A method and system for diagnosing construction progress deviation based on cluster analysis

Abstract

This invention belongs to the field of engineering progress diagnosis technology, specifically disclosing a method and system for diagnosing engineering construction progress deviations based on cluster analysis. The method includes: slicing the original progress data of the target project into time-series states to obtain a continuous progress state sequence; measuring the similarity of the evolution trajectory of the continuous progress state sequence and performing topological mapping on the measurement results to obtain a trajectory similarity matrix; based on the trajectory similarity matrix, performing trajectory pattern clustering on the continuous progress state sequence to obtain progress state trajectory clusters; locating key turning points in the evolution trajectory to obtain state transition points; performing critical condition backtracking on the state transition points to obtain preconditions, and performing correlation mining on the preconditions to obtain common conditions; and based on the common conditions, performing predictive risk analysis on the target project to obtain deviation diagnosis results. This invention can improve diagnostic efficiency and decision support capabilities.

Description

Technical Field

[0001] This invention belongs to the field of engineering progress diagnosis technology, and in particular relates to a method and system for diagnosing engineering construction progress deviations based on cluster analysis. Background Technology

[0002] Existing engineering schedule deviation diagnosis technologies lack effective time-series structuring methods for processing raw schedule data, failing to fully analyze the dependencies and temporal relationships between data, resulting in messy data formats and insufficient extraction of effective information. Furthermore, the similarity measurement of evolution trajectories does not integrate multi-scale morphological features of the data and task risk attributes, performing only simple surface comparisons, leading to insufficient accuracy in similarity assessment results and an inability to accurately capture the inherent patterns and differences in schedule changes.

[0003] In the trajectory clustering and deviation tracing stages, existing technologies fail to optimize clustering logic by incorporating the characteristics of different construction phases and the inherent risk weights of tasks. This results in trajectory cluster division lacking specificity and failing to adapt to the progress characteristics of different construction phases. The identification of key turning points lacks a deep connection with the construction plan, and the mining of preconditions for deviation causes lacks structured analysis of event chains and extraction of common patterns. Consequently, deviation tracing is inaccurate and incomplete, ultimately leading to low efficiency in progress deviation diagnosis and an inability to effectively predict and prevent potential risks. Therefore, improving the efficiency of a clustering-based approach to diagnose engineering construction progress deviations has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for diagnosing engineering construction progress deviations based on cluster analysis. By using time-series state slicing, trajectory cluster analysis, and mining of preconditions for state transition points, and integrating task risk attributes, it achieves accurate source tracing and predictive risk analysis of engineering progress deviations, thereby improving diagnostic efficiency and decision support capabilities.

[0005] To achieve the above objectives, this invention provides a method for diagnosing engineering construction schedule deviations based on cluster analysis, comprising the following steps: S01. Perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project; S02. Perform similarity measurement on the evolution trajectory of the continuous progress state sequence, and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. S03. Based on the trajectory similarity matrix, the continuous progress state sequence is divided into trajectory pattern clusters to obtain the progress state trajectory clusters of the continuous progress state sequence. S04. Based on the progress state trajectory cluster and the construction plan of the target project, locate the key turning points of the evolution trajectory to obtain the state transition points of the evolution trajectory. S05. Perform critical condition backtracking on the state transition points to obtain the preconditions of the state transition points, and perform correlation mining on the preconditions to obtain the common conditions of the state transition points. S06. Based on common conditions, conduct predictive risk analysis on the target project to obtain deviation diagnosis results for the target project.

[0006] Preferably, in step S01, the process of obtaining the continuous progress sequence of the target project includes: Dependency analysis is performed on the original schedule data of the target project to obtain the event logic diagram of the original schedule data; Based on the event logic diagram, the original progress data is timestamped to obtain the time-series baseline event flow of the original progress data of the target project; The time-series reference event stream is divided into non-uniform density segments to obtain adaptive time segments of the unified time-series reference event stream; By performing cross-segment correlation and integration on adaptive time segments, a continuous progress status sequence of the target project is obtained.

[0007] Preferably, in step S02, the process of obtaining the trajectory similarity matrix of the continuous progress state sequence includes: Multi-scale morphological waveform decomposition is performed on the continuous progress state sequence to obtain the morphological components of the continuous progress state sequence. Based on morphological components, adaptive time warp alignment is performed on the evolution trajectory of a continuous progress state sequence to obtain the alignment path of the evolution trajectory. Based on the alignment path, the evolutionary trajectory is compared point by point to obtain the morphological difference sequence of the evolutionary trajectory; By weighted fusion of morphological difference sequences, the comprehensive morphological distance of the evolutionary trajectory is obtained; Based on the comprehensive morphological distance, a nonlinear dimension reduction projection is performed on the evolution trajectory to obtain the low-dimensional manifold coordinates of the evolution trajectory; Based on low-dimensional manifold coordinates, the comprehensive morphological distance is normalized and reconstructed to obtain the trajectory similarity matrix of the continuous progress state sequence.

[0008] Preferably, in step S03, the process of obtaining the progress state trajectory cluster of the continuous progress state sequence includes: A topological vulnerability analysis is performed on the task dependencies of the target project to obtain the inherent risk weights of the nodes in the task dependencies. Based on the inherent risk weights of nodes, a perception enhancement transformation is performed on the trajectory similarity matrix to obtain the corrected similarity matrix of the continuous progress state sequence. Based on the construction plan of the target project, the modified similarity matrix is ​​reorganized into a phase-consistent form, and the reorganized matrix is ​​then subjected to adaptive sub-block segmentation to obtain the intra-phase similarity sub-blocks of the target project. Based on the intra-stage similarity sub-blocks, the stage progress state sequence of the continuous progress state sequence is mapped in the feature space to obtain the feature coordinates of the stage progress state sequence. Density peak aggregation is performed on the feature coordinates to obtain the intra-stage trajectory clusters of the stage progress state sequence; Cross-stage pattern linking is performed on the trajectory clusters within a stage to obtain a progress state trajectory cluster of a continuous progress state sequence.

[0009] Preferably, the process of obtaining the corrected similarity matrix of the continuous progress state sequence includes: Based on the inherent risk weight of nodes, a vulnerability neighborhood propagation simulation is performed on task nodes with task dependencies to obtain the infection risk value of task nodes. Based on the infection risk value, element-wise gain optimization is performed on the trajectory similarity matrix to obtain the corrected similarity of the trajectory similarity matrix. The formula for calculating the corrected similarity is as follows: ; In the formula, This represents the corrected similarity between trajectory p and trajectory q in the trajectory similarity matrix. This represents the original similarity between trajectory p and trajectory q in the trajectory similarity matrix. It is the natural logarithm function. This is the weighted average of the infection risk of trajectory p in the trajectory similarity matrix. This is the weighted average of the infection risk of trajectory q in the trajectory similarity matrix. It is a minimum value function. For all infection risk values, It is an exponential function. The preset sensitivity coefficient for vulnerability differences; The corrected similarity is reconstructed by normalizing the matrix to obtain the corrected similarity matrix of the continuous progress state sequence.

