Fabricated pipeline construction progress prediction method based on deep learning

Abstract

The invention relates to the technical field of building construction, and discloses a fabricated pipeline construction progress prediction method based on deep learning. According to the method, multi-source engineering data, including structured construction parameters, unstructured construction logs and the like, of fabricated pipeline construction are collected, and a standardized feature set is generated through preprocessing; extracting feature vectors of each data source by using a depth feature extraction model, constructing a construction progress knowledge base, generating global fusion features through multi-source feature fusion, and predicting a construction process sequence and progress by combining knowledge base data and by means of a time sequence prediction algorithm. Meanwhile, analyzing sensor data by adopting wavelet transform, and correcting a process sequence by utilizing a grey system theory when the sensor data is abnormal; and constructing a construction specification knowledge graph, and verifying the consistency of the process sequence and the specification by using a graph embedding algorithm. The method can accurately predict the construction progress, responds to abnormal conditions in real time, ensures that the construction accords with specifications, and improves the construction management level of the fabricated pipeline.

Description

Technical Field

[0001] The present invention relates to the field of building construction technology, and in particular to a method for predicting the construction progress of an assembled pipeline based on deep learning. Background Art

[0002] In the construction industry, prefabricated pipeline construction, with its advantages of high efficiency and environmental friendliness, has gradually become a key development direction for the industry. With the acceleration of urbanization, the scale of construction projects continues to expand and their structures become increasingly complex. Precisely controlling the progress of prefabricated pipeline construction has become a key factor in ensuring smooth project progress, reducing costs, and improving overall efficiency. However, current prefabricated pipeline construction progress management still faces many challenges.

[0003] Traditional construction progress prediction methods rely heavily on empirical evidence and simple mathematical models. These methods struggle to fully account for the complex factors inherent in prefabricated pipeline construction. Prefabricated pipeline construction involves numerous steps, such as prefabrication, transportation, and installation, each of which is influenced by multiple factors. The spatial layout of the construction site can restrict the stacking and lifting of pipelines, and the investment in manpower and equipment resources at different construction stages can significantly impact the construction progress. Traditional methods are unable to effectively integrate these complex data sources, including construction parameters, unstructured information from construction logs, and spatial information derived from the three-dimensional layout of pipelines, leading to significant deviations in prediction results.

[0004] On the other hand, traditional methods lack the ability to track dynamic changes during the construction process in real time. In actual construction, unexpected events such as weather changes, equipment failures, and design changes often occur. When these events occur, traditional forecasting models are unable to adjust their predictions in a timely manner, resulting in a significant disconnect between the construction schedule and actual construction. For example, inclement weather can delay outdoor pipeline installation, while sudden equipment failures can stall construction. However, traditional forecasting methods struggle to respond quickly to these changes, making it difficult for construction teams to make timely and appropriate adjustments, which in turn impacts the entire project's duration and cost.

[0005] In addition, with the advancement of the digital transformation of the construction industry, the application of building information modeling (BIM) technology in prefabricated pipeline construction has gradually become popular. Although BIM technology can provide a three-dimensional visual model of the pipeline, existing construction progress prediction methods fail to fully utilize this advantage. It is impossible to deeply integrate the geometric features of the pipeline layout in the BIM model with other construction data, and it is impossible to achieve refined prediction and management of the construction progress. Moreover, the existing prediction methods are also difficult to meet the construction industry's higher requirements for construction progress prediction accuracy, real-time and intelligence, which restricts the further development and application of prefabricated pipeline construction technology. Therefore, it is urgent to develop a method that can effectively integrate multi-source engineering data, track construction dynamic changes in real time, and accurately predict construction progress. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for predicting the construction progress of assembled pipelines based on deep learning to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for predicting the construction progress of an assembled pipeline based on deep learning, the method comprising:

[0009] Collect multi-source engineering data during the construction of prefabricated pipelines, including structured construction parameters, unstructured construction logs, pipeline 3D layout models, and real-time monitoring sensor data;

[0010] Preprocessing the multi-source engineering data to generate a standardized feature set, wherein the standardized feature set includes pipeline layout geometric features, construction parameter time series features, log text semantic features, and sensor monitoring dynamic features;

[0011] Extracting feature vectors from each data source based on a deep feature extraction model, wherein the deep feature extraction model includes a three-dimensional convolutional network, a time series encoder, and a text embedding network;

[0012] Building a construction progress knowledge base, which stores historical construction case data, including pipeline construction process sequences, logical relationships between processes, and corresponding historical progress records;

[0013] Mapping the feature vectors to a unified semantic space through a multi-source feature fusion method to generate a global fusion feature;

[0014] Based on the global fusion features and the historical data in the construction progress knowledge base, a time series prediction algorithm is used to generate a process sequence and progress prediction result of the current pipeline construction.

