House steel structure construction prediction progress and deployment system based on deep learning

Abstract

The invention relates to the technical field of building engineering construction management, and discloses a house steel building construction prediction progress and deployment system based on deep learning. According to the system, periodic stable paragraphs and non-periodic fluctuation paragraphs are recognized by analyzing the time sequence form of construction flow data, and construction links are stripped accordingly. And performing cross mapping on the hoisting event sequence in the link and the environmental monitoring reading of the associated link to generate a link toughness spectrogram representing the construction robustness. And determining a resource demand based on the spectrum graph, and forming a multi-dimensional resource demand vector, so as to drive a graph convolutional network to construct a dynamic construction deduction graph. And carrying out message passing and neighborhood aggregation on the graph, analyzing a key bottleneck path, and iteratively generating a progress prediction and resource allocation scheme. According to the method, the construction dynamic toughness can be evaluated, and accurate positioning and dynamic optimization of resource bottlenecks are realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of building engineering construction management, in particular to a housing steel structure construction construction progress prediction and deployment system based on deep learning. BACKGROUND

[0002] In the current housing steel structure construction construction management, progress prediction and resource deployment mainly rely on traditional project management tools based on fixed plans or monitoring systems combined with part of Internet of Things data. These existing technical solutions usually simply summarize or independently analyze various types of collected construction data, and calculate the progress and allocate resources according to historical experience or static models. The extensive data processing makes it difficult to finely distinguish the inherent laws of construction activities from time series and the influence mechanism of environmental fluctuations and other factors on complex operation chains is not clearly quantified.

[0003] The defects of the existing method are that the evaluation of construction robustness is static and isolated, the internal relationship between periodic steady-state operation and non-periodic disturbance events in the construction flow is not deconstructed from the data level, and the real performance of a specific link when the associated link is disturbed cannot be dynamically described. At the same time, the resource deployment logic is often based on preset rules or isolated analysis of a single link, and lacks the ability to simulate resource competition, dependency relationship and dynamically identify bottleneck paths in the global resource flow network. This causes the progress prediction to be easily inaccurate under disturbance, and the resource deployment scheme to be rigid and difficult to adapt to the dynamic changes of the construction site. SUMMARY

[0004] The purpose of the present application is to provide a housing steel structure construction construction progress prediction and deployment system based on deep learning to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides a housing steel structure construction construction progress prediction and deployment system based on deep learning, which comprises:

[0006] A data acquisition module acquires real-time construction flow data of multiple sources and heterogeneous in the housing steel structure construction process. The construction flow data covers material flow trajectory, component hoisting event sequence and environmental monitoring readings;

[0007] A link identification module identifies periodic stable paragraphs and non-periodic fluctuation paragraphs in the material flow trajectory based on the time sequence form of the construction flow data, and accordingly separates a number of to-be-evaluated construction links from all construction activities;

[0008] A resilience spectrum module integrates the component hoisting event sequence in the periodic stable paragraph of each to-be-evaluated construction link, and cross-maps it with the environmental monitoring readings in the non-periodic fluctuation paragraph of the associated construction link, to generate a link resilience spectrum representing the construction robustness;

[0009] a resource calibration module, which calibrates the demand intensity and sensitivity of each construction resource in the periodic stable paragraph according to the link resilience spectrum, to form a multi-dimensional resource demand vector;

[0010] a graph deduction module, which drives the node feature update of the graph convolution network by using the multi-dimensional resource demand vector, and constructs a dynamic construction deduction graph with the resource flow path as the edge and the construction link as the node;

[0011] a prediction and deployment module, which analyzes the key bottleneck path in the dynamic construction deduction graph by performing multiple rounds of message passing and neighborhood aggregation, and iteratively generates the construction progress prediction and real-time resource deployment scheme based on the analysis result.

[0012] Preferably, the component hoisting event sequence includes the start and end time stamp of each hoisting operation, the spatial coordinate displacement of the hoisted component, and the load rate change curve of the hoisting machine; and the environmental monitoring readings include the continuous sampling values of wind speed and direction, the time sequence record of light intensity, and the measurement data stream of temperature and humidity sensors.

[0013] Preferably, the generation step of the link resilience spectrum specifically comprises:

[0014] For any construction link to be evaluated, all component hoisting event sequences in the periodic stable paragraph of the construction link are integrated into a link internal event set, and the environmental monitoring readings in the non-periodic fluctuation paragraph of all other construction links having material supply or spatial adjacent relationship with the construction link are integrated into an associated external disturbance set;

[0015] The morphological features of the load rate change curve of each hoisting operation in the link internal event set are extracted, and the statistical features of the continuous sampling values of wind speed and direction in the same period in the associated external disturbance set are extracted synchronously, and the morphological coupling coefficient of the morphological features and the statistical features is calculated;

[0016] The spatial coordinate displacement corresponding to each hoisting operation is convoluted with the morphological coupling coefficient to obtain the environmental disturbance response value of the hoisting operation;

[0017] The environmental disturbance response values of all hoisting operations in the link internal event set are aggregated, and time sequence alignment and smoothing processing are performed according to the start time stamp of the hoisting operation, to generate the link resilience spectrum of the construction link to be evaluated.

[0018] Preferably, the demand intensity and sensitivity of each construction resource in the periodic stable paragraph are calibrated in the following manner:

[0019] According to the distribution interval of the environmental disturbance response value in the link resilience spectrum, a high-robustness interval, a medium-robustness interval, and a low-robustness interval are divided;

[0020] For the hoisting operation in the high robustness interval, the corresponding component type and machine type are analyzed, and the average flow rate in the time interval of the material flow trajectory is combined to calibrate the demand intensity of the hoisting machine and the specific component;

[0021] For the hoisting operation in the low robustness interval, the wind speed or light intensity abnormal value exceeding the preset threshold in the corresponding environmental monitoring reading is extracted, and the frequency and average amplitude of the wind speed or light intensity abnormal value are calculated to calibrate the sensitivity to the construction environment condition;

[0022] The demand intensity and sensitivity are used as different dimensions of a vector to construct a multi-dimensional resource demand vector for each periodic stable segment of the construction link to be evaluated.