[0010] Preferably, in step S04, the process of obtaining the state transition point of the evolution trajectory includes: By performing trend convergence on the schedule state trajectory cluster, the dominant induced trajectory of the schedule state trajectory cluster is obtained. Based on the construction plan, the phase synchronization alignment of the main artificial trajectory is performed to obtain the phase deviation characterization of the main artificial trajectory. By locating the inflection point of the phase deviation characterization, the curvature extremum point of the phase deviation characterization is obtained; By performing spatiotemporal fusion on the curvature extrema, the state transition points of the evolution trajectory are obtained.

[0011] Preferably, in step S05, the process of obtaining the common conditions for state transition points includes: Multi-level progressive backtracking is performed on the state transition points to obtain the hierarchical preceding events of the state transition points; Based on the evolutionary trajectory, the hierarchical preceding events are reorganized by temporal linking to obtain the candidate preceding event chain of state transition points; Structural pattern encoding is performed on the candidate preceding event chains to obtain the event chain encoding sequence of state transition points; Frequent continuous sequence pattern mining is performed on the event chain encoding sequence to obtain the common sequence patterns of state transition points; By performing semantic description mapping on common sequence patterns, common conditions for state transition points are obtained.

[0012] Preferably, the process of obtaining the common sequence pattern of state transition points includes: A sliding window traversal is performed on the event chain encoding sequence to obtain the potential event sequence of the event chain encoding sequence; Based on state transition points, the potential event sequences are weighted according to their importance to obtain weighted candidate events for the potential event sequences; The time span of the weighted candidate events is analyzed to obtain the event timestamp sequence of the weighted candidate events; Based on the event timestamp sequence, a compactness measure is performed on the weighted candidate events to obtain the temporal distribution density of the weighted candidate events; Based on the time distribution density, the significance score of the weighted candidate events is calculated. The formula for calculating the significance score is as follows: ; In the formula, Here, k is the significance score, k is the number of coded events in the weighted candidate events, and j is the index of the event timestamp sequence. Encode the occurrence timestamp of event j in the event timestamp sequence. Encode the occurrence timestamp of event j+1 in the event timestamp sequence. The average time interval between consecutive encoded events in the event timestamp sequence. For the weighted candidate event, the encoded event i, Here, i is the index of the weighted candidate event, representing the predefined weighting function for project importance. Based on the significance score, the weighted candidate events are prioritized and filtered to obtain the highly significant events among the weighted candidate events; By summarizing and integrating highly significant events, common sequence patterns of state transition points are obtained.

[0013] Preferably, in step S06, the process of obtaining the deviation diagnosis result of the target project includes: The applicability of common conditions is evaluated to obtain the applicability weights of the common conditions; Based on applicability weights, a pattern scan is performed on the current progress status of the target project to obtain potential risk matching points for the current progress status. By performing risk transmission simulation on potential risk matching points, the risk impact range of potential risk matching points can be obtained; Based on the scope of risk impact, a comprehensive analysis of the current progress status is conducted to obtain the deviation diagnosis results for the target project.

[0014] A cluster analysis-based engineering construction schedule deviation diagnosis system, used to implement the aforementioned cluster analysis-based engineering construction schedule deviation diagnosis method, includes: The state serialization module is used to perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project. The similarity matrix construction module is used to measure the similarity of the evolution trajectory of a continuous progress state sequence and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. The trajectory clustering module is used to perform trajectory pattern clustering on a continuous progress state sequence based on the trajectory similarity matrix, so as to obtain the progress state trajectory cluster of the continuous progress state sequence. The key point localization module is used to locate key turning points of the evolution trajectory based on the progress state trajectory cluster and the construction plan of the target project, and obtain the state transition points of the evolution trajectory. The precondition mining module is used to backtrack the critical conditions of state transition points to obtain the precondition parameters of state transition points, and to perform correlation mining on the precondition parameters to obtain the common conditions of state transition points. The risk prediction module is used to perform predictive risk analysis on the target project based on common conditions, and obtain the deviation diagnosis results of the target project.

[0015] The present invention has the following beneficial effects: This invention analyzes the dependencies of original progress data by slicing time-series state data, and combines multi-scale morphological waveform decomposition and adaptive time warp alignment technology to accurately capture the morphological features of the progress evolution trajectory and improve the accuracy of trajectory similarity assessment. At the same time, it incorporates task topology vulnerability analysis and node inherent risk weights to optimize and correct the similarity matrix, thereby achieving phased and highly correlated progress state trajectory cluster division. This makes the progress state representation more in line with the characteristics of engineering construction and greatly improves the accuracy and relevance of data processing.

[0016] This invention relies on the phase synchronization and alignment of the author's programmed trajectory with the construction plan to accurately locate state transition points. Through multi-level progressive backtracking and frequent continuous sequence pattern mining, it extracts the common conditions for the causes of deviations in a structured manner. Based on these common conditions, it conducts predictive risk analysis to accurately identify potential risk matching points and simulate the scope of impact, thereby improving the foresight and comprehensiveness of deviation diagnosis, effectively increasing the efficiency of project schedule deviation diagnosis, and providing scientific and reliable decision support for schedule adjustments. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a functional block diagram of the system of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Example 1: As Figure 1 As shown, a method for diagnosing engineering construction schedule deviations based on cluster analysis includes the following steps: S01. Perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project; S02. Perform similarity measurement on the evolution trajectory of the continuous progress state sequence, and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. S03. Based on the trajectory similarity matrix, the continuous progress state sequence is divided into trajectory pattern clusters to obtain the progress state trajectory clusters of the continuous progress state sequence. S04. Based on the progress state trajectory cluster and the construction plan of the target project, locate the key turning points of the evolution trajectory to obtain the state transition points of the evolution trajectory. S05. Perform critical condition backtracking on the state transition points to obtain the preconditions of the state transition points, and perform correlation mining on the preconditions to obtain the common conditions of the state transition points. S06. Based on common conditions, conduct predictive risk analysis on the target project to obtain deviation diagnosis results for the target project.

[0019] In S01, the continuous schedule sequence of the target project is obtained, including: Dependency analysis is performed on the original schedule data of the target project to obtain the event logic diagram of the original schedule data; Based on the event logic diagram, the original progress data is timestamped to obtain the time-series baseline event flow of the original progress data of the target project; The time-series reference event stream is divided into non-uniform density segments to obtain adaptive time segments of the unified time-series reference event stream; By performing cross-segment correlation and integration on adaptive time segments, a continuous progress status sequence of the target project is obtained.

[0020] The process involves analyzing all tasks included in the original progress data of the target project, clarifying the prerequisites for each task and the subsequent tasks associated with it, defining the objective execution order of all tasks, determining the dependencies of each task on other tasks within the overall project progress, using each task as an independent unit, connecting dependent tasks with directional lines, clearly demonstrating the logical connections between all tasks, and ultimately forming an event logic diagram of the original progress data.

[0021] Based on the task dependency sequence clearly defined in the event logic diagram, the specific time of each task's actual start and final completion is extracted and used as a timestamp. All tasks are rearranged according to the order shown by the timestamps to ensure that the order of tasks in the time dimension fully matches the dependency dimension requirements specified in the event logic diagram, thus constructing a logically coherent time-series benchmark event flow that conforms to the actual execution situation.