[0015] Preferably, the method of generating the process sequence and progress prediction results of the current pipeline construction using a time series prediction algorithm includes:

[0016] generating an initial candidate process set according to the global fusion feature and the correlation between the processes in the knowledge base;

[0017] Building a construction process network based on the logical relationship between the processes, wherein nodes in the network represent processes and edges represent logical dependencies or resource constraints between processes;

[0018] The time series data is segmented by a sliding window mechanism, combined with the construction process network, and the initial candidate process set is optimized using a recursive neural network to output the final process sequence and progress prediction value.

[0019] Preferably, the preprocessing of the multi-source engineering data includes:

[0020] A gridding algorithm is used to generate spatial topological features of the pipeline three-dimensional layout model, a one-dimensional convolutional network is used to extract local fluctuation features of the construction parameter time series features, and a bidirectional gated recurrent unit is used to extract context-related features of the log text semantic features.

[0021] Preferably, mapping the feature vector to a unified semantic space by a multi-source feature fusion method includes:

[0022] Calculate the weight distribution of features from different data sources through a multi-head attention mechanism;

[0023] The weighted features are input into the residual connection network to fuse multi-scale features and eliminate redundant information.

[0024] Preferably, the time series prediction algorithm further includes:

[0025] A progress optimization model is constructed based on a generative adversarial network. The model dynamically adjusts the rationality of the prediction sequence through a discriminator. The discriminator input includes process timing continuity, resource occupancy rate and construction specification compliance.

[0026] Preferably, the recursive neural network is an improved long short-term memory network, comprising:

[0027] Encoding the construction process network into a memory unit state, wherein the state includes timestamps and dependency relationships of process nodes;

[0028] The gating mechanism is used to control the forgetting and updating of temporal information, and skip connections are introduced to alleviate the gradient vanishing problem.

[0029] The network weights are adjusted by backpropagating the prediction error, which is calculated as the root mean square deviation of the actual progress from the predicted value.

[0030] Preferably, the method further includes: using wavelet transform analysis on the real-time monitoring sensor data to extract abnormal signal characteristics during the construction process; when an abnormal signal is detected, triggering real-time correction of the process sequence, and updating the progress prediction result according to the corrected sequence.

[0031] Preferably, the real-time correction adopts grey system theory, and the specific method includes:

[0032] Define the grey correlation function of process adjustment threshold, resource reallocation rules and construction interruption tolerance;

[0033] Correction candidate plans are generated through the grey prediction model, and the optimal adjustment strategy is selected based on the whitening weight function.

[0034] Preferably, the method further comprises:

[0035] Constructing a construction specification knowledge graph, where nodes represent construction standard clauses and edges represent logical relationships or conflict constraints between clauses;

[0036] During the progress prediction process, the consistency between the process sequence and the knowledge graph is verified through a graph embedding algorithm.

[0037] Preferably, the graph embedding algorithm is a representation learning method based on a hierarchical graph neural network, including: extracting subgraph structural features from the current process sequence, hierarchically aggregating node neighbor information in the knowledge graph, and generating a global consistency vector through graph pooling operations.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The deep learning-based prefabricated pipeline construction progress prediction method proposed in this invention optimizes and innovates the prefabricated pipeline construction progress management from multiple aspects, bringing significant beneficial effects. In terms of data processing and feature extraction, this method collects multi-source engineering data such as structured construction parameters, unstructured construction logs, pipeline three-dimensional layout models, and real-time monitoring sensor data during the prefabricated pipeline construction process, and pre-processes them to generate a standardized feature set, which comprehensively and accurately reflects all aspects of the construction process. For example, the use of a gridding algorithm to generate spatial topological features from the pipeline three-dimensional layout model can fully explore the geometric information of the pipeline layout and provide a basis for subsequent analysis; a one-dimensional convolutional network is used to extract the local fluctuation characteristics of the construction parameter time series characteristics, effectively capturing the change patterns of construction parameters over time; and a bidirectional gated recurrent unit is used to extract the context-related features of the log text semantic features to fully understand the key information in the construction log. These operations ensure the integrity and accuracy of the data, laying a solid foundation for accurate prediction.

[0040] The use of deep feature extraction models further improves the quality of feature extraction. Three-dimensional convolutional networks, time series encoders, and text embedding networks extract feature vectors from different data sources, enabling deep exploration of latent features within the data and making subsequent feature fusion and prediction more accurate. For example, the 3D convolutional network can better process the spatial information in the 3D pipeline layout model, the time series encoder can accurately grasp the temporal changes in construction parameters, and the text embedding network can deeply understand the semantic content of construction logs. All of these contribute to a more comprehensive understanding of the construction status.

