Tunnel large machine matching operation control analysis method based on multi-source data fusion

Abstract

The present application belongs to the technical field of tunnel construction intelligent analysis and management, and relates to a tunnel large machine supporting operation control analysis method based on multi-source data fusion. First, the process duration data of continuous multiple construction cycles of a tunnel operation surface is obtained, and a process tree structure containing longitudinal membership and horizontal connection is constructed. Then, the deviation duration of each component unit in the process tree structure relative to the historical cycle is calculated, and the abnormal source node is identified according to the threshold value performance. Subsequently, the data is processed according to the abnormal source level selection resolution dimension, the abnormal type and confidence are traced back, and the optimization suggestion is generated and pushed according to the rules. Finally, the adoption state is detected in the subsequent cycle and closed-loop feedback is performed. The present application improves the accuracy and interpretability of abnormal source tracing through hierarchical modeling, quantifies the abnormal reasons through multi-dimensional analysis, realizes the iteration of optimization strategies relying on closed-loop verification, and improves the tunnel construction diagnosis efficiency and collaborative operation level.

Description

Technical Field

[0001] This invention belongs to the field of intelligent analysis and management technology for tunnel construction, and relates to a method for controlling and analyzing the operation of large tunnel machinery based on multi-source data fusion. Background Technology

[0002] Tunnel engineering is characterized by dynamic construction. As excavation progresses, the working face continuously advances, and each cycle requires the coordinated completion of multiple sub-processes, such as drilling, charging explosives, ventilation, muck removal, and support. Due to the concealment of underground spaces and the time-varying nature of rock mass conditions, continuously comparing and analyzing the process duration data of the same working face in different cycles is a conventional technical means to assess construction efficiency and identify progress bottlenecks.

[0003] Currently, construction analysis at tunnel working faces typically employs the following methods: On-site technicians manually record or log the start and end times of each major process, and calculate the total time consumed for each process; the monitoring system built into the tunnel boring machine collects operational status data for each individual piece of equipment. During the analysis phase, the total duration of the current cycle is usually compared with the average duration of historical cycles. When the total duration exceeds expectations, engineering technicians, based on experience, systematically investigate any potential contributing factors.

[0004] However, existing technologies have the following shortcomings in practical applications: 1. Existing technologies suffer from insufficient granularity in process data and isolated multi-source data, making it difficult to pinpoint the source of deviations. Current recording methods primarily focus on the total process time without standardizing the hierarchical breakdown within the process. This makes it impossible to further decompose the deviation into its specific sub-step when discrepancies occur in the total time. Furthermore, equipment operating data, surrounding rock parameter data, and process time data belong to different acquisition systems, lacking unified timeline alignment and correlation mapping. This makes it difficult to combine geological characteristic parameters to eliminate the objective impact of rock strata changes on process time, resulting in a lack of accurate data support for identifying the causes of anomalies.

[0005] 2. Existing technologies lack multi-cycle benchmark construction and root cause tracing mechanisms. They typically employ simple arithmetic averages or extreme value comparisons, without establishing a time-weighted multi-cycle benchmark extraction method, making it difficult to effectively identify trend deviations in process duration. When anomalies occur, there is a lack of means to match anomaly types and calculate confidence levels based on multi-dimensional features, leading to improvement suggestions relying heavily on empirical judgments and failing to conduct closed-loop verification of optimization effects through data feedback from subsequent cycles. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a method for control and analysis of tunnel machinery operation based on multi-source data fusion is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: This invention provides a method for controlling and analyzing the operation of tunnel machinery based on multi-source data fusion, including: acquiring the process duration data of multiple consecutive construction cycles at the tunnel working face, decomposing them into process nodes, sub-process nodes, equipment condition nodes and execution action nodes according to a preset hierarchy, and constructing a process tree structure through the vertical subordinate relationship and horizontal connection relationship between nodes.

[0008] The deviation duration of each component unit within the process tree structure in the current construction cycle relative to multiple historical construction cycles is calculated. The component unit includes hierarchical nodes and sub-process connection items. The abnormal source node is identified based on the deviation duration exceeding the threshold.

[0009] Based on the level to which the anomaly source node belongs, the corresponding device parsing dimension is selected to process the original running data fragment, and the current anomaly type and confidence level are traced. The device parsing dimension includes instruction response dimension, parameter association dimension, or action dynamic dimension.