[0022] The distribution of all tasks in the time-series baseline event stream is analyzed at different time periods. The density of tasks on the time axis is observed. For time periods with relatively concentrated task distribution, the length of the time segments in those time periods is set to a shorter level. For time periods with relatively dispersed task distribution, the length of the time segments in those time periods is set to a longer level. This ensures that each time segment contains a moderate number of tasks and that the division of time segments accurately matches the actual distribution characteristics of the tasks, ultimately forming an adaptive time segment for the unified time-series baseline event stream.

[0023] Extract the execution status, completion status, and inter-task relationships of each task within each adaptive time segment. Conduct in-depth analysis of the task connection relationship between adjacent adaptive time segments, clarify the succession logic between unfinished tasks in the previous time segment and related tasks in the next time segment, connect all adaptive time segments sequentially according to the time progression, and comprehensively integrate the task relationship information between each segment to eliminate information gaps between different segments. This will enable the progress status of all time segments to form a continuous and complete whole, resulting in a continuous progress status sequence of the target project.

[0024] Step S01 clearly presents the logical relationships between all tasks in the original progress data. By clarifying the preparatory requirements for each task, subsequent related tasks, and their execution order, it accurately defines the dependencies between tasks, avoiding data chaos caused by logical ambiguity. Using the task dependency order as a framework, it constructs a time-series baseline event flow based on actual start and completion times, ensuring a high degree of alignment between tasks in both time and logical dimensions, guaranteeing the consistency and accuracy of progress data. Time segments are differentiated according to the density of task distribution, ensuring each segment contains an appropriate number of tasks, accurately matching the actual distribution characteristics of tasks, and improving the rationality and adaptability of time segment division. By integrating task information within each segment and the connection logic between adjacent segments, it eliminates information gaps, forming a continuous and complete progress state sequence. This comprehensively and accurately reflects the evolution of project progress, providing high-quality data support for subsequent evolution trajectory analysis and trajectory clustering, and facilitating the efficient advancement of project progress deviation diagnosis.

[0025] In S02, the trajectory similarity matrix of the continuous progress state sequence is obtained, including: Multi-scale morphological waveform decomposition is performed on the continuous progress state sequence to obtain the morphological components of the continuous progress state sequence. Based on morphological components, adaptive time warp alignment is performed on the evolution trajectory of a continuous progress state sequence to obtain the alignment path of the evolution trajectory. Based on the alignment path, the evolutionary trajectory is compared point by point to obtain the morphological difference sequence of the evolutionary trajectory; By weighted fusion of morphological difference sequences, the comprehensive morphological distance of the evolutionary trajectory is obtained; Based on the comprehensive morphological distance, a nonlinear dimension reduction projection is performed on the evolution trajectory to obtain the low-dimensional manifold coordinates of the evolution trajectory; Based on low-dimensional manifold coordinates, the comprehensive morphological distance is normalized and reconstructed to obtain the trajectory similarity matrix of the continuous progress state sequence.

[0026] The continuous progress state sequence is decomposed layer by layer according to different time scales, gradually refining it from macro time span to micro time segment, separating the morphological features of different frequencies in the sequence, including overall trend features and local fluctuation features. The independent morphological features extracted at each time scale together constitute the morphological components of the continuous progress state sequence.

[0027] Based on morphological components, the occurrence order and change pattern of corresponding morphological features in different evolutionary trajectories are analyzed. By stretching or compressing the time axis of the trajectory, the time position of morphological feature points on the trajectory is adjusted so that similar morphological feature points in different evolutionary trajectories correspond precisely, eliminating morphological matching deviations caused by different time progression speeds, and forming a time correspondence between evolutionary trajectories, i.e., the alignment path of evolutionary trajectories.

[0028] According to the time correspondence rules determined by the alignment path, the state data of each evolutionary trajectory at the same corresponding time point are compared one by one, and the difference value of the state data of different trajectories at each corresponding time point is recorded. The difference values ​​of all corresponding time points are arranged in chronological order to form a morphological difference sequence that can reflect the point-by-point differences between evolutionary trajectories.

[0029] Based on the importance of morphological components in the overall evolutionary trajectory, a fixed weight is assigned to the difference value corresponding to each morphological component. Each difference value in the morphological difference sequence is multiplied by its corresponding weight to obtain a weighted difference value. Then, all weighted difference values ​​are summarized and integrated to obtain a comprehensive morphological distance that can comprehensively reflect the overall morphological differences of the evolutionary trajectory.

[0030] By preserving the relative differences and correlations between evolutionary trajectories, high-dimensional comprehensive morphological distance information is projected into a low-dimensional space. By preserving key difference information and eliminating redundant information, a unique coordinate position is assigned to each evolutionary trajectory in the low-dimensional space. These coordinate positions together constitute the low-dimensional manifold coordinates of the evolutionary trajectory.

[0031] The range of values ​​for all comprehensive morphological distances is determined, and each comprehensive morphological distance is normalized by converting it to the same value range according to a unified standard. Combining the correlation between evolutionary trajectories reflected by low-dimensional manifold coordinates, a two-dimensional matrix is ​​constructed. Each element in the matrix corresponds to the similarity between two evolutionary trajectories after normalization. Finally, the trajectory similarity matrix of the continuous progress state sequence is obtained.

[0032] By breaking down continuous progress state sequences layer by layer according to different time scales, and refining the extraction of overall trends and local fluctuations from macro to micro levels, multi-dimensional morphological information of the sequences can be comprehensively captured, laying a solid foundation for subsequent evolutionary trajectory analysis. Adjusting the trajectory time axis based on morphological components ensures precise correspondence of similar morphological feature points across different evolutionary trajectories, completely eliminating morphological matching deviations caused by differences in time progression speed and ensuring the accuracy of trajectory comparison. Trajectory state data is compared point-by-point along the alignment path, and difference values ​​are recorded in an orderly manner to form a morphological difference sequence, fully presenting the point-by-point differences between trajectories. By setting fixed weights based on the importance of morphological components and summarizing weighted difference values, the resulting comprehensive morphological distance fully reflects the overall morphological differences of the evolutionary trajectories. High-dimensional information is projected into a low-dimensional space, retaining key difference information and eliminating redundancy, allowing low-dimensional manifold coordinates to accurately represent the correlation between trajectories. A trajectory similarity matrix is ​​constructed through normalization processing and correlation degree combination. Elements in the matrix accurately represent the similarity between any two trajectories, providing high-quality data support for subsequent trajectory pattern clustering and ensuring the accuracy of project progress deviation diagnosis.

[0033] In S03, the progress state trajectory cluster of the continuous progress state sequence is obtained, including: A topological vulnerability analysis is performed on the task dependencies of the target project to obtain the inherent risk weights of the nodes in the task dependencies. Based on the inherent risk weights of nodes, a perception enhancement transformation is performed on the trajectory similarity matrix to obtain the corrected similarity matrix of the continuous progress state sequence. Based on the construction plan of the target project, the modified similarity matrix is ​​reorganized into a phase-consistent form, and the reorganized matrix is ​​then subjected to adaptive sub-block segmentation to obtain the intra-phase similarity sub-blocks of the target project. Based on the intra-stage similarity sub-blocks, the stage progress state sequence of the continuous progress state sequence is mapped in the feature space to obtain the feature coordinates of the stage progress state sequence. Density peak aggregation is performed on the feature coordinates to obtain the intra-stage trajectory clusters of the stage progress state sequence; Cross-stage pattern linking is performed on the trajectory clusters within a stage to obtain a progress state trajectory cluster of a continuous progress state sequence.