[0041] Regarding construction progress prediction, the constructed construction progress knowledge base stores a wealth of historical construction case data, including pipeline construction process sequences, logical relationships between processes, and corresponding historical progress records. A multi-source feature fusion method maps feature vectors from different data sources into a unified semantic space to generate a global fusion feature. Combined with the historical data in the knowledge base, a time series prediction algorithm is used to generate the process sequence and progress prediction results for the current pipeline construction. This approach leverages historical experience while integrating it with the actual conditions of current construction, significantly improving prediction accuracy. For example, when faced with similar construction conditions and process schedules, historical cases can be quickly referenced and adjusted based on current real-time data to produce more realistic prediction results.

[0042] In addition, various innovative technologies in the time series forecasting algorithm further optimize the forecasting process. A progress optimization model based on a generative adversarial network dynamically adjusts the rationality of the forecast sequence through a discriminator. The discriminator inputs key indicators such as process time sequence continuity, resource utilization, and compliance with construction specifications, ensuring that the forecast results conform to construction logic while meeting resource utilization and regulatory requirements. An improved long-short-term memory network encodes the construction process network as memory cell states, controls the forgetting and updating of time series information through a gating mechanism, and introduces skip connections to alleviate the vanishing gradient problem. This allows for better processing of long-sequence data and improves the stability and reliability of forecasts.

[0043] To address anomalies during construction, wavelet transform analysis is applied to real-time monitoring sensor data, enabling timely extraction of abnormal signal characteristics. Once an abnormal signal is detected, gray system theory is used to make real-time corrections to the process sequence and update the progress forecast. This real-time response mechanism helps construction teams quickly adjust their plans, minimizing delays and increased costs caused by abnormalities. For example, if sensors detect an equipment anomaly, the system can quickly initiate a correction process, re-plan the process sequence, and rationally allocate resources to ensure continuous and efficient construction.

[0044] The construction of a construction specification knowledge graph and its application in the progress forecasting process ensures that the construction process sequence complies with construction standards. A graph embedding algorithm based on a hierarchical graph neural network verifies the consistency of the process sequence with the knowledge graph, thus preventing construction problems caused by violations of construction specifications at the source and improving construction quality and safety. For example, when scheduling construction procedures, the system automatically checks for compliance with relevant specifications, promptly identifying and correcting potential violations to ensure smooth project progress. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a working principle diagram of the method for predicting the construction progress of assembled pipelines based on deep learning according to the present invention;

[0046] Figure 2 Flowchart for preprocessing multi-source engineering data;

[0047] Figure 3 Flowchart for fusing multi-source features into a unified semantic space;

[0048] Figure 4 Flowchart for adjusting process sequence and progress forecast based on sensor data anomalies. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] See also Figures 1-4 The present invention provides a method for predicting the construction progress of an assembled pipeline based on deep learning, and its specific implementation method will be described in detail below.

[0051] During the construction of prefabricated pipelines, various types of data are collected extensively. These include structured construction parameters, such as the dimensions of the construction site, the specifications and models of the pipelines, and the parameters of the construction equipment; unstructured construction logs, which cover daily construction status records, personnel changes, and problems encountered; pipeline three-dimensional layout models, which accurately present the position and direction of the pipelines in space; and real-time monitoring sensor data, such as data on monitoring the construction environment temperature, humidity, pipeline pressure, etc.

[0052] The collected multi-source engineering data is preprocessed and converted into a standardized feature set. Pipeline layout geometric features are obtained by processing the three-dimensional pipeline layout model; construction parameter temporal features are extracted from structured construction parameters; log text semantic features are derived from unstructured construction logs; and sensor monitoring dynamic features are derived from real-time monitoring sensor data. These features reflect information about the construction process from different perspectives.

[0053] A deep feature extraction model, comprising a 3D convolutional network, a time series encoder, and a text embedding network, processes each data source and extracts corresponding feature vectors. The 3D convolutional network extracts spatial features from the 3D pipeline layout model, the time series encoder analyzes the temporal characteristics of construction parameters, and the text embedding network mines the semantic features of log text.

[0054] Build a construction progress knowledge base, which stores a large amount of historical construction case data. These case data include the pipeline construction process sequence, that is, the sequence of each process in the construction process; the logical relationship between processes, such as some processes must start after other processes are completed; and the corresponding historical progress records, which record the time progress of each process in the actual construction.

[0055] Through the multi-source feature fusion method, the extracted feature vectors are mapped to a unified semantic space to generate global fusion features. This process enables features from different sources to be effectively fused in the same space, providing more comprehensive and accurate information for subsequent predictions.

