude data is read point by point and the maximum and minimum amplitude values are filtered out. The difference is calculated to obtain the amplitude diffusion range value. By introducing the time distribution within the change cluster area, arranging the time intervals of each change segment and statistically analyzing the coverage time range, the value of the fluctuation coverage span is obtained. The amplitude diffusion range is recorded in segments to identify increasing, decreasing, and alternating segments, and to determine the rhythm of change. By combining the numerical change rhythm of amplitude diffusion range with the numerical value of fluctuation coverage span, the predicted attention level corresponding to each change cluster area is adjusted to obtain a dynamic resource allocation strategy.
9. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 8, characterized in that, The increasing change segments in the amplitude diffusion range and the continuously changing segments in the fluctuation coverage span are marked accordingly. The predicted attention level within the corresponding time interval is continuously adjusted. At the same time, the decreasing change segments and the coverage interruption segments are separated and marked. By arranging the change clusters in time sequence, the continuous change processing of the predicted attention level between adjacent areas is completed.
10. The method for predicting the complexity of shield tunneling parameter variables by grouping and sampling in a multi-process manner according to claim 8, characterized in that, The trajectory of attention changes corresponding to the dynamic resource allocation strategy is processed, and dynamic resource allocation is completed by extracting migration rate and dwell ratio values. The specific steps are as follows: Read the corresponding change clustering areas of the dynamic resource tilt strategy in chronological order, extract and arrange the attention level values of each time node, record the change nodes and connect the time intervals to obtain the trajectory of attention level change. By referring to the trajectory of changes in attention, the time intervals and magnitudes of changes of adjacent change nodes are extracted. The time intervals and magnitudes of changes are paired and organized to obtain migration rate values, which are then arranged in chronological order. The time intervals are divided based on the migration rate value, the intervals with stable attention levels are identified and the time lengths are counted, and the time lengths are matched with the overall time range to obtain the dwell ratio value. The migration rate and the dwell ratio are combined for classification, and the expansion area and the fallback area are divided and the input changes are adjusted accordingly. Based on the time distribution of the expansion and fallback regions, the computing resources are transferred and allocated, and the changes at each time node are recorded to complete the dynamic allocation of resources.