[0034] Topological vulnerability refers to the inherent risk transmission characteristics of nodes (tasks) in an engineering construction task dependency network due to their different network positions. It is quantified by analyzing the dependency connection structure of task nodes: if a task is a prerequisite for multiple subsequent tasks (large out-degree), lacks alternative execution paths (concentrated in-degree), or is located at the intersection of critical paths, then that node has high topological vulnerability. This vulnerability does not stem from the execution difficulty of the task itself, but from its structural position in the overall engineering network: once this task is delayed, the risk will be transmitted to downstream tasks along the dependency chain, forming a structural risk amplification effect. By simulating the transmission of inherent risk weights and vulnerability neighborhoods within nodes, topological vulnerability is transformed into a contagion risk value, thereby correcting the trajectory similarity matrix and making the clustering results more closely reflect the actual risk transmission patterns in engineering projects.

[0035] Based on the inherent risk weights of nodes, a perception enhancement transformation is performed on the trajectory similarity matrix to obtain a corrected similarity matrix for the continuous progress state sequence, including: Based on the inherent risk weight of nodes, a vulnerability neighborhood propagation simulation is performed on task nodes with task dependencies to obtain the infection risk value of task nodes. Based on the infection risk value, element-wise gain optimization is performed on the trajectory similarity matrix to obtain the corrected similarity of the trajectory similarity matrix. The formula for calculating the corrected similarity is as follows: ; In the formula, This represents the corrected similarity between trajectory p and trajectory q in the trajectory similarity matrix. This represents the original similarity between trajectory p and trajectory q in the trajectory similarity matrix. It is the natural logarithm function. This is the weighted average of the infection risk of trajectory p in the trajectory similarity matrix. This is the weighted average of the infection risk of trajectory q in the trajectory similarity matrix. It is a minimum value function. For all infection risk values, It is an exponential function. The preset vulnerability difference sensitivity coefficient can be set based on experience; The corrected similarity is reconstructed by normalizing the matrix to obtain the corrected similarity matrix of the continuous progress state sequence.

[0036] The dependency structure of all tasks in the target project is analyzed, the position and role of each task node in the overall dependency network are analyzed, and it is determined whether a node is a prerequisite for multiple subsequent tasks or lacks alternative execution paths. Based on the degree of influence of the node on the overall progress, a fixed risk weight is assigned to each task node to obtain the inherent risk weight of the node in the task dependency relationship.

[0037] Based on the inherent risk weight of each node, the directly related task nodes of each task node are tracked, and the inherent risk of the node is passed to the neighboring nodes according to the degree of dependence. The neighboring nodes then combine their own risk with the risk passed on and continue to pass it to the next level node they are associated with, and so on, to complete the risk transmission process of all nodes and obtain the infection risk value of each task node.

[0038] For each original similarity element in the trajectory similarity matrix corresponding to two trajectories, the infection risk value of all task nodes involved in these two trajectories is extracted. Based on the correlation of these infection risk values, the original similarity is adjusted in a targeted manner so that the similarity elements corresponding to trajectories with close risk correlation receive reasonable gains, forming the corrected similarity of the trajectory similarity matrix.

[0039] Determine the value range of all corrected similarities, adjust all corrected similarities to the same numerical range using a unified standard to eliminate scale differences between different elements, and then reorganize and arrange all normalized corrected similarities according to the correspondence between trajectories to construct a complete corrected similarity matrix for the continuous progress state sequence.

[0040] Referring to the construction stages divided in the target project's construction plan, the trajectory similarity elements corresponding to the same construction stage in the modified similarity matrix are grouped and reorganized to ensure that the matrix structure is consistent with the construction stage division. Then, based on the number of trajectories and similarity distribution characteristics within each stage, the reorganized matrix is ​​divided into appropriately sized independent sub-blocks. Each sub-block contains only the similarity information of trajectories within the same construction stage, thus obtaining the intra-stage similarity sub-blocks of the target project.

[0041] Based on the degree of trajectory similarity reflected in the similarity sub-blocks within a stage, the core features of each stage progress state sequence are transformed into spatial coordinates. In the preset feature space, a unique position is assigned to each stage progress state sequence according to the similarity relationship between sequences. This position can accurately reflect the characteristic attributes of the sequence, thus obtaining the feature coordinates of the stage progress state sequence.

[0042] In the feature space, regions with dense distribution of feature coordinates are retrieved, and the core peak points of these dense regions are identified. Feature coordinates distributed around the same core peak point are grouped into a set. Each set represents a set of stage progress state sequences with highly similar features. These sets are the intra-stage trajectory clusters of the stage progress state sequences.

[0043] Analyze the correlation patterns between trajectory clusters within different construction stages, identify the succession logic between trajectory clusters of the previous construction stage and trajectory clusters of the next construction stage, and link the trajectory clusters within the stages with succession relationships in sequence according to the time sequence of project construction to form a continuous trajectory cluster covering the entire construction cycle of the target project, and finally obtain the progress state trajectory cluster of the continuous progress state sequence.

[0044] The initial similarity between trajectory p and trajectory q comes from the trajectory similarity matrix formed after measuring the similarity of the evolutionary trajectories of the continuous progress state sequence and performing topological mapping. This matrix represents the initial similarity data between the corresponding trajectories p and q. The contagion risk weighted average of trajectory p is the average of the contagion risk values ​​of all task nodes involved in trajectory p, calculated by assigning fixed weights to each task node according to its importance in trajectory p. The contagion risk weighted average of trajectory q is obtained in the same way. All contagion risk values ​​are the set of contagion risk values ​​for all task nodes obtained after simulating the vulnerability neighborhood propagation of task nodes with task dependencies based on their inherent risk weights. The vulnerability difference sensitivity coefficient is a fixed value pre-set according to the task characteristics and risk assessment standards of the target project.

[0045] This formula is based on the original similarity between trajectory p and trajectory q. It incorporates the weighted average of the infection risk of the two trajectories, the minimum value of all infection risk values, and information related to the vulnerability difference sensitivity coefficient. It performs targeted gain optimization on the original similarity, so that the corrected similarity can not only reflect the similarity between the two trajectories in morphological evolution, but also accurately reflect the risk association characteristics of the task nodes involved in the two trajectories. This makes the similarity results more in line with the consideration of risk factors in the diagnosis of project schedule deviation.

[0046] When the ratio of the product of the weighted averages of the infection risks of two trajectories to the minimum of all infection risk values ​​increases, and the difference between the weighted averages of the infection risks of the two trajectories decreases, the result of the exponential part will approach 1, and the result of the natural logarithmic part will increase accordingly. The gain of the corrected similarity relative to the original similarity will be more prominent. When the difference between the weighted averages of the infection risks of two trajectories widens, the result of the exponential part will decrease, and the result of the natural logarithmic part will also decrease accordingly. The gain of the corrected similarity relative to the original similarity will tend to level off. This differentiated gain achieves reasonable optimization of the similarity of trajectories with different risk associations.