[0023] Preferably, the method for constructing a dynamic construction deduction graph with resource flow paths as edges and construction links as nodes comprises the following steps:

[0024] Each construction link to be evaluated is taken as a node in the graph, and the initial feature vector of the node is obtained by dimension reduction mapping of the corresponding multi-dimensional resource demand vector;

[0025] The flow path of the material in the actual physical space in the construction flow data is taken as the basis for constructing the directed edges between the nodes in the graph, and the average passing time of the historical material on the flow path and the current congestion state are used to assign initial weights to the directed edges;

[0026] The initial graph structure with node features and directed edge weights is input into a preset graph convolution network, and each layer of the graph convolution network performs the following operations: each node aggregates the feature vectors of all its in-edge neighbor nodes, and updates the aggregated features based on the weights of the in-edges. The updated node features will be used for neighborhood aggregation in the next layer or as the final node representation;

[0027] After multiple rounds of iterative updates, the graph structure with stable node features is output as the dynamic construction deduction graph.

[0028] Preferably, the process of analyzing the key bottleneck path in the graph is specifically:

[0029] In the dynamic construction deduction graph, the sum of the edge weights of all directed edges and the inner product of the feature vectors of the two end nodes connected by the directed edges is calculated to obtain the flow efficiency evaluation value of each directed edge;

[0030] All directed edges with a flow efficiency evaluation value lower than the system preset threshold are identified and marked as potential bottleneck edges.

[0031] If the simulation result shows that the accumulated delay of resources at the potential bottleneck edge exceeds a preset delay threshold, the potential bottleneck edge and the complete path where the potential bottleneck edge is located are determined as a critical bottleneck path.

[0032] Preferably, the construction progress prediction and real-time resource allocation scheme are iteratively generated based on the analysis result, and the iteration logic comprises:

[0033] In each iteration, the weight of the corresponding directed edge in the dynamic construction deduction graph is dynamically adjusted according to the currently identified critical bottleneck path, and the adjustment range of the weight is inversely proportional to the flow efficiency evaluation value of the critical bottleneck path;

[0034] Using the dynamic construction deduction graph with adjusted edge weights, the aggregation and update of node features and the analysis of critical bottleneck paths are performed again;

[0035] The set of critical bottleneck paths obtained in the new round of analysis is compared with the set obtained in the last round, and if the composition or number of the core bottleneck paths changes, an updated resource pre-allocation instruction and a progress correction amount are generated according to the new set of bottleneck paths;

[0036] The updated resource pre-allocation instruction and the progress correction amount are applied to the actual material flow trajectory and the component hoisting event sequence scheduling, and the new construction flow data generated thereby is collected as the input of the next iteration, until the set of critical bottleneck paths tends to be stable or reaches the maximum number of iterations, and the final construction progress prediction timeline and resource allocation scheme detailed table are output.

[0037] Preferably, the method for identifying periodic stable segments and non-periodic fluctuation segments in the material flow trajectory is:

[0038] The material flow trajectory data is sliced with a fixed time window, and the variance and autocorrelation coefficient of the material flow distance in each time window are calculated;

[0039] The sequence of time windows with continuous variance below the variance threshold and continuous autocorrelation coefficient above the correlation coefficient threshold is determined and merged as a periodic stable segment;

[0040] The sequence of time windows with continuous variance above the variance threshold or continuous autocorrelation coefficient below the correlation coefficient threshold is determined and merged as a non-periodic fluctuation segment.

[0041] Preferably, the method for calculating the morphological coupling coefficient of the shape feature and the statistical feature is further refined as:

[0042] The wavelet packet decomposition is performed on the load rate change curve, and the energy of the specified frequency band is extracted as the load shape feature vector;

[0043] Empirical mode decomposition is performed on the continuous sampling values of the wind speed and wind direction to obtain a plurality of intrinsic mode functions, and sample entropy of the first few intrinsic mode functions is calculated as a wind speed pattern feature vector;

[0044] A cosine similarity of the load pattern feature vector and the wind speed pattern feature vector in the standard orthogonal basis is calculated, and a value obtained by normalizing the cosine similarity through an S-shaped function is defined as the pattern coupling coefficient.

[0045] Preferably, the dynamic construction deduction graph after adjusting the edge weight is used to re-aggregate and update the node features and analyze the key bottleneck paths, specifically including the following steps:

[0046] The dynamic construction deduction graph after adjusting the edge weight is input into a preset graph convolution network model to perform the aggregation and update operation of the node features, wherein each node aggregates the latest feature vectors of all incoming edge neighbor nodes, and the aggregated feature vectors are weighted and fused based on the current weight values of the incoming edges to generate updated node feature vectors;

[0047] Based on the updated node feature vectors, the flow efficiency evaluation values of all directed edges in the dynamic construction deduction graph are recalculated;

[0048] According to the recalculated flow efficiency evaluation values, all directed edges with a value lower than a system preset threshold are identified, and the directed edges are marked as potential bottleneck edges in the next round;

[0049] The traffic propagation simulation algorithm is performed on the complete path containing the potential bottleneck edges, and if the simulation result shows that the accumulated delay time of resources at the potential bottleneck edge exceeds a preset delay threshold, the path is determined as a key bottleneck path.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] Based on the timing pattern of the construction flow data, the periodic stable paragraphs and the non-periodic fluctuation paragraphs are identified, and a stable hoisting event of a link is cross-mapped with a fluctuation environment reading of an associated link to generate a link resilience spectrum. The internal rhythm and external disturbance of the construction process can be separated from the data source, and through the establishment of dynamic association mapping across links and data types, the robustness level of the construction link in the face of random interference from associated operations is directly quantified. The obtained spectrum goes beyond traditional single indicators or static thresholds, and provides visual and calculable representation of the dynamic resilience of the construction process.

[0052] The resource is calibrated according to the link toughness spectrum diagram, the demand intensity and the sensitivity are formed, the multi-dimensional resource demand vector is formed, and the dynamic construction deduction diagram is constructed and updated by driving the graph convolution network. The diagram connects each construction link through the resource flow path, and through the multi-round message passing and neighborhood aggregation operation of the graph convolution network, the dynamic allocation, consumption and competition process of the resource in the complex construction network can be simulated. Based on the learning and deduction mechanism of the graph structure, the system can automatically discover and analyze the key bottleneck path which restricts the overall progress from the global perspective of resource flow, rather than only optimizing the local link. Based on the iterative analysis of the bottleneck path, the generated resource allocation scheme has dynamic adaptability and global optimization characteristics, and can continuously adjust with the construction progress and external condition changes. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The timing diagram of the deep learning-based housing steel construction construction progress prediction and allocation system is described.