[0010] Based on the preset association rules between anomaly type and optimization suggestions, targeted optimization suggestions are generated and pushed to the execution end.

[0011] In subsequent construction cycles, the adoption status of optimization suggestions is monitored, and closed-loop feedback is implemented based on the actual changes in the duration of abnormal source nodes.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention constructs a process tree structure containing vertical subordinate relationships and horizontal connection relationships, decomposes the total process time into sub-processes, equipment conditions and execution action nodes, and combines bottom-up node traversal with parent-child node coupling verification to gradually narrow the tracing range of the total time deviation, and finally locates the abnormal source to a specific level node, thereby improving the accuracy and interpretability of abnormal tracing.

[0013] (2) This invention corrects the deviation duration by introducing geological feature vectors, thereby eliminating the interference of rock strata changes on process duration analysis. Furthermore, for anomaly source nodes at different levels, analysis is performed using process time sequence logic comparison, physical parameter mapping relationship analysis, or action fluctuation feature extraction. The confidence level of the anomaly type is calculated by combining Bayesian probabilistic inference, thereby achieving quantitative identification of the cause of the anomaly and providing a basis for the generation of subsequent optimization suggestions.

[0014] (3) This invention detects the adoption status of optimization suggestions in subsequent construction cycles and compares the changes in deviation duration after adoption with historical benchmarks. Based on the comparison results, effective optimizations are stored in the case library, or alternative suggestions are generated when optimizations are ineffective, forming a complete closed loop from anomaly identification, suggestion push to effect verification, and realizing continuous iteration of optimization strategies. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the implementation steps of the method of the present invention.

[0017] Figure 2 This is a diagram illustrating the drilling process as an example in the process tree structure of this invention.

[0018] Figure 3 This is a logical diagram illustrating the present invention for identifying abnormal source nodes based on the deviation duration exceeding a threshold. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, the present invention provides a tunnel machinery supporting operation control and analysis method based on multi-source data fusion, including: S1. Obtaining the process duration data of multiple consecutive construction cycles at the tunnel working face, decomposing them into process nodes, sub-process nodes, equipment condition nodes and execution action nodes according to a preset hierarchy, and constructing a process tree structure through the vertical subordinate relationship and horizontal connection relationship between nodes.

[0021] In this embodiment, constructing the process tree structure includes the following steps: S11. Using sensor timestamp information, identify the start and end times of processes, sub-processes, equipment conditions, and executed actions, and calculate the corresponding duration.

[0022] S12. Using the process as the root node, and sub-processes, equipment conditions, and execution actions as branch nodes from top to bottom, encapsulate the corresponding duration into the corresponding node.

[0023] A process refers to a phased task in the construction process, including but not limited to drilling, charging, blasting, and slag removal. A sub-process refers to the process items required to complete a process. Equipment operating condition refers to the operating mode performed by the equipment under a certain sub-process. Execution action refers to the smallest identifiable action unit of the equipment when executing a certain operating mode.

[0024] Reference Figure 2 As shown, taking the drilling process as an example, it can be decomposed into three consecutive sub-processes: drilling rig positioning, drilling execution, and drilling rig relocation. Drilling rig positioning corresponds to the walking and positioning mode of the rock drilling rig. The decomposed execution actions are, in sequence, rig travel, boom extension, outrigger fixation, and rough positioning of the drill arm.

[0025] The drilling process follows the drilling mode of the corresponding rock drilling rig, and the execution actions are broken down into the following sequence: drill arm advance, drill bit rotation, advance and rotation coordination, and drill bit retraction.

[0026] The drilling rig relocation corresponds to the relocation and retraction mode of the rock drilling rig, which can be further broken down into the following actions: outrigger retraction, boom folding, and rig travel.

[0027] S13. Establish horizontal timing relationships only between sub-process nodes and mark the connection duration.

[0028] S14. Construct vertical membership constraints to ensure that the duration of a parent node is equal to the sum of the durations of all its child nodes and the connection durations between child nodes.

[0029] S15. Connect multiple single-cycle tree structures corresponding to continuous construction cycles according to time sequence to form a process tree structure.

[0030] It should be specifically explained that the process tree structure proposed in this invention is constructed independently for each process stage. That is, a dedicated tree structure is formed for each individual process, and the tree structures of different processes are independent of each other.