[0047] By analyzing the task dependency connection structure and the position and role of nodes in the dependency network, a fixed risk weight is assigned to each task node based on its impact on the overall progress. This accurately obtains the inherent risk weight of nodes in the task dependency relationship, laying a solid foundation for subsequent risk correlation analysis. Based on the degree of dependency tightness, the inherent risk of nodes is transmitted and combined with the node's own risk to complete the entire node risk transmission process. The resulting contagion risk value comprehensively reflects the overall state of task nodes after risk transmission. By combining the correlation degree of contagion risk values ​​of task nodes involved in two trajectories, the original similarity elements in the trajectory similarity matrix are adjusted to ensure that the corrected similarity accurately reflects the risk correlation characteristics between trajectories.

[0048] By standardizing all corrected similarity values ​​to the same range and eliminating scale differences, the matrix is ​​reorganized to ensure data consistency and structural standardization. The matrix is ​​reorganized and segmented according to construction stages to obtain intra-stage similarity sub-blocks, allowing for precise matching of similarity information with the characteristics of each construction stage and improving application targeting. The core features of the stage progress status sequence are transformed into feature coordinates to accurately represent the sequence's characteristic attributes. Densely distributed coordinates in the feature space are clustered to form intra-stage trajectory clusters, enabling effective classification of sequences with highly similar features. The logical connection between trajectory clusters of each stage is identified and linked according to construction time sequence, forming progress status trajectory clusters covering the entire construction cycle of the target project. This provides high-quality clustering support for locating key turning points in the subsequent evolution trajectory, ensuring the orderly and efficient progress of project progress deviation diagnosis.

[0049] This step clarifies the reliable sources and acquisition methods of each parameter. The original similarity is taken from the trajectory similarity matrix. The weighted average of contagion risk is calculated by assigning weights according to the importance of task nodes. All contagion risk values ​​are obtained through vulnerability neighborhood transmission simulation. The vulnerability difference sensitivity coefficient is preset in combination with engineering characteristics, providing accurate basic data for correcting the similarity calculation. The formula integrates trajectory morphology similarity and risk association information, and optimizes the gain of the original similarity in a targeted manner. This ensures that the corrected result reflects both the similarity of trajectory evolution morphology and the risk association characteristics of task nodes, aligning with the consideration of risk factors in schedule deviation diagnosis. Through differentiated gain adjustment, the similarity is optimized based on the ratio and difference changes of the weighted average of trajectory contagion risk, ensuring that the similarity results of trajectories with different risk associations are reasonable. This enhances the supporting role of similarity data in the subsequent division of schedule status trajectory clusters, helping to accurately diagnose engineering schedule deviations.

[0050] In S04, the state transition points of the evolution trajectory are obtained, including: By performing trend convergence on the schedule state trajectory cluster, the dominant induced trajectory of the schedule state trajectory cluster is obtained. Based on the construction plan, the phase synchronization alignment of the main artificial trajectory is performed to obtain the phase deviation characterization of the main artificial trajectory. By locating the inflection point of the phase deviation characterization, the curvature extremum point of the phase deviation characterization is obtained; By performing spatiotemporal fusion on the curvature extrema, the state transition points of the evolution trajectory are obtained.

[0051] Traverse all evolutionary trajectories in the progress state trajectory cluster, extract the time node corresponding to each trajectory and the progress state information under that node, sort out the common features of the progress state of all trajectories at the same time node, integrate the common features of each time node to form a unified trend, and discard isolated local fluctuation information in the trajectory that does not have commonality, and finally form a trajectory that can accurately represent the core evolution direction of the entire progress state trajectory cluster, that is, the master-directed evolutionary trajectory of the progress state trajectory cluster.

[0052] Extract the key stage nodes clearly defined in the target project construction plan, including the planned start and completion nodes of each sub-project and the key milestone nodes of the overall project. Use these nodes as a benchmark to adjust the time phase of the master-directed trajectory, so that the progress stage transitions on the master-directed trajectory are precisely synchronized with the stage division of the construction plan. Then, compare the aligned master-directed trajectory with the progress requirements of the corresponding stage in the construction plan, record the difference between the actual progress status and the planned progress status at each stage node, and arrange all differences in chronological order to obtain the phase deviation characterization of the master-directed trajectory.

[0053] The phase deviation characterization is analyzed segment by segment to obtain the change amplitude and direction of change between two adjacent difference values. When the change amplitude of the difference value at a certain position reaches the preset standard and the change direction changes from increasing to decreasing or from decreasing to increasing, that position is the inflection point of the phase deviation characterization. At the same time, this position is also the point where the curvature of the deviation change reaches the extreme value. The time coordinates and deviation values ​​corresponding to these inflection points are accurately recorded to form the curvature extreme points of the phase deviation characterization.

[0054] Collect time information and deviation information corresponding to all curvature extrema, analyze the distribution pattern of these extrema in the time dimension, and the engineering progress execution background corresponding to each extrema. Integrate curvature extrema that are adjacent in time and have logical correlation in deviation changes, eliminate isolated extrema that have no practical engineering significance, and retain key extrema that can mark a substantial change in the engineering progress status. These key extrema together constitute the state transition points of the evolution trajectory.

[0055] Step S04 extracts the time nodes and progress status information of all evolution trajectories, identifies common characteristics at the same time nodes, and discards isolated local fluctuations. The resulting master-directed trajectories accurately represent the core evolution direction of the progress status trajectory clusters, providing a reliable foundation for subsequent alignment with the construction plan. Key stage nodes in the construction plan are extracted to adjust the time phase of the master-directed trajectories, synchronizing progress stage transitions with the construction plan. Phase deviation representations obtained by comparing recorded differences clearly show the discrepancies between actual and planned progress. The curvature extrema points located by segmentally analyzing deviation changes accurately capture key locations of deviation changes. Collecting extrema point information, integrating logically related points, and eliminating isolated points yields state transition points that mark substantial changes in project progress status, providing crucial node support for subsequent project progress deviation diagnosis.

[0056] In S05, the common conditions for obtaining state transition points include: Multi-level progressive backtracking is performed on the state transition points to obtain the hierarchical preceding events of the state transition points; Based on the evolutionary trajectory, the hierarchical preceding events are reorganized by temporal linking to obtain the candidate preceding event chain of state transition points; Structural pattern encoding is performed on the candidate preceding event chains to obtain the event chain encoding sequence of state transition points; Frequent continuous sequence pattern mining is performed on the event chain encoding sequence to obtain the common sequence patterns of state transition points; By performing semantic description mapping on common sequence patterns, common conditions for state transition points are obtained.