[0054] Figure 2 The flowchart for generating the link toughness spectrum diagram is described.

[0055] Figure 3 The flowchart for constructing the dynamic construction deduction diagram is described.

[0056] Figure 4 The construction period comparison diagram before and after the resource allocation of the housing steel construction link is described.

[0057] Figure 5 The timing comparison diagram of the housing steel construction material flow distance sequence is described. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0059] Please refer to Figure 1The application provides a deep learning-based housing steel structure construction progress prediction and deployment system, which comprises a data acquisition module that acquires real-time construction flow data of multiple sources and heterogeneous in the housing steel structure construction process, the construction flow data covering material flow trajectory, component hoisting event sequence and environmental monitoring readings; a link identification module that identifies periodic stable paragraphs and non-periodic fluctuation paragraphs in the material flow trajectory based on the time sequence form of the construction flow data, and accordingly separates a plurality of to-be-evaluated construction links from all construction activities; a toughness spectrum module that, for each to-be-evaluated construction link, integrates the component hoisting event sequence in the periodic stable paragraph, and cross-maps the environmental monitoring readings in the non-periodic fluctuation paragraph associated with the construction link, to generate a link toughness spectrum representing construction robustness; a resource calibration module that, according to the link toughness spectrum, calibrates the demand intensity and sensitivity of each construction resource in the periodic stable paragraph, to form a multi-dimensional resource demand vector; a graph deduction module that drives node feature updating of a graph convolution network using the multi-dimensional resource demand vector, to construct a dynamic construction deduction graph with resource flow paths as edges and construction links as nodes; and a prediction and deployment module that analyzes key bottleneck paths in the dynamic construction deduction graph by performing multiple rounds of message passing and neighborhood aggregation on the graph, and iteratively generates a construction progress prediction and real-time resource deployment scheme based on the analysis result.

[0060] In embodiment 1, the component hoisting event sequence includes the start and end time stamps of each hoisting operation, the spatial coordinate displacement of the hoisted component, and the load rate change curve of the hoisting machine; the environmental monitoring readings include continuous sampling values of wind speed and direction, time sequence records of light intensity, and measurement data streams of temperature and humidity sensors.

[0061] In a specific implementation, the component hoisting event sequence includes the start and end time stamps of each hoisting operation, the spatial coordinate displacement of the hoisted component, and the load rate change curve of the hoisting machine, the start and end time stamps are recorded in international standard time format, for example, the start time stamp of a column component hoisting operation is 2023-11-05T08:30:00Z, and the end time stamp is 2023-11-05T08:45:15Z, the spatial coordinate displacement is measured by a total station, including incremental values in a three-dimensional rectangular coordinate system, for example, displacement of 20.5 meters along the east-west direction, displacement of 10.2 meters along the south-north direction, and vertical displacement of 15.8 meters, and the load rate change curve is generated by a force sensor built in the hoisting machine at a frequency of 10 times per second, the curve data point sequence reflects the change trend of the hoisting load percentage over time.

[0062] In some embodiments, the environmental monitoring readings include continuous sampling values of wind speed and direction outputted by an ultrasonic anemometer at a frequency of 5 times per second, the sampling values including instantaneous wind speed in meters per second and wind direction in degrees, for example, a sequence of wind speed sampling values during a morning period shows a wind speed range of 0.5 meters per second to 3.2 meters per second and a wind direction range of 90 degrees to 120 degrees, a time series of light intensity recorded by a digital light meter once per minute, in lux, for example, a sequence of recorded values during a sunny noon period rises from 850 lux to 1200 lux, and a measurement data stream of a temperature and humidity sensor including temperature values in degrees Celsius and relative humidity values in percentage, the sensor uploading data every 20 seconds, for example, a sequence of temperature values fluctuates from 18 degrees Celsius to 25 degrees Celsius and a sequence of relative humidity values changes from 45% to 65% in a day. Optionally, the spatial coordinate displacement amounts in the sequence of component hoisting events can be used to calculate a geometric path length of the hoisting operation, the geometric path length formula being expressed as:

[0063]

[0064] wherein the symbol represents the geometric path length, in meters; the symbol represents the spatial coordinate displacement amount of the hoisted component in the east-west direction; the symbol represents the spatial coordinate displacement amount of the hoisted component in the north-south direction; and the symbol represents the spatial coordinate displacement amount of the hoisted component in the vertical direction. By calculating the geometric path length, the movement distance difference of different hoisting operations can be compared, for example, the geometric path length of a beam component hoisting is 28.3 meters and the geometric path length of a plate component hoisting is 12.7 meters.

[0065] In a specific implementation, the start and end time stamps of the sequence of component hoisting events are captured through a high-precision timer and a hoisting action sensor linkage, with a time stamp resolution reaching the millisecond level, the spatial coordinate displacement amounts are obtained by calculating the difference between the hoisting point and the installation point through a global navigation satellite system receiver, and the load rate change curve is stored in an array format corresponding to the time stamps and the load percentage. In some embodiments, the continuous sampling values of wind speed and direction of the environmental monitoring readings are sent to a central database in a streaming manner through a data collector, the time series of light intensity is stored in a ring buffer in chronological order, and the measurement data stream of the temperature and humidity sensor is aggregated through a wireless sensor network. Optionally, the data comparison of the sequence of component hoisting events can be realized by displaying the load rate change curves of multiple operations side by side, with the horizontal axis being a standardized time unit and the vertical axis being a load percentage, and the data comparison of the environmental monitoring readings can be realized by superimposedly drawing the wind speed continuous sampling value curve and the light intensity time series curve, with the horizontal axis being the same time axis and the vertical axis being the wind speed value in meters per second and the light intensity value in lux, respectively.