[0031] S2. Calculate the deviation duration of each component unit in the process tree structure in the current construction cycle relative to multiple historical construction cycles. The component unit includes hierarchical nodes and sub-process connection items. Identify abnormal source nodes based on the deviation duration exceeding the threshold.

[0032] In this embodiment, hierarchical nodes and sub-process connection items have different physical meanings and data characteristics in the process tree structure. The duration of hierarchical nodes reflects the duration of the interaction process between the equipment and the surrounding rock. Affected by geological conditions and the condition of the equipment itself, the data shows a slow drifting trend over time. The duration of sub-process connection items reflects the waiting, preparation, or idle time between processes. It is mainly affected by immediate factors such as spatial resource conflicts and scheduling command responses. The data fluctuates randomly around a stable value without obvious trend changes. Therefore, it is necessary to handle different cases when calculating the deviation duration.

[0033] Specifically, the following implementation is applied to hierarchical nodes: Reference weights are assigned to the duration data of the same hierarchical node in each historical construction cycle according to the time interval between the current construction cycle and the current construction cycle. In this embodiment, to reflect that the early equipment status has higher reference value for building a health benchmark, the principle of assigning greater weights to construction cycles with larger time intervals from the current construction cycle is followed. Specifically, the ratio is calculated by using the time interval between each historical construction cycle and the current construction cycle as the numerator and the sum of the time intervals between all historical construction cycles and the current construction cycle as the denominator; the resulting ratio is the assigned reference weight.

[0034] By using weighted average calculation, the baseline reference duration of each level node in multiple historical construction cycles is obtained.

[0035] Subtract the corresponding benchmark duration from the duration of each node at each level of the current construction cycle to obtain the original deviation duration of the current construction cycle relative to multiple historical construction cycles.

[0036] The thrust, torque, and penetration parameters of the tunneling equipment are obtained to construct a geological feature vector.

[0037] Based on the differences in geological feature vectors between the current construction cycle and multiple historical construction cycles, the permissible duration variation of each level node under the influence of geological fluctuation factors is determined. Specifically, the geological feature vectors of the level nodes in multiple historical construction cycles and the corresponding node durations are retrieved to construct a historical instance set.

[0038] Using geological feature vectors as independent variables and node duration as dependent variables, a multiple linear regression is performed on the historical instance set to obtain a regression coefficient vector. The regression coefficient vector reflects the degree of influence of each geological parameter on the unit duration.

[0039] Calculate the arithmetic mean of the durations of all nodes in the historical instance set, and use it as the historical average duration.

[0040] Calculate the arithmetic mean of all historical geological feature vectors in the historical instance set, and use it as the historical average geological vector.

[0041] Subtracting the current geological feature vector from the historical average geological vector yields the geological deviation vector, which includes thrust deviation, torque deviation, and penetration deviation, along with their positive and negative directions.

[0042] By performing a dot product operation between the geological deviation vector and the regression coefficient vector, the permissible duration variation of the hierarchical nodes under the influence of geological fluctuation factors is obtained.

[0043] Subtracting the allowable duration variation from the original deviation duration yields the deviation duration of each node after eliminating the interference of geological fluctuation factors.

[0044] Specifically, the following measures are implemented for sub-process connection items: the connection time between sub-process nodes at the same location in each historical construction cycle is statistically analyzed, and their arithmetic mean is calculated as the historical benchmark reference time for sub-process connection items.

[0045] The deviation time of the sub-process connection item is obtained by subtracting the historical benchmark reference time from the actual connection time between the corresponding sub-process nodes in the current construction cycle.

[0046] Based on this, refer to Figure 3 As shown, the abnormal source node is identified based on the deviation duration of each component unit exceeding the threshold. Specifically, this includes: setting a warning threshold for the deviation duration of each component unit, and marking the component unit whose deviation duration exceeds the corresponding warning threshold as being in a deviation state.

[0047] Traverse the process tree structure from bottom to top, prioritizing the detection of deviations at the execution action level.

[0048] If there are nodes in a deviated state at the execution action level, they are marked as candidate abnormal source nodes; otherwise, the deviated state at the equipment condition level and sub-process level is checked level by level upwards.

[0049] A parent-child node coupling check is performed on candidate anomaly source nodes: if a parent node deviates from all its child nodes, and the sum of the absolute values ​​of the child node deviations covers the absolute value of the parent node deviation, then the parent node is removed, and the lowest-level child node is retained as the anomaly source node. This check mechanism ensures that the anomaly source is located at the action execution or equipment operating condition level to reflect anomalies in the equipment's own operating efficiency.