[0057] Frequent continuous sequence pattern mining of the event chain encoded sequence yields common sequence patterns of state transition points, including: A sliding window traversal is performed on the event chain encoding sequence to obtain the potential event sequence of the event chain encoding sequence; Based on state transition points, the potential event sequences are weighted according to their importance to obtain weighted candidate events for the potential event sequences; The time span of the weighted candidate events is analyzed to obtain the event timestamp sequence of the weighted candidate events; Based on the event timestamp sequence, a compactness measure is performed on the weighted candidate events to obtain the temporal distribution density of the weighted candidate events; Based on the time distribution density, the significance score of the weighted candidate events is calculated, where the formula for calculating the significance score is: ; In the formula, Here, is the significance score (representing significance per unit time), k is the number of coded events in the weighted candidate events (k≥2), and j is the index of the event timestamp sequence. Encode the occurrence timestamp of event j in the event timestamp sequence. Encode the occurrence timestamp of event j+1 in the event timestamp sequence. The average time interval between consecutive encoded events in the event timestamp sequence. For the weighted candidate event, the encoded event i, Here, i is the index of the weighted candidate event, representing the predefined weighting function for project importance. To avoid the denominator being 0, we can further introduce a minimal positive constant Ɛ with the dimension of time square: ; Based on the significance score, the weighted candidate events are prioritized and filtered to obtain the highly significant events among the weighted candidate events; By summarizing and integrating highly significant events, common sequence patterns of state transition points are obtained.

[0058] Starting from the time node corresponding to the state transition point, trace back step by step the preceding events at different levels. First, identify the first-level preceding event that directly caused the state transition, then trace back the second-level preceding event that triggered the first-level preceding event, and so on to complete the multi-level tracing. Organize all the traced events according to the level of influence of the events to form the hierarchical preceding events of the state transition point.

[0059] By referring to the time progression of the evolutionary trajectory, the actual occurrence time of each event in the hierarchical preceding events is sorted out. The preceding events of different levels and types are linked in chronological order to eliminate logical gaps between events and construct a complete event chain that conforms to the actual occurrence logic, thus obtaining the candidate preceding event chain of state transition points.

[0060] The events in each candidate preceding event chain are classified by type, and a unique identifier is assigned to each type of event. According to the order of events in the event chain, the identifiers corresponding to each event are arranged in sequence to form a symbol sequence that can reflect the event chain structure pattern, that is, the event chain encoding sequence of state transition points.

[0061] Set a fixed-length sliding window. Starting from the beginning of the event chain encoding sequence, slide the window backward in steps of one symbol. After each slide, extract the encoded fragment covered by the window. Treat all extracted encoded fragments as sequences that may contain common patterns to obtain the potential event sequence of the event chain encoding sequence.

[0062] Based on the causes and characteristics of state transition points, the influence of each potential event sequence on the occurrence of state transition points is analyzed. Fixed weights are assigned to potential event sequences with different degrees of influence. The weight assignment criteria are determined according to the closeness of the correlation between the event and the state transition point. The weighted potential event sequence is the weighted candidate event of the potential event sequence.

[0063] Extract the timestamps of all events contained in each weighted candidate event, and arrange the corresponding timestamps in the order in which the events are arranged in the weighted candidate events to ensure that the timestamp of each event corresponds one-to-one with the event itself, thus forming a sequence of event timestamps for the weighted candidate events.

[0064] The difference between two adjacent event timestamps in the event timestamp sequence is calculated to obtain the time interval between each adjacent event. The number of events occurring within a unit time range is counted. This number directly reflects the density of events in the time dimension among the weighted candidate events, thus obtaining the time distribution density of the weighted candidate events.

[0065] By combining the temporal distribution density of weighted candidate events with the previously assigned importance weights, a comprehensive evaluation is conducted according to fixed rules. The density corresponding to the temporal distribution density is combined with the influence corresponding to the weight to comprehensively measure the importance of each weighted candidate event and obtain the significance score of the weighted candidate event.

[0066] All weighted candidate events are sorted according to their significance scores. A fixed screening criterion is set, and weighted candidate events whose scores meet the criterion are selected. These screened events can highlight the key impact on the state transition point, and are therefore considered high significance events among the weighted candidate events.

[0067] By analyzing the structural composition and event arrangement patterns of all highly significant events one by one, the recurring event combinations and their order are extracted. These consistent combinations and orders are then summarized and integrated, and redundant information is removed to form a sequence pattern that can represent the common characteristics of multiple highly significant events, thus obtaining the common sequence pattern of state transition points.

[0068] By reconstructing each symbol in the common sequence pattern into its corresponding semantic description of the actual event, clarifying the specific event content represented by each symbol, and combining the order and combination of events, we summarize the common preconditions corresponding to these patterns in clear and standardized language, clarify the causal relationship between these conditions and the occurrence of state transition points, and obtain the common conditions of state transition points.

[0069] The number of coded events in the weighted candidate events is derived from the weighted candidate events themselves; it can be obtained by directly counting the total number of coded events contained within the weighted candidate events. The index of the event timestamp sequence is the position identifier of the coded event within the event timestamp sequence, naturally generated according to the order of events in the sequence. The occurrence timestamps of coded events j and j+1 are the time records of the corresponding coded events directly extracted from the event timestamp sequence of the weighted candidate events. The average time interval of consecutive coded events is obtained by first calculating the time intervals of all adjacent coded events in the event timestamp sequence, summing all intervals, and then dividing by the total number of intervals. Coded event i is a single coded event contained within the weighted candidate events, and is determined as the weighted candidate events are identified. The preset project importance weighting function is pre-set based on the task characteristics and importance evaluation criteria of the target project, and is used to assign a corresponding importance weight value to each coded event.

[0070] The importance weighting function for a project can be defined as a critical path weighting function based on the critical path attribute of the task (according to...). Whether it belongs to the critical path determines the output value, or it is defined as a schedule elasticity weight function based on the sensitivity of task schedule delay (assigning the highest weight to critical tasks with a total float time of zero, and decreasing the weight of tasks with a larger float time), or it is defined as a resource concentration weight function based on the intensity of resource input (quantifying the weight according to the proportion of human, mechanical, and financial resource input to the total project input).

[0071] This formula comprehensively measures the importance of weighted candidate events by integrating the importance weights of coded events with their temporal distribution characteristics. The numerator summarizes the importance weights of all coded events, reflecting the overall importance level of the weighted candidate events; the denominator combines the results of calculations related to the number of coded events and the dispersion of timestamps, reflecting the uniformity of the distribution of coded events over time. The combined result can objectively and comprehensively quantify the impact of weighted candidate events on state transition points, providing a reliable basis for subsequent screening of highly significant events.

[0072] When the sum of the importance weights of all encoded events increases, and the dispersion of the timestamps of the encoded events decreases (i.e., the difference between the time interval of adjacent encoded events and the average time interval narrows), the denominator value will decrease accordingly, ultimately leading to an increase in the overall score. Conversely, when the sum of the importance weights of the encoded events decreases, or the dispersion of the timestamps of the encoded events increases, the denominator value will increase accordingly, and the overall score will decrease. Through this correlation between importance weights and time distribution characteristics, accurate differentiation of the importance of different weighted candidate events can be achieved.