[0066] Embodiment 2: refer to Figure 2 For any one construction link to be evaluated, all component hoisting event sequences in the periodic stable paragraph of the link are integrated into a link internal event set, and all environmental monitoring readings of other construction links in non-periodic fluctuation paragraphs that have material supply or spatial adjacent relationship with the link are integrated into an associated external interference set. The morphological feature of the load rate change curve of each hoisting operation in the link internal event set is extracted, and the statistical feature of the wind speed and direction continuous sampling values in the same period in the associated external interference set is extracted synchronously. The morphological coupling coefficient of the morphological feature and the statistical feature is calculated. The spatial coordinate displacement corresponding to each hoisting operation is convolved with the morphological coupling coefficient to obtain the environmental disturbance response value of the hoisting operation. The environmental disturbance response values of all hoisting operations in the link internal event set are aggregated, and time sequence alignment and smoothing processing are performed according to the starting time stamp of the hoisting operation to generate the link resilience spectrum of the construction link to be evaluated. The method of calculating the morphological coupling coefficient of the morphological feature and the statistical feature is further refined as follows: the load rate change curve is subjected to wavelet packet decomposition, and the energy of a specified frequency band is extracted as a load morphological feature vector; the wind speed and direction continuous sampling values are subjected to empirical mode decomposition to obtain a plurality of intrinsic mode functions, and the sample entropy of the first few intrinsic mode functions is calculated as a wind speed morphological feature vector; the cosine similarity of the load morphological feature vector and the wind speed morphological feature vector in the standard orthogonal basis is calculated, and the value obtained by normalizing the cosine similarity through an S-shaped function is defined as the morphological coupling coefficient.

[0067] In specific implementation, for any one construction link to be evaluated, all component hoisting event sequences in the periodic stable paragraph of the link are integrated into a link internal event set, which is a data set containing the starting and ending time stamps, spatial coordinate displacement and load rate change curve data of all hoisting operations in the paragraph. All environmental monitoring readings of other construction links in non-periodic fluctuation paragraphs that have material supply or spatial adjacent relationship with the link are integrated into an associated external interference set, which is a data set containing wind speed and direction continuous sampling values, light intensity time sequence records and temperature and humidity measurement data streams. In some embodiments, the morphological feature of the load rate change curve of each hoisting operation in the link internal event set is extracted, the morphological feature being a numerical vector reflecting the fluctuation pattern of the curve. The statistical feature of the wind speed and direction continuous sampling values in the same period in the associated external interference set is extracted synchronously, the statistical feature being a numerical vector describing the data distribution and concentration trend. The morphological coupling coefficient of the morphological feature and the statistical feature is calculated, the morphological coupling coefficient being a scalar value between 0 and 1.

[0068] In a specific implementation, the method of calculating the morphological coupling coefficient of the morphological features and the statistical features is further refined as follows: wavelet packet decomposition is performed on the load rate change curve, wavelet packet decomposition is a signal processing method, and the energy of a specified frequency band is extracted after decomposition as a load morphological feature vector; empirical mode decomposition is performed on the continuous sampling values of the wind speed and direction, empirical mode decomposition is an adaptive time-frequency analysis method, and a plurality of intrinsic mode functions are obtained after decomposition; the sample entropy of the first few intrinsic mode functions is calculated as a wind speed morphological feature vector, and the sample entropy is an index for measuring the complexity of a time series. Optionally, the cosine similarity of the load morphological feature vector and the wind speed morphological feature vector in the standard orthogonal basis is calculated, and the cosine similarity calculation formula is expressed as:

[0069]

[0070] wherein the symbol represents the similarity value without normalization; the symbol represents the load morphological feature vector; the symbol represents the wind speed morphological feature vector; the symbol represents the dot product operation of vectors; and the symbol represents the Euclidean norm of vectors. The value obtained after normalizing the cosine similarity by the S-shaped function is defined as the morphological coupling coefficient, and the S-shaped function is a mathematical function with an S-shaped curve, which is used to map any real number to the range of 0 to 1.

[0071] In a specific implementation, convolution operation is performed on the spatial coordinate displacement amount corresponding to each hoisting operation and the morphological coupling coefficient, the convolution operation is a mathematical integral transformation used to fuse two signals to obtain the environmental disturbance response value of the hoisting operation, the environmental disturbance response value is a scalar, the environmental disturbance response values of all hoisting operations in the internal event set of the link are aggregated, the aggregation operation is to arrange a plurality of scalar values in the corresponding time sequence to form a sequence, and to perform time alignment and smoothing processing according to the start time stamp of the hoisting operation, the smoothing processing adopts the moving average method, and the link resilience spectrum of the construction link to be evaluated is generated, the link resilience spectrum is a continuous curve with time as the horizontal axis and the environmental disturbance response value as the vertical axis. In some embodiments, the internal event set of the link and the associated external disturbance set are strictly aligned in time, ensuring that the data point time stamps of the load rate change curve and the continuous sampling values of the wind speed and direction are completely matched, and the calculation of the morphological coupling coefficient is independently performed for each aligned time window. Optionally, the energy of the specified frequency band of the load rate change curve is selected to form the load morphological feature vector together with the energy of the high frequency band, and the wind speed morphological feature vector is composed of the sample entropy values of the first three intrinsic mode functions.

[0072] Embodiment 3: refer to Figure 3, according to the distribution interval of the environmental disturbance response value in the link toughness spectrum, the high robustness interval, the medium robustness interval and the low robustness interval are divided; for the hoisting operation in the high robustness interval, the corresponding component type and machine type are analyzed, and the demand intensity for the hoisting machine and the specific component is calibrated combined with the average flow rate in the time period in the material flow track; for the hoisting operation in the low robustness interval, the wind speed or light intensity abnormal value exceeding the preset threshold in the corresponding environmental monitoring reading is extracted, the frequency and average amplitude of the wind speed or light intensity abnormal value are calculated, and the sensitivity to the construction environment condition is calibrated; the demand intensity and the sensitivity are taken as different dimensions of the vector, and a multi-dimensional resource demand vector is constructed for each periodic stable paragraph of the construction link to be evaluated. The method for constructing a dynamic construction deduction graph with resource flow paths as edges and construction links as nodes includes the following steps: taking each construction link to be evaluated as a node in the graph, the initial feature vector of the node is obtained by dimension reduction mapping of the corresponding multi-dimensional resource demand vector; the flow path of the material in the actual physical space in the construction flow data is taken as the basis for constructing the directed edge between the nodes in the graph, and the initial weight of the directed edge is given according to the average passing time of the historical material on the flow path and the current congestion state; the initial graph structure with node features and directed edge weights is input into the preset graph convolution network, each layer of the graph convolution network performs the following operations: each node aggregates the feature vectors of all the in-edge neighbor nodes, and updates the aggregated features based on the weight of the in-edge, the updated node features will be used for neighborhood aggregation of the next layer or as the final node representation; after multiple rounds of iteration and update, the graph structure with stable node features is output as the dynamic construction deduction graph.