[0050] If a sub-process connection item is in a deviated state, then the sub-process node connected to that connection item that is later in the sequence is marked as an abnormal source node.

[0051] It should be noted that the specific process of setting the warning threshold for the deviation duration of each component unit is as follows: For the execution action node, select historical data within the first maintenance cycle after the new equipment is put into operation, extract the duration of all normal construction cycles for each execution action, and calculate the mode as the ideal reference duration for this action; then calculate the difference between the actual duration of the execution action in each cycle and the ideal reference duration to obtain a set of deviation samples, and take the maximum positive value of the set of deviation samples as the warning threshold for the deviation duration.

[0052] For a device operating condition node, the warning threshold for its deviation duration is jointly determined by the deviation thresholds of all the execution action nodes it contains. Specifically, considering that the deviations of each execution action may be superimposed or canceled out, the deviation thresholds of all sub-execution action nodes can be summed to obtain the warning threshold for the deviation duration of the device operating condition node.

[0053] For sub-process connection items, extract the standard interval time between processes specified in the construction organization design document and take 10% of this time as the warning threshold for deviation time.

[0054] S3. Based on the level to which the anomaly source node belongs, select the corresponding device parsing dimension to process the original running data fragment, trace the current anomaly type and confidence level, and the device parsing dimension includes instruction response dimension, parameter association dimension or action dynamic dimension.

[0055] (1) When the abnormal source node belongs to the sub-process level, select the instruction response dimension to process the original running data fragment, specifically including: tracing down the first equipment condition node and the first execution action node for the abnormal source node in the process tree structure.

[0056] The equipment response data is formed by extracting the start time of the equipment condition node, the sequence of operating parameters of the first execution action node, and the equipment standby power consumption data within the time period corresponding to the sub-process connection item.

[0057] The equipment response data is compared with the equipment design operating condition response standard, and the abnormal response mode is identified through the following steps: Start-up delay identification: Obtain the end time of the sub-process connection item as the planned start time of the next sub-process; calculate the time difference between the start time of the equipment operating condition node and the planned start time as the start-up delay duration; if the start-up delay duration is greater than zero, it is determined that there is an abnormal start-up delay mode, and the start-up delay duration is used as the characteristic parameter of this mode.

[0058] Standby Status Anomaly Identification: During the connection period corresponding to the sub-process connection item, the average standby power consumption in the equipment's standby power consumption data is calculated. If the average standby power consumption exceeds the product of the standard standby power consumption and a preset tolerance coefficient, an abnormal standby power consumption is determined. The preset tolerance coefficient is used to allow normal power consumption fluctuations of the equipment in standby mode, avoiding false alarms caused by sensor noise, changes in ambient temperature, or slight load fluctuations. An exemplary value of 1.2 can be used. The standard standby power consumption is obtained by consulting the technical documents provided by the equipment manufacturer, which show the rated power consumption value of the equipment in standby mode.

[0059] The value of the average standby power consumption exceeding the product of the standard standby power consumption and the preset tolerance coefficient is used as the characteristic parameter of the abnormal standby mode.

[0060] Operating condition switching lag identification: Inflection point detection is performed on the operating parameter sequence of the first execution action node, and the interval formed by multiple consecutive preset sampling points of parameter change rate being lower than a preset change rate threshold is identified as a stagnation interval; the cumulative duration of the stagnation interval is counted as the lag duration; if the lag duration is greater than zero, it is determined that there is an abnormal operating condition switching lag mode, and the lag duration is used as a characteristic parameter of the operating condition switching lag mode; wherein, the change rate threshold is set according to the lower limit of the normal change rate of parameters specified in the equipment design operating condition response standard, and can be exemplarily taken as 5MPa / s.

[0061] Action execution failure identification: Peak detection is performed on the operating parameter sequence of the first action node. The number of peaks where parameter fluctuations exceed a preset fluctuation range during startup is counted as the number of attempts. If the number of attempts is greater than 1, an abnormal action execution failure mode is determined to exist, and the number of attempts is used as a characteristic parameter of the action execution failure mode. The fluctuation range threshold is set according to the upper limit of the normal fluctuation range of parameters specified in the equipment design operating condition response standard, and can be exemplarily set to 2 MPa.

[0062] Based on the identified abnormal response patterns, the corresponding response feature parameters are extracted to form an instruction response feature vector.