[0073] By tracing back through multiple levels, we can identify pre-existing influencing events at different levels. These hierarchical pre-existing events, organized by influence level, comprehensively cover the factors related to state transitions. Linking events at different levels chronologically forms a candidate pre-existing event chain, eliminating logical gaps and ensuring the integrity and rationality of the event chain. The event chain coding sequence obtained from event classification clearly presents the event chain structure pattern. A sliding window is used to extract potential event sequences, comprehensively uncovering possible common patterns. Weighted candidate events, weighted by the degree of correlation, highlight key influencing factors. A significance score, obtained through a comprehensive evaluation of time distribution density and importance weight, accurately filters highly significant events. The common sequence patterns formed through summarization and integration extract core common features. Common conditions obtained through semantic mapping clarify the common pre-existing conditions for state transitions, providing accurate basis for subsequent predictive risk analysis and facilitating efficient diagnosis of project schedule deviations.

[0074] This step clarifies the source and acquisition method of each parameter. The number of coded events is obtained by statistically counting the total number of coded events in the weighted candidate events. Timestamps are extracted from the event timestamp sequence. The average time interval is obtained by calculating the ratio of the sum of adjacent intervals to the number of times the time interval is used. The engineering importance weight function is preset in conjunction with task characteristics to ensure the accuracy and reliability of the basic data. The formula integrates the importance weight of coded events with time distribution characteristics. The numerator reflects the overall importance, and the denominator reflects the uniformity of time distribution, objectively quantifying the degree of influence of weighted candidate events on state transition points, providing reliable support for screening highly significant events. By observing the correlation between the sum of importance weights and the dispersion of timestamps, the importance of different weighted candidate events is accurately distinguished, ensuring the accuracy of subsequent common sequence pattern mining.

[0075] In S06, the deviation diagnosis results of the target project are obtained, including: The applicability of common conditions is evaluated to obtain the applicability weights of the common conditions; Based on applicability weights, a pattern scan is performed on the current progress status of the target project to obtain potential risk matching points for the current progress status. By performing risk transmission simulation on potential risk matching points, the risk impact range of potential risk matching points can be obtained; Based on the scope of risk impact, a comprehensive analysis of the current progress status is conducted to obtain the deviation diagnosis results for the target project.

[0076] Analyze the compatibility of each common condition with the current construction stage of the target project. Combine the project task type, risk control standards, and actual execution scenarios to verify the applicable scenarios and the possibility of the common conditions playing a role in the current project. Assign a fixed weight value to each common condition based on the degree of compatibility and actual application value, and finally form the applicability weight of the common conditions.

[0077] Based on the applicability weight as the priority, and in descending order of weight, the core features of each common condition are comprehensively compared with the actual situation of the current progress status of the target project. The state segments in the current progress that match the description of the common condition are scanned to accurately identify those state nodes that may cause progress deviations. These nodes are the potential risk matching points of the current progress status.

[0078] Starting from the task node corresponding to each potential risk matching point, we track the subsequent tasks associated with that node, the pre-processes that it depends on, and the relevant resource configurations. We simulate the transmission path of risk in the task chain and process flow, analyze the impact of risk on the execution progress, quality requirements, and resource needs of associated tasks, clarify the scope of tasks, process links, and time cycles affected by the risk, and accurately define the risk impact range of potential risk matching points.

[0079] Integrate specific information on the scope of risk impact, combine the current progress status with the deviation from the construction plan, analyze the types of progress deviations caused by the risk, identify the key tasks and core processes involved in the deviation, assess the degree of interference of the deviation on the overall project progress, sort out the root causes of the deviation, summarize all analysis results in standardized and clear language, and form a deviation diagnosis result that can comprehensively reflect the progress deviation of the target project.

[0080] By combining project task types, risk control standards, and actual execution scenarios, the suitability and applicability of common conditions are verified and assigned fixed weights. The resulting applicability weights accurately reflect the actual application value of each common condition, providing a scientific priority basis for subsequent risk analysis. The core characteristics of common conditions are comprehensively compared with the current progress status in descending order of weight, accurately scanning and identifying potential risk matching points that may cause schedule deviations, ensuring no risk nodes are missed. Starting from potential risk matching points, related tasks, preceding processes, and resource allocation are tracked, the risk transmission path is simulated, and the impact is analyzed. The scope of tasks, process links, and time cycles affected by the risk are clearly defined, providing a specific basis for deviation assessment. Integrating the scope of risk impact with schedule deviations, in-depth analysis of deviation types, key tasks, interference levels, and root causes is conducted. A comprehensive deviation diagnosis result is formed using standardized and clear language, providing accurate and reliable decision support for project schedule adjustments and facilitating efficient project schedule management.

[0081] Example 2: As Figure 2 As shown, a project construction schedule deviation diagnosis system based on cluster analysis is used to implement the method in Example 1, including: The state serialization module is used to perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project. The similarity matrix construction module is used to measure the similarity of the evolution trajectory of a continuous progress state sequence and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. The trajectory clustering module is used to perform trajectory pattern clustering on a continuous progress state sequence based on the trajectory similarity matrix, so as to obtain the progress state trajectory cluster of the continuous progress state sequence. The key point localization module is used to locate key turning points of the evolution trajectory based on the progress state trajectory cluster and the construction plan of the target project, and obtain the state transition points of the evolution trajectory. The precondition mining module is used to backtrack the critical conditions of state transition points to obtain the precondition parameters of state transition points, and to perform correlation mining on the precondition parameters to obtain the common conditions of state transition points. The risk prediction module is used to perform predictive risk analysis on the target project based on common conditions, and obtain the deviation diagnosis results of the target project.

[0082] Example 3: A device for diagnosing engineering construction progress deviations based on cluster analysis, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the method in Example 1 is implemented by the processor executing the program.

Claims

1. A method for diagnosing engineering construction schedule deviations based on cluster analysis, characterized in that, Includes the following steps: S01. Perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project; S02. Perform similarity measurement on the evolution trajectory of the continuous progress state sequence, and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. S03. Based on the trajectory similarity matrix, the continuous progress state sequence is divided into trajectory pattern clusters to obtain the progress state trajectory clusters of the continuous progress state sequence. S04. Based on the progress state trajectory cluster and the construction plan of the target project, locate the key turning points of the evolution trajectory to obtain the state transition points of the evolution trajectory. S05. Perform critical condition backtracking on the state transition points to obtain the preconditions of the state transition points, and perform correlation mining on the preconditions to obtain the common conditions of the state transition points. S06. Based on common conditions, conduct predictive risk analysis on the target project to obtain deviation diagnosis results for the target project.

2. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S01, the process of obtaining the continuous progress state sequence of the target project includes: Dependency analysis is performed on the original schedule data of the target project to obtain the event logic diagram of the original schedule data; Based on the event logic diagram, the original progress data is timestamped to obtain the time-series baseline event flow of the original progress data of the target project; The time-series reference event stream is divided into non-uniform density segments to obtain adaptive time segments of the unified time-series reference event stream; By performing cross-segment correlation and integration on adaptive time segments, a continuous progress status sequence of the target project is obtained.

3. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S02, the process of obtaining the trajectory similarity matrix of the continuous progress state sequence includes: Multi-scale morphological waveform decomposition is performed on the continuous progress state sequence to obtain the morphological components of the continuous progress state sequence. Based on morphological components, adaptive time warp alignment is performed on the evolution trajectory of a continuous progress state sequence to obtain the alignment path of the evolution trajectory. Based on the alignment path, the evolutionary trajectory is compared point by point to obtain the morphological difference sequence of the evolutionary trajectory; By weighted fusion of morphological difference sequences, the comprehensive morphological distance of the evolutionary trajectory is obtained; Based on the comprehensive morphological distance, a nonlinear dimension reduction projection is performed on the evolution trajectory to obtain the low-dimensional manifold coordinates of the evolution trajectory; Based on low-dimensional manifold coordinates, the comprehensive morphological distance is normalized and reconstructed to obtain the trajectory similarity matrix of the continuous progress state sequence.

4. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S03, the process of obtaining the progress state trajectory cluster of a continuous progress state sequence includes: A topological vulnerability analysis is performed on the task dependencies of the target project to obtain the inherent risk weights of the nodes in the task dependencies. Based on the inherent risk weights of nodes, a perception enhancement transformation is performed on the trajectory similarity matrix to obtain the corrected similarity matrix of the continuous progress state sequence. Based on the construction plan of the target project, the modified similarity matrix is ​​reorganized into a phase-consistent form, and the reorganized matrix is ​​then subjected to adaptive sub-block segmentation to obtain the intra-phase similarity sub-blocks of the target project. Based on the intra-stage similarity sub-blocks, the stage progress state sequence of the continuous progress state sequence is mapped in the feature space to obtain the feature coordinates of the stage progress state sequence. Density peak aggregation is performed on the feature coordinates to obtain the intra-stage trajectory clusters of the stage progress state sequence; Cross-stage pattern linking is performed on the trajectory clusters within a stage to obtain a progress state trajectory cluster of a continuous progress state sequence.

5. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 4, characterized in that, The process of obtaining the corrected similarity matrix of the continuous progress state sequence includes: Based on the inherent risk weight of nodes, a vulnerability neighborhood propagation simulation is performed on task nodes with task dependencies to obtain the infection risk value of task nodes. Based on the infection risk value, element-wise gain optimization is performed on the trajectory similarity matrix to obtain the corrected similarity of the trajectory similarity matrix. The formula for calculating the corrected similarity is as follows: ； In the formula, This represents the corrected similarity between trajectory p and trajectory q in the trajectory similarity matrix. This represents the original similarity between trajectory p and trajectory q in the trajectory similarity matrix. It is the natural logarithm function. This is the weighted average of the infection risk of trajectory p in the trajectory similarity matrix. This is the weighted average of the infection risk of trajectory q in the trajectory similarity matrix. It is a minimum value function. For all infection risk values, It is an exponential function. The preset sensitivity coefficient for vulnerability differences; The corrected similarity is reconstructed by normalizing the matrix to obtain the corrected similarity matrix of the continuous progress state sequence.

6. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S04, the process of obtaining the state transition point of the evolution trajectory includes: By performing trend convergence on the schedule state trajectory cluster, the dominant induced trajectory of the schedule state trajectory cluster is obtained. Based on the construction plan, the phase synchronization alignment of the main artificial trajectory is performed to obtain the phase deviation characterization of the main artificial trajectory. By locating the inflection point of the phase deviation characterization, the curvature extremum point of the phase deviation characterization is obtained; By performing spatiotemporal fusion on the curvature extrema, the state transition points of the evolution trajectory are obtained.

7. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S05, the process of obtaining the common conditions for state transition points includes: Multi-level progressive backtracking is performed on the state transition points to obtain the hierarchical preceding events of the state transition points; Based on the evolutionary trajectory, the hierarchical preceding events are reorganized by temporal linking to obtain the candidate preceding event chain of state transition points; Structural pattern encoding is performed on the candidate preceding event chains to obtain the event chain encoding sequence of state transition points; Frequent continuous sequence pattern mining is performed on the event chain encoding sequence to obtain the common sequence patterns of state transition points; By performing semantic description mapping on common sequence patterns, common conditions for state transition points are obtained.

8. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 7, characterized in that, The process of obtaining the common sequence pattern of state transition points includes: A sliding window traversal is performed on the event chain encoding sequence to obtain the potential event sequence of the event chain encoding sequence; Based on state transition points, the potential event sequences are weighted according to their importance to obtain weighted candidate events for the potential event sequences; The time span of the weighted candidate events is analyzed to obtain the event timestamp sequence of the weighted candidate events; Based on the event timestamp sequence, a compactness measure is performed on the weighted candidate events to obtain the temporal distribution density of the weighted candidate events; Based on the time distribution density, the significance score of the weighted candidate events is calculated. The formula for calculating the significance score is as follows: ； In the formula, Here, k is the significance score, k is the number of coded events in the weighted candidate events, and j is the index of the event timestamp sequence. Encode the occurrence timestamp of event j in the event timestamp sequence. Encode the occurrence timestamp of event j+1 in the event timestamp sequence. The average time interval between consecutive encoded events in the event timestamp sequence. For the weighted candidate event, the encoded event i, Here, i is the index of the weighted candidate event, representing the predefined weighting function for project importance. Based on the significance score, the weighted candidate events are prioritized and filtered to obtain the highly significant events among the weighted candidate events; By summarizing and integrating highly significant events, common sequence patterns of state transition points are obtained.

9. The method for diagnosing engineering construction progress deviations based on cluster analysis as described in claim 1, characterized in that, In S06, the process of obtaining the deviation diagnosis result of the target project includes: The applicability of common conditions is evaluated to obtain the applicability weights of the common conditions; Based on applicability weights, a pattern scan is performed on the current progress status of the target project to obtain potential risk matching points for the current progress status. By performing risk transmission simulation on potential risk matching points, the risk impact range of potential risk matching points can be obtained; Based on the scope of risk impact, a comprehensive analysis of the current progress status is conducted to obtain the deviation diagnosis results for the target project.

10. A project construction schedule deviation diagnosis system based on cluster analysis, used to implement the project construction schedule deviation diagnosis method based on cluster analysis as described in any one of claims 1-9, characterized in that, include: The state serialization module is used to perform time-series state slicing on the original progress data of the target project to obtain a continuous progress state sequence of the target project. The similarity matrix construction module is used to measure the similarity of the evolution trajectory of a continuous progress state sequence and perform topological mapping on the measurement results to obtain the trajectory similarity matrix of the continuous progress state sequence. The trajectory clustering module is used to perform trajectory pattern clustering on a continuous progress state sequence based on the trajectory similarity matrix, so as to obtain the progress state trajectory cluster of the continuous progress state sequence. The key point localization module is used to locate key turning points of the evolution trajectory based on the progress state trajectory cluster and the construction plan of the target project, and obtain the state transition points of the evolution trajectory. The precondition mining module is used to backtrack the critical conditions of state transition points to obtain the precondition parameters of state transition points, and to perform correlation mining on the precondition parameters to obtain the common conditions of state transition points. The risk prediction module is used to perform predictive risk analysis on the target project based on common conditions, and obtain the deviation diagnosis results of the target project.

Images