[0073] In specific implementations, according to the distribution interval of the environmental disturbance response value in the link toughness spectrum, a high robustness interval, a medium robustness interval and a low robustness interval are divided. The division basis is a preset environmental disturbance response value threshold. For example, an interval where the environmental disturbance response value is less than 0.3 is set as the high robustness interval, an interval where the environmental disturbance response value is between 0.3 and 0.7 is set as the medium robustness interval, and an interval where the environmental disturbance response value is greater than 0.7 is set as the low robustness interval. In some embodiments, for hoisting operations in the high robustness interval, the component type and tool model corresponding to the hoisting operation in the high robustness interval are analyzed, and the average flow rate in the same period in the material flow trajectory is combined to calibrate the demand intensity for the hoisting tool and the specific component. The demand intensity is represented by a value between 0 and 1, and the larger the value, the stronger the demand. For hoisting operations in the low robustness interval, abnormal values of wind speed or illumination intensity exceeding the preset threshold in the environmental monitoring readings corresponding to the hoisting operation in the low robustness interval are extracted, the frequency and average amplitude of the wind speed or illumination intensity abnormal values are calculated, and the sensitivity to the construction environment is calibrated. The sensitivity is also represented by a value between 0 and 1. Optionally, the calculation of demand intensity and sensitivity can be integrated into a quantitative calibration function. The quantitative calibration function of demand intensity and sensitivity is expressed as:

[0074]

[0075] wherein the symbol represents the calibrated value for the construction resource , which will be separated into a demand intensity component or a sensitivity component in the subsequent; the symbol represents the number of hoisting operations using the resource in the high robustness interval; the symbol represents the total number of hoisting operations using the resource in the entire evaluation period; the symbol represents the average flow rate of materials involving the resource in the relevant period; the symbol represents a preset reference maximum flow rate; and the symbol represents the frequency of wind speed or illumination abnormal events corresponding to the resource in the low robustness interval; the symbol represents the average amplitude of the wind speed or illumination abnormal events corresponding to the resource in the low robustness interval; the symbol , , are preset weight coefficients, and . For hoisting tools and specific components, the results are mainly calculated according to the first two items of the formula, and are used as demand intensity. For "stable wind speed conditions" and "sufficient illumination conditions" and other environmental resources, the results are mainly calculated according to the third item of the formula, and are used as sensitivity.

[0076] In specific implementations, the demand intensity and the sensitivity are taken as different dimensions of a vector, and a multi-dimensional resource demand vector is constructed for each periodic stable segment of the construction link to be evaluated, for example, the multi-dimensional resource demand vector of a construction link can be expressed as [crane demand intensity: 0.85, steel column demand intensity: 0.72, stable wind speed sensitivity: 0.45]. The method for constructing a dynamic construction deduction graph with resource flow paths as edges and construction links as nodes includes the following steps: taking each construction link to be evaluated as a node in the graph, and the initial feature vector of the node is obtained by dimension reduction mapping of the corresponding multi-dimensional resource demand vector, and the dimension reduction mapping adopts the principal component analysis method. In some embodiments, the flow path of the material in the actual physical space in the construction flow data is taken as the basis for constructing the directed edges between nodes in the graph, for example, the material transportation path from the "steel member pre-assembly area" to the "main truss hoisting area" constitutes a directed edge, and the initial weight of the directed edge is assigned according to the average passing time of historical materials on the flow path and the current congestion state, and the product of the inverse of the average passing time and the congestion state evaluation coefficient is often used as the initial weight. Optionally, the initial graph structure with node features and directed edge weights is input into a preset graph convolution network, and each layer of the graph convolution network performs the following operations: each node aggregates the feature vectors of all its in-edge neighbor nodes, and updates the aggregated features based on the weights of the in-edges, and the updated node features will be used for neighborhood aggregation in the next layer or as the final node representation.

[0077] In the dynamic construction deduction graph, the sum of the edge weights of all directed edges and the inner product of the feature vectors of the two end nodes connected by the directed edges is calculated to obtain the flow efficiency evaluation value of each directed edge. All directed edges with a flow efficiency evaluation value lower than the system preset threshold value are identified and marked as potential bottleneck edges. Flow propagation simulation is performed on the path containing the potential bottleneck edge. If the simulation result shows that the cumulative delay of resources at the potential bottleneck edge exceeds the preset delay threshold value, the potential bottleneck edge and the complete path where it is located are determined as the critical bottleneck path. Based on the analysis result, the construction progress prediction and real-time resource allocation scheme is iteratively generated, and the iteration logic includes: in each iteration, the weight of the corresponding directed edge in the dynamic construction deduction graph is dynamically adjusted according to the currently identified critical bottleneck path, and the adjustment range of the weight is inversely proportional to the flow efficiency evaluation value of the critical bottleneck path; the node feature is aggregated and updated and the critical bottleneck path is analyzed using the dynamic construction deduction graph after adjusting the edge weight; the set of critical bottleneck paths obtained by the new analysis is compared with the set obtained in the last round. If the composition or number of the core bottleneck path changes, an updated resource pre-allocation instruction and progress correction amount are generated according to the new bottleneck path set; the updated resource pre-allocation instruction and progress correction amount are applied to the actual material flow trajectory and component hoisting event sequence scheduling, and new construction flow data generated thereby is collected as the input of the next iteration, until the set of critical bottleneck paths tends to be stable or reaches the maximum number of iterations, and the final construction progress prediction timeline and resource allocation scheme detail table are output. Using the dynamic construction deduction graph after adjusting the edge weight, the node feature is aggregated and updated and the critical bottleneck path is analyzed, which includes the following steps: inputting the dynamic construction deduction graph after adjusting the edge weight into the preset graph convolution network model to perform the aggregation and updating operation of the node feature, wherein each node aggregates the latest feature vector of all incoming edge neighbor nodes, and the aggregated feature vector is weighted and fused based on the current weight value of the incoming edge to generate an updated node feature vector; based on the updated node feature vector, the flow efficiency evaluation value of all directed edges in the dynamic construction deduction graph is recalculated; based on the recalculated flow efficiency evaluation value, all directed edges with a value lower than the system preset threshold value are identified and marked as potential bottleneck edges in the new round; the flow propagation simulation algorithm is performed on the complete path containing the potential bottleneck edge. If the simulation result shows that the cumulative delay time of resources at the potential bottleneck edge exceeds the preset delay threshold value, the path is determined as the critical bottleneck path.