[0063] It should be noted that the above response anomaly patterns correspond to the anomaly types traced under the instruction response dimension.

[0064] (2) When the abnormal source node belongs to the equipment operating condition level, select the parameter association dimension to process the original operating data segment, including: extracting the synchronous time sequence data of at least one pair of operating parameters with physical association within the time period corresponding to the abnormal source node.

[0065] A sliding window is used to segment the synchronous time series data, and the least squares method is used to calculate the linear fitting residual sequence and goodness of fit between pairs of operating parameters within each window.

[0066] The goodness-of-fit scores of each window are arranged in descending order of numerical value, and windows at the tail end of the goodness-of-fit distribution are selected as suspected abnormal windows. The preset proportion is between 5% and 10%, with a default value of 5%.

[0067] For each suspected abnormal window, the mean, standard deviation, and autocorrelation coefficient of the linearly fitted residual sequence are extracted to form a parameter-related feature vector.

[0068] It should be noted that the anomaly types traced under the parameter correlation dimension include physical mapping relationship failure, system response lag, operating condition instability, and operating point drift. The parameter correlation feature vectors corresponding to different anomaly types are shown in Table 1 below. Data preparation: The mean, standard deviation, and average value of the linear fitting residual sequences of all non-suspected anomaly windows are calculated respectively, and used as reference mean, reference standard deviation, and reference autocorrelation coefficient.

[0069] Table 1. Explanation of the performance of different anomaly types under parameter correlation dimensions.

[0070]

[0071] (3) When the abnormal source node belongs to the execution action level, select the action dynamic dimension to process the original running data segment, including: extracting the time series data of at least one running parameter within the time period corresponding to the abnormal source node.

[0072] A sliding window is used to segment the time-series data of the operating parameters, and time-domain statistical features are calculated in each window. The time-domain statistical features include mean, variance, peak factor and waveform factor.

[0073] A fast Fourier transform is performed on the time-series data of each window to extract frequency domain features, which include the main frequency amplitude and the energy proportion of a specified frequency band.

[0074] Cluster analysis is performed on the time-domain statistical features and frequency-domain features of each window to screen out windows that are far from the mainstream cluster center as suspected abnormal windows. The specific process is as follows: the time-domain statistical features and frequency-domain features of each window are combined to form the multi-dimensional feature vector of the window. The feature vectors of all windows constitute the set of sample points in the feature space.

[0075] Density clustering algorithm is used to cluster sample points in the feature space, and the cluster with the densest distribution of sample points is identified as the mainstream cluster, with the centroid of the mainstream cluster serving as the mainstream cluster center. The mainstream cluster corresponds to the feature distribution range of the executed action under normal conditions.

[0076] Calculate the Euclidean distance from each sample point to the mainstream cluster center to obtain the distance set.

[0077] Calculate the mean and standard deviation of the distance set, and set a distance threshold by the sum of the mean and a preset multiple of the standard deviation. The preset multiple ranges from 2 to 3. Mark the windows corresponding to sample points whose Euclidean distance is greater than the distance threshold as suspected abnormal windows.

[0078] In addition, if no window in the current batch is marked as suspected abnormal, the preset multiplier value will be automatically reduced by a fixed step size (e.g., 0.1) and re-evaluated until at least one suspected abnormal window is detected or the preset multiplier value drops to 1.5. If the number of suspected abnormal windows in the current batch exceeds 10% of the total number of windows, the preset multiplier value will be automatically increased by a fixed step size and re-evaluated until the proportion of suspected abnormal windows drops to less than 10% or the preset multiplier value rises to 3.5.

[0079] For each suspected abnormal window, its time-domain statistical features and frequency-domain features are combined to form a dynamic feature vector of the action.

[0080] It should be noted that the abnormality types traced under the dynamic dimension of action include high-frequency fluctuation abnormality, low-frequency fluctuation abnormality, initiation oscillation, slow response and multiple attempts. The dynamic feature vectors corresponding to different abnormality types are shown in Table 2 below: Data preparation: The average values ​​of the time domain statistical features and frequency domain features of all non-suspected abnormality windows are calculated as reference data. In the following table, the increase and rise are relative to the reference values.

[0081] Table 2. Explanation of Different Abnormal Types under the Dynamic Dimension of Action

[0082]

[0083] In this embodiment, tracing the current anomaly type and confidence level includes: matching the feature vector generated under the parsing dimension corresponding to the anomaly source node with the sample feature vector in the historical anomaly case library based on similarity. Specifically, the matching score can be obtained by calculating the cosine similarity between the two vectors.

[0084] Mark all anomaly types in the historical anomaly case library that are at the same level as the current anomaly source node as candidate anomaly types.

[0085] The occurrence frequency of each candidate anomaly type in the historical anomaly case database is statistically analyzed. The prior probability of each candidate anomaly type is quantified by the ratio of the occurrence frequency of a single candidate anomaly type to the sum of the occurrence frequencies of all candidate anomaly types. If the number of samples in the historical anomaly case database is less than 100, a uniform distribution method is adopted, and the prior probability value of each candidate anomaly type is assigned as the reciprocal of the total number of candidate anomaly types.

[0086] Extract the feature vectors of all samples belonging to a certain candidate anomaly type from the historical case library, and match them with the current feature vector. Select the largest matching score as the likelihood that the current feature vector belongs to that candidate anomaly type. Use this as the output of the constructed likelihood function to obtain the likelihood that the current feature vector belongs to each candidate anomaly type.

[0087] The unnormalized posterior probability is obtained by multiplying the prior probability of each candidate anomaly type by the likelihood of the current feature vector under that type.

[0088] The normalization factor is obtained by summing the unnormalized posterior probabilities of all candidate anomaly types. According to Bayes' theorem, the posterior probability of each candidate anomaly type is equal to its unnormalized posterior probability divided by the normalization factor, thus realizing the quantitative calculation of the posterior probability.

[0089] The candidate anomaly type corresponding to the maximum posterior probability is taken as the current anomaly type, and the posterior probability is taken as the confidence level of the anomaly type.

[0090] The corresponding optimization suggestion generation process is triggered only when the confidence level exceeds a preset threshold. The preset threshold is set to 0.7, and on-site personnel can adjust it within the range of 0.5-0.9 through the human-computer interaction interface.

[0091] S4. Based on the preset association rules between the anomaly type and the optimization suggestions, generate targeted optimization suggestions and push them to the execution end.

[0092] It should be noted that the preset association rules refer to the pre-established mapping relationship between anomaly types and optimization suggestions. This mapping relationship is based on the following information formulated in advance by professional technicians: I. Equipment design principles: consult the technical manuals and design specifications provided by the equipment manufacturer to obtain the standard response characteristics and allowable fluctuation range of each operating condition and each execution action of the equipment.

[0093] II. Maintenance Manual: Refer to the equipment maintenance manual for the correspondence between common fault phenomena, possible causes, and solutions.

[0094] III. On-site Engineering Experience: Summarize the fault diagnosis and handling experience accumulated by front-line operators, maintenance personnel, and construction management personnel in practice, and formulate standardized recommendations through expert review.

[0095] The above rules, after being organized, form a structured data table, which is stored in the cloud database for direct system access. It covers various anomalies under the dimensions of instruction response, parameter association, and action dynamics.

[0096] For example, if the anomaly type is startup delay under the command response dimension, the corresponding optimization suggestion is to check whether the device positioning sensor signal is normal and to review the automatic positioning program parameters.

[0097] If the anomaly type is system response lag under the parameter correlation dimension, the corresponding optimization suggestions are to check the response characteristics of the pilot valve or proportional valve, optimize the controller PID parameters, clean or replace the filter element, and check the hydraulic oil viscosity.

[0098] S5. In subsequent construction cycles, monitor the adoption status of optimization suggestions and provide closed-loop feedback based on the actual changes in the duration of abnormal source nodes.

[0099] Specifically, in the next construction cycle, sensor signals and equipment operation logs are used to detect whether the operations or parameters involved in the optimization suggestions have been adjusted. If they have been adjusted, they are marked as adopted.

[0100] After the loop ends, the deviation duration of the abnormal source node is recalculated and compared with the historical benchmark. If the deviation duration returns to the normal range, the optimization is deemed effective and the optimization record is stored in the effective case library.

[0101] If the deviation duration is not recovered and the suggestion is not adopted, the correctness of the original optimization suggestion will be re-verified based on the data of the new construction cycle: if the verification is correct, the original optimization suggestion will be pushed again and the operator will be prompted for confirmation; if the verification is incorrect, the alternative suggestion generation mechanism will be triggered. Here, "correct verification" means that the optimization suggestion generated after re-verification is the same as the original optimization suggestion.