[0078] In a specific implementation, in the dynamic construction deduction graph, the sum of the edge weight of all directed edges and the inner product of the feature vectors of the two end nodes connected by the directed edge is calculated to obtain the flow efficiency evaluation value of each directed edge, which is a scalar value. The process of analyzing the key bottleneck path in the graph is specifically to identify all directed edges whose flow efficiency evaluation value is lower than the system preset threshold, mark all directed edges lower than the system preset threshold as potential bottleneck edges, and perform flow propagation simulation on the path containing the potential bottleneck edge. The flow propagation simulation is an algorithm for simulating the process of resource or information flow on a graph. If the simulation result shows that the cumulative delay of resources at the potential bottleneck edge exceeds the preset delay threshold, the potential bottleneck edge and the complete path where it is located are determined as the key bottleneck path. Based on the analysis result, the construction progress prediction and real-time resource allocation scheme are iteratively generated, and the iteration logic includes that in each iteration, according to the currently identified key bottleneck path, the weight of the corresponding directed edge in the dynamic construction deduction graph is dynamically adjusted, and the edge weight adjustment formula is expressed as:

[0079]

[0080] Wherein, the symbol represents the adjusted directed edge weight; the symbol represents the unadjusted directed edge weight; the symbol represents the current flow efficiency evaluation value of the directed edge; the symbol represents the system preset flow efficiency evaluation value threshold; and the symbol is an adjustment coefficient greater than zero, which is used to control the rate of weight decay. The adjustment range of the weight is inversely proportional to the flow efficiency evaluation value of the key bottleneck path. The lower the flow efficiency evaluation value, the greater the weight down adjustment range. The dynamic construction deduction graph after adjusting the edge weight is used to update the aggregation of node features and analyze the key bottleneck path again.

[0081] In some embodiments, using the dynamic construction evolution graph with adjusted edge weights, the aggregation update of node features and the analysis of key bottleneck paths are performed again, including the following steps: inputting the dynamic construction evolution graph with adjusted edge weights into a preset graph convolution network model to perform the aggregation update of node features, wherein each node aggregates the latest feature vectors of all its in-edge neighbor nodes, and the aggregated feature vectors are weighted and fused based on the current weight values of the in-edges to generate updated node feature vectors. Based on the updated node feature vectors, the flow efficiency evaluation values of all directed edges in the dynamic construction evolution graph are recalculated. According to the recalculated flow efficiency evaluation values, all directed edges with values lower than the system preset threshold are identified and marked as potential bottleneck edges in the new round. The traffic propagation simulation algorithm is performed on the complete path containing the potential bottleneck edges, and if the simulation result shows that the accumulated delay time of resources at the potential bottleneck edges exceeds the preset delay threshold, the path is determined as a key bottleneck path. Optionally, the traffic propagation simulation algorithm adopts a queue-based discrete event simulation method, and resource packets are distributed to downstream nodes according to the weight proportion of the edges, and consume time on each edge in inverse proportion to the current edge weight. The specific implementation of the traffic propagation simulation algorithm adopts a queue-based discrete event simulation method, in which resource packets are modeled as discrete events, and the simulation process is advanced by maintaining an event queue. Resource packets flow from the source node of the dynamic construction evolution graph, and each resource packet is distributed to the downstream node according to the weight proportion of the directed edge, and the edge with higher weight obtains a larger proportion of resource packet distribution. On each directed edge, the time consumed by resource packets to pass through is inversely proportional to the current edge weight, that is, the higher the edge weight, the shorter the passing time. During the simulation process, the system continuously tracks the accumulated delay time of resource packets at each potential bottleneck edge, and compares the accumulated delay with the preset delay threshold to determine the key bottleneck path.

[0082] In a specific implementation, the key bottleneck path set obtained in the new round of analysis is compared with the set in the last round. If the composition or number of the core bottleneck path changes, an updated resource pre-allocation instruction and progress correction amount are generated according to the new bottleneck path set. The resource pre-allocation instruction is a scheduling command for adjusting the crane and transport vehicle resources to be inclined to the non-bottleneck path. The progress correction amount is a time adjustment amount for the original planned construction period. The updated resource pre-allocation instruction and progress correction amount are applied to the actual material flow trajectory and component hoisting event sequence scheduling, and new construction flow data generated thereby is collected as input for the next round of iteration, until the key bottleneck path set tends to be stable or reaches the maximum number of iterations, and the final construction progress prediction timeline and resource allocation scheme detailed table are output. In some embodiments, the comparison of the key bottleneck path set is realized by calculating the Jaccard similarity coefficient. If the Jaccard similarity coefficient of three consecutive iterations is higher than the stability threshold, it is determined that the set tends to be stable. During the iteration process, referring to Table 1, the key bottleneck path and its attributes resolved in each round can be recorded as shown in Table 1.

[0083] Table 1: Key bottleneck path analysis process table

[0084] Iteration round Critical bottleneck path (sequence of nodes) Cumulative latency (minutes) 1 Node A -> Node C -> Node F 45 2 Node A -> Node C 38 2 Node D -> Node F 52 3 Node D -> Node F 25

[0085] Optionally, the construction progress prediction timeline is presented in the form of a Gantt chart, indicating the start and end times of each construction link after adjustment. The resource allocation scheme detailed table lists the machine number, allocation time, and target position that need to be adjusted.

[0086] Referring to Figure 4 This is a comparison chart of the construction period before and after resource allocation for housing steel construction. Steel component hoisting (original 10 days → 8 days after allocation), material transportation (original 8 days → 7 days after allocation), component assembly (original 9 days → 7.5 days after allocation); foundation pouring (maintain 12 days); quality detection (original 5 days → 5.5 days after allocation). Such charts are used for the effect evaluation of construction resource allocation. By comparing the construction period changes in different links, the speed-up effect of resource optimization (such as machine / personnel inclination) on key links can be directly reflected, and the slightly delayed links after allocation that need to be paid attention to are identified, providing a basis for subsequent construction progress control.