[0102] If the same optimization suggestion is pushed to the same abnormal source node more than a preset number of times (e.g., 3 times) and is not adopted, it will no longer be pushed repeatedly. Instead, an advanced warning will be triggered, marking the problem as a persistent unresponsive issue on the operating end and reporting it to the management level for manual intervention and coordination to break the potential vicious cycle.

[0103] If the deviation duration is not recovered but the suggestion has been adopted, the current optimization is marked as invalid, triggering the alternative suggestion generation mechanism. The mechanism includes: re-checking the status of the current abnormal source node to confirm whether the abnormality still exists; if the abnormality still exists, selecting the second-best matching suggestion for the same abnormality type from the preset association rules between the abnormality type and the optimization suggestion and pushing it; if there are no other alternative suggestions for the abnormality type, or the number of invalid optimizations accumulates to 3, the equipment deep inspection warning or benchmark self-calibration process is automatically triggered.

[0104] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for controlling and analyzing the operation of large tunnel machinery based on multi-source data fusion, characterized in that, include: The process duration data of multiple consecutive construction cycles at the tunnel working face is obtained and decomposed into process nodes, sub-process nodes, equipment condition nodes and execution action nodes according to the preset hierarchy. A process tree structure is constructed through the vertical subordinate relationship and horizontal connection relationship between nodes. The deviation duration of each component unit within the process tree structure in the current construction cycle relative to multiple historical construction cycles is calculated. The component unit includes hierarchical nodes and sub-process connection items. Abnormal source nodes are identified based on the deviation duration exceeding the threshold. Based on the level to which the anomaly source node belongs, the corresponding device parsing dimension is selected to process the original running data fragment, and the current anomaly type and confidence level are traced. The device parsing dimension includes instruction response dimension, parameter association dimension or action dynamic dimension. Based on the preset association rules between exception types and optimization suggestions, targeted optimization suggestions are generated and pushed to the execution end; In subsequent construction cycles, the adoption status of optimization suggestions is monitored, and closed-loop feedback is carried out in combination with the actual changes in the duration of abnormal source nodes. When the abnormal source node belongs to the sub-process level, the original runtime data fragment is processed according to the instruction response dimension, including: Within the process tree structure, trace down to the first equipment condition node and the first action execution node for the abnormal source node; Extract the start time of the equipment operating condition node, the sequence of operating parameters of the first execution action node, and the equipment standby power consumption within the time period corresponding to the sub-process connection item to form equipment response data; By comparing with the equipment design operating condition response standard, at least one abnormal response mode is identified, including startup delay, abnormal standby state, stuck operating condition switching, or unsmooth action execution. Based on the identified abnormal response patterns, the corresponding response feature parameters are extracted to form an instruction response feature vector. When the abnormal source node belongs to the device operating condition level, select the parameter association dimension to process the original running data fragment, including: Extract the synchronization time-series data of at least one pair of operating parameters with physical correlation within the time period corresponding to the abnormal source node; A sliding window is used to segment the synchronous time series data, and the linear fit residual sequence and goodness of fit between pairs of running parameters are calculated in each window; The goodness of fit of each window is arranged in descending order of value, and the windows with a predetermined proportion of goodness of fit at the tail end are selected as suspected abnormal windows. For each suspected abnormal window, the mean, standard deviation, and autocorrelation coefficient of the linear fitting residual sequence are extracted to form a parameter-related feature vector; When the execution action level belongs to the abnormal source node, the original runtime data fragment is processed using the dynamic dimension of the action, including: Extract time-series data of at least one operating parameter within the time period corresponding to the anomaly source node; A sliding window is used to segment the time-series data of the operating parameters, and time-domain statistical features are calculated in each window. The time-domain statistical features include mean, variance, peak factor and waveform factor. A fast Fourier transform is performed on the time-series data of each window to extract frequency domain features, which include the main frequency amplitude and the energy proportion of a specified frequency band. Cluster analysis was performed on the time-domain statistical characteristics and frequency-domain characteristics of each window to screen out windows that are far from the mainstream cluster centers as suspected abnormal windows; For each suspected abnormal window, its time-domain statistical features and frequency-domain features are combined to form a dynamic feature vector of the action.

2. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 1, characterized in that, The tree-like structure of the construction process includes: Using sensor timestamp information, the start and end times of processes, sub-processes, equipment conditions, and executed actions are identified, and the corresponding durations are calculated. With the process as the root node, and sub-processes, equipment conditions, and execution actions as the various branch nodes from top to bottom, the corresponding duration is encapsulated into the corresponding node; Establish horizontal temporal relationships only between sub-process nodes and mark the connection duration; Construct vertical membership constraints to ensure that the duration of a parent node is equal to the sum of the durations of all its child nodes and the connection durations between child nodes; Multiple single-cycle tree structures corresponding to continuous construction cycles are connected in time sequence to form a process tree structure.

3. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 2, characterized in that, The calculation process tree structure specifies the deviation time of each component unit in the current construction cycle relative to multiple historical construction cycles, specifically including the following for hierarchical nodes: Based on the time interval with the current construction cycle, reference weights are assigned to the duration data of the same level node in each historical construction cycle. By weighted average calculation, the baseline reference duration of each level node in multiple historical construction cycles is obtained. Subtract the corresponding benchmark reference duration from the duration of each node at each level of the current construction cycle to obtain the original deviation duration of the current construction cycle relative to multiple historical construction cycles. Obtain the thrust, torque, and penetration parameters of the tunneling equipment to construct a geological feature vector; Based on the differences in geological feature vectors between the current construction cycle and multiple historical construction cycles, the permissible duration variation of each level node under the influence of geological fluctuation factors is determined. Subtracting the allowable duration variation from the original deviation duration yields the deviation duration of each node after eliminating the interference of geological fluctuation factors.

4. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 2, characterized in that, The calculation process tree structure specifies the deviation time of each component unit in the current construction cycle relative to multiple historical construction cycles, specifically including the sub-process connection items: The connection time between sub-process nodes at the same location in each historical construction cycle is statistically analyzed, and their arithmetic mean is calculated as the historical benchmark reference time for sub-process connection items. The deviation time of the sub-process connection item is obtained by subtracting the historical benchmark reference time from the actual connection time between the corresponding sub-process nodes in the current construction cycle.

5. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 1, characterized in that, The method of identifying abnormal source nodes based on the threshold performance of deviation duration includes: Preset warning thresholds for the duration of deviation for each component, and mark the component whose deviation duration exceeds the corresponding warning threshold as being in a deviation state; Traverse the process tree structure from bottom to top, and prioritize detecting deviations in the execution action level; If there are nodes in a deviated state at the action execution level, they are marked as candidate abnormal source nodes; otherwise, the deviation state at the equipment condition level and sub-process level is checked level by level upwards. Perform parent-child node coupling verification on candidate anomaly source nodes, and retain the lowest-level child node as the anomaly source node; If a sub-process connection item is in a deviated state, then the sub-process node connected to that connection item that is later in the sequence is marked as an abnormal source node.

6. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 1, characterized in that, The process of tracing the current anomaly type and confidence level includes: The feature vector generated under the corresponding parsing dimension of the abnormal source node is matched with the sample feature vector in the historical abnormal case library to obtain the matching score; Based on the matching scores between the current feature vector and the historical samples of each candidate anomaly type, a likelihood function is constructed, and combined with the prior probabilities of each candidate anomaly type obtained from the historical anomaly case library, the posterior probability of each candidate anomaly type is calculated using Bayes' theorem. The candidate anomaly type corresponding to the maximum posterior probability is taken as the current anomaly type, and the posterior probability is taken as the confidence level of the anomaly type. The process of generating corresponding optimization suggestions is triggered only when the confidence level exceeds a preset threshold.

7. The tunnel machinery operation control and analysis method based on multi-source data fusion according to claim 1, characterized in that, The process of detecting the adoption status of optimization suggestions in subsequent construction cycles, and combining this with closed-loop feedback based on actual changes in the duration of abnormal source nodes, includes: In the next construction cycle, sensor signals and equipment condition logs are used to detect whether the operations or parameters involved in the optimization suggestions have been adjusted. If they have been adjusted, they are marked as adopted. After the loop ends, the deviation duration of the abnormal source node is recalculated. If the deviation duration returns to the normal range, the optimization is deemed effective, and the optimization record is stored in the effective case library. If the deviation duration is not recovered and the suggestion is not adopted, the correctness of the original optimization suggestion will be re-verified based on the data of the new construction cycle. If the deviation duration is not recovered but the suggestion has been adopted, then mark this optimization as invalid and trigger the alternative suggestion generation mechanism.

Images