[0087] Embodiment 5: Slice the material flow trajectory data with fixed time windows, calculate the variance and autocorrelation coefficient of the material flow transfer distance in each time window; judge and merge the sequence of time windows with continuous variance below the variance threshold and continuous autocorrelation coefficient above the correlation coefficient threshold as a periodic stationary paragraph; judge and merge the sequence of time windows with continuous variance above the variance threshold or continuous autocorrelation coefficient below the correlation coefficient threshold as a non-periodic fluctuation paragraph. In specific implementation, the method of identifying the periodic stationary paragraph and the non-periodic fluctuation paragraph in the material flow trajectory is to slice the material flow trajectory data with fixed time windows, the length of the fixed time window is set to 1 hour, the material flow trajectory data is a sequence recording the change of the position of the material in the three-dimensional space with time, and each time window contains a series of material coordinate points arranged in time sequence. The variance and autocorrelation coefficient of the material flow transfer distance in each time window are calculated, the variance is a statistical quantity measuring the fluctuation degree of the distance between adjacent coordinate points in the time window, and the autocorrelation coefficient is a statistical quantity measuring the correlation between the material flow transfer distance sequence itself and the lag version. In some embodiments, the material flow transfer distance is calculated by the Euclidean distance between consecutive coordinate points to form a distance sequence, the variance is calculated for this distance sequence, and the autocorrelation coefficient is calculated by using the lag 1 order Pearson correlation coefficient calculation method. Alternatively, the calculation formula of the autocorrelation coefficient is expressed as:

[0088]

[0089] wherein the symbol represents the calculated autocorrelation coefficient value, which ranges from -1 to 1; the symbol represents the total number of data points of the material flow transfer distance sequence in the current fixed time window; the symbol represents the th material flow transfer distance value in the sequence; the symbol represents the th material flow transfer distance value in the sequence, i.e. the data lagging one order; and the symbol represents the arithmetic mean of all material flow transfer distance values in the current fixed time window.

[0090] In a specific implementation, a sequence of time windows with continuous variance lower than a variance threshold and continuous autocorrelation coefficient higher than a correlation coefficient threshold is determined and merged into a periodic stationary paragraph, the variance threshold is a preset numerical constant, the correlation coefficient threshold is another preset numerical constant, and continuous means that multiple time windows are adjacent in time and all satisfy the conditions. In some embodiments, if the variance values of time window 1, time window 2, and time window 3 are 0.5, 0.6, and 0.55, respectively, and the autocorrelation coefficient values are 0.85, 0.82, and 0.88, respectively, the three time windows are merged into a periodic stationary paragraph. A sequence of time windows with continuous variance higher than the variance threshold or continuous autocorrelation coefficient lower than the correlation coefficient threshold is determined and merged into a non-periodic fluctuation paragraph. Optionally, if the variance values of time window 4 and time window 5 are 2.1 and 1.9, respectively, regardless of their autocorrelation coefficient values, the two time windows can be merged into a non-periodic fluctuation paragraph; if the variance values of time window 6 and time window 7 are not higher than the variance threshold, but the autocorrelation coefficients are 0.4 and 0.3, respectively, the two time windows can also be merged into a non-periodic fluctuation paragraph.

[0091] Referring to Figure 5 This is a time series comparison chart of the material flow transfer distance sequence of a house steel structure. The lag 1 order sequence is a time delay version of the original sequence; the fluctuation trends of the two are highly consistent, and the distance values of the corresponding points are similar, indicating that the material flow transfer distance has strong time correlation. Such a chart is used in the autocorrelation coefficient calculation link of the material flow transfer trajectory. By comparing the correlation of the original sequence and the lag sequence, the periodicity of the material flow transfer can be determined (the higher the correlation, the stronger the periodicity), providing data basis for subsequent identification of "periodic stationary paragraphs".

[0092] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0093] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A deep learning based steel building construction progress and deployment system, characterized in that, The system comprises the following processing procedures: A data acquisition module acquires real-time construction flow data of multiple sources and heterogeneous in the process of building steel construction, and the construction flow data covers material flow trajectory, component hoisting event sequence and environmental monitoring readings; A link identification module identifies periodic stable paragraphs and non-periodic fluctuation paragraphs in the material flow trajectory based on the time sequence form of the construction flow data, and accordingly separates a plurality of to-be-evaluated construction links from all construction activities; A toughness spectrum module integrates the component hoisting event sequence in the periodic stable paragraph for each to-be-evaluated construction link, and cross-maps the environmental monitoring readings in the non-periodic fluctuation paragraph associated with the construction link, to generate a link toughness spectrum representing construction robustness; A resource calibration module calibrates the demand intensity and sensitivity of each construction resource in the periodic stable paragraph according to the link toughness spectrum, to form a multi-dimensional resource demand vector; A graph deduction module drives node feature updating of a graph convolution network using the multi-dimensional resource demand vector, to construct a dynamic construction deduction graph with resource flow paths as edges and construction links as nodes; A prediction and deployment module analyzes key bottleneck paths in the dynamic construction deduction graph by performing multiple rounds of message passing and neighborhood aggregation, and iteratively generates construction progress prediction and real-time resource deployment schemes based on the analysis results. 2.The deep learning-based housing steel construction progress prediction and scheduling system of claim 1, wherein The component hoisting event sequence includes the start and end time stamps of each hoisting operation, the spatial coordinate displacement of the hoisted component, and the load rate change curve of the hoisting machine; and the environmental monitoring readings include continuous sampling values of wind speed and direction, time sequence records of light intensity, and measurement data stream of temperature and humidity sensors. 3.The deep learning-based housing steel construction progress prediction and deployment system of claim 2, wherein, The generation steps of the link toughness spectrum specifically include: For any to-be-evaluated construction link, integrate all component hoisting event sequences in the periodic stable paragraph into a link internal event set, and integrate environmental monitoring readings in the non-periodic fluctuation paragraph of all other construction links having material supply or spatial adjacent relationship with the to-be-evaluated construction link into an associated external disturbance set; Extract the morphological features of the load rate change curve of each hoisting operation in the link internal event set, and simultaneously extract the statistical features of the continuous sampling values of wind speed and direction in the same period in the associated external disturbance set, and calculate the morphological coupling coefficient of the morphological features and the statistical features; Perform convolution operation on the spatial coordinate displacement and the morphological coupling coefficient corresponding to each hoisting operation to obtain the environmental disturbance response value of the hoisting operation; Aggregate the environmental disturbance response values of all hoisting operations in the link internal event set, and perform time alignment and smoothing processing according to the start time stamp of the hoisting operation, to generate the link toughness spectrum of the to-be-evaluated construction link.

4. The deep learning-based housing steel construction construction progress and deployment system according to claim 3, wherein, The demand intensity and sensitivity calibration of each construction resource in the periodic stable paragraph is specifically implemented as follows: According to the distribution interval of the environmental disturbance response value in the link toughness spectrum, divide the high-robustness interval, the medium-robustness interval and the low-robustness interval; For the hoisting operation in the high-robustness interval, analyze the corresponding component type and machine type, and calibrate the demand intensity for the hoisting machine and the specific component in combination with the average flow rate in the period in the material flow trajectory; For hoisting operations in the low robustness interval, extract the wind speed or light intensity outliers in the corresponding environmental monitoring readings that exceed the preset threshold, calculate the frequency and average amplitude of the wind speed or light intensity outliers, and use them to determine the sensitivity to the construction environment; The demand intensity and sensitivity are used as different dimensions of a vector to construct a multi-dimensional resource demand vector for each periodic and smooth segment of the construction process to be evaluated.

5. The deep learning-based housing steel construction construction progress and deployment system according to claim 4, wherein, The method for constructing a dynamic construction deduction graph with resource flow paths as edges and construction steps as nodes includes the following steps: Each construction step to be evaluated is used as a node in the graph, and the initial feature vector of the node is obtained by dimensionality reduction mapping of the corresponding multi-dimensional resource demand vector; The flow path of the material in the actual physical space in the construction flow data is used as the basis for constructing the directed edges between nodes in the graph, and the average passing time of historical materials on the flow path and the current congestion state are used to assign initial weights to the directed edges; The initial graph structure with node features and directed edge weights is input into a preset graph convolution network, and each layer of the graph convolution network performs the following operations: each node aggregates the feature vectors of all its incoming edge neighbor nodes, and updates the aggregated features based on the weights of the incoming edges, and the updated node features are used for neighborhood aggregation in the next layer or as the final node representation; After multiple rounds of iterative updates, the graph structure with stable node features is output as the dynamic construction deduction graph. 6.The deep learning-based house steel construction progress prediction and deployment system according to claim 5, wherein, The process of analyzing key bottleneck paths in the graph is as follows: In the dynamic construction deduction graph, calculate the sum of the edge weights of all directed edges and the inner product of the feature vectors of the two end nodes connected by the directed edges to obtain the flow efficiency evaluation value of each directed edge; Identify all directed edges with a flow efficiency evaluation value below the system's preset threshold, and mark all directed edges below the system's preset threshold as potential bottleneck edges; Perform traffic propagation simulation on the path containing the potential bottleneck edge, and if the simulation result shows that the cumulative delay of resources at the potential bottleneck edge exceeds the preset delay threshold, the potential bottleneck edge and the complete path it is in are determined as the key bottleneck path. 7.The deep learning-based house steel construction progress prediction and deployment system according to claim 6, wherein, The iterative logic for generating construction progress prediction and real-time resource allocation scheme based on the analysis results includes: In each iteration, dynamically adjust the weights of the corresponding directed edges in the dynamic construction deduction graph based on the currently identified key bottleneck path, and the adjustment amplitude of the weights is inversely proportional to the flow efficiency evaluation value of the key bottleneck path; Use the dynamic construction deduction graph with adjusted edge weights to re-aggregate and update the node features and analyze the key bottleneck paths; Compare the set of key bottleneck paths obtained in the new iteration with the set obtained in the previous iteration, and if the composition or number of the core bottleneck paths changes, generate updated resource pre-allocation instructions and progress correction amounts based on the new set of bottleneck paths. The updated resource pre-assignment instruction is applied to the actual material flow trajectory and the component hoisting event sequence scheduling, and new construction flow data generated thereby is collected as input for the next round of iteration, until the key bottleneck path set tends to be stable or the maximum number of iteration rounds is reached, and a final construction progress prediction timeline and resource allocation scheme detailed table is output. 8.The deep learning-based house steel construction progress prediction and deployment system according to claim 1, wherein, The method for identifying periodic stable segments and non-periodic fluctuation segments in the material flow trajectory is: The material flow trajectory data is sliced with a fixed time window, and the variance and autocorrelation coefficient of the material flow transfer distance in each time window are calculated; The sequence of time windows with continuous variance below the variance threshold and continuous autocorrelation coefficient above the correlation coefficient threshold is determined and merged into a periodic stable segment; The sequence of time windows with continuous variance above the variance threshold or continuous autocorrelation coefficient below the correlation coefficient threshold is determined and merged into a non-periodic fluctuation segment. 9.The deep learning-based house steel construction progress prediction and deployment system according to claim 3, wherein, The method for calculating the morphological coupling coefficient of the morphological feature and the statistical feature is further refined as: Wavelet packet decomposition is performed on the load rate change curve, and the energy of the specified frequency band is extracted as the load morphological feature vector; Empirical mode decomposition is performed on the continuous sampling values of wind speed and direction, and a number of intrinsic mode functions are obtained. The sample entropy of the first few intrinsic mode functions is calculated as the wind speed morphological feature vector; The cosine similarity of the load morphological feature vector and the wind speed morphological feature vector in the standard orthogonal basis is calculated, and the value obtained by normalizing the cosine similarity through the S-shaped function is defined as the morphological coupling coefficient. 10.The deep learning-based house steel construction progress prediction and deployment system according to claim 7, wherein, The adjusted dynamic construction deduction graph is used to re-aggregate and update the node features and analyze the key bottleneck paths, which includes the following steps: The adjusted dynamic construction deduction graph is input into the pre-set graph convolution network model to perform node feature aggregation and update operation, wherein each node aggregates the latest feature vector of all incoming edge neighbor nodes, and the aggregated feature vector is weighted and fused based on the current weight value of the incoming edge to generate an updated node feature vector; Based on the updated node feature vector, the flow efficiency evaluation value of all directed edges in the dynamic construction deduction graph is recalculated; According to the recalculated flow efficiency evaluation value, all directed edges with a value below the system preset threshold are identified and marked as potential bottleneck edges in the next round. The traffic propagation simulation algorithm is executed on the complete path containing the potential bottleneck edge, and if the simulation result shows that the accumulated delay time of resources at the potential bottleneck edge exceeds the preset delay threshold, the path is determined as a key bottleneck path.

Images