[0056] Based on the generated global fusion features and the historical data in the construction progress knowledge base, a time series prediction algorithm is used to generate the process sequence and progress prediction results of the current pipeline construction. In this way, it is possible to combine historical experience and the actual situation of the current construction to make a relatively accurate prediction of the construction progress.

[0057] Example 1:

[0058] When preprocessing multi-source engineering data, the following specific methods are adopted. For the pipeline three-dimensional layout model, a grid division algorithm is used to generate spatial topology features. This algorithm divides the pipeline three-dimensional layout model into multiple grid cells, and constructs a topological structure reflecting the pipeline spatial layout by analyzing the connection relationship, adjacent relationship, etc. between these grid cells. This topological structure can clearly show the relative positions and connection methods of each part of the pipeline in space, providing an important basis for subsequent analysis.

[0059] For the time series features of construction parameters, a one-dimensional convolutional network is used to extract local fluctuation features. The one-dimensional convolutional network slides the convolutional kernel on the time series to extract local features of the change of construction parameters over time. Let the input time series data of construction parameters be X = [x1, x2, …, x n , and the convolutional kernel be W = [w1, w2, …, w m , where n is the length of the time series and m is the size of the convolutional kernel, and m < n. After the convolutional operation, the obtained feature map Y = [y1, y2, …, y n-m+1 , where In this way, it is possible to capture the fluctuation trends and change characteristics of construction parameters in local time periods, such as the sudden acceleration or deceleration of construction progress, the abnormal fluctuation of construction equipment parameters, etc.

[0060] For the semantic features of log text, a bidirectional gated recurrent unit (Bi-GRU) is used to extract contextual features. Bi-GRU processes text sequences from both forward and reverse directions, making full use of the information before and after the text. Suppose the input text sequence is S = [s1, s2, ..., s T ], where T is the length of the text sequence. In the forward propagation process, the GRU unit is based on the current input s t and the hidden state at the previous moment Calculate the hidden state at the current moment During the back propagation process, according to the current input s T-t+1 And the reverse hidden state of the previous moment Calculate the reverse hidden state at the current moment The final hidden state h t It is composed of the forward and reverse hidden states, that is, This can help us better understand the contextual semantics of each word in the text and extract key information related to construction, such as problems encountered during construction and solutions taken.

[0061] Example 2:

[0062] When mapping feature vectors to a unified semantic space through multi-source feature fusion, the specific operation is as follows. First, the weight distribution of features from different data sources is calculated through a multi-head attention mechanism. The multi-head attention mechanism allows the model to learn different feature relationships in different representation subspaces. Assume that the input feature vectors are F1, F2, ..., F k , where k is the number of data sources. For each head i, calculate the attention score A i , where Q i , K i The query matrix and key matrix are obtained by linearly transforming the input feature vector, d k is the dimension of the key matrix. Attention score A i It reflects the degree of correlation between different feature vectors. The output of each head is obtained by weighted summing of the attention scores. i , O i =A i V i , where V i Is a value matrix. The outputs of multiple heads are concatenated to obtain the final weighted feature F weighted .

[0063] Next, the weighted features are input into the residual connection network to fuse multi-scale features and eliminate redundant information. The residual connection network uses jump connections to enable the network to directly learn the residual between input and output, effectively avoiding the gradient vanishing problem and enhancing the network's ability to fuse features of different scales. Let the weighted features be F weighted After a series of convolutional layers and activation functions in the residual connection network, the fused feature F is obtained. fused In this process, convolutional layers at different levels can extract features of different scales. For example, shallow convolutional layers extract local detail features, while deep convolutional layers extract global abstract features. These features are fused together through residual connections, removing redundant information and retaining the most valuable features for construction progress prediction.

[0064] Example 3:

[0065] When using the time series prediction algorithm to generate the process sequence and progress prediction results of the current pipeline construction, the specific steps are as follows. First, based on the correlation between the global fusion features and the processes in the knowledge base, an initial set of candidate processes is generated. By calculating the similarity between the global fusion features and the features of each process in the knowledge base, the processes most relevant to the current construction situation are found to form the initial set of candidate processes. For example, cosine similarity can be used to measure the similarity between features. Let the global fusion feature be G and the feature of process j in the knowledge base be F. j , then the similarity The processes with higher similarity are selected as the initial candidate processes.

[0066] Next, a construction process network is constructed based on the logical relationships between processes. In this network, nodes represent processes, and edges represent logical dependencies or resource constraints between processes. For example, if process A must complete before process B can begin, there is a directed edge from process A to process B. Similarly, if processes C and D require the same resource, and that resource can only be used by one process at a time, there is a resource constraint edge between processes C and D. By constructing such a network, the relationships between processes can be clearly represented, providing a foundation for subsequent optimization.

[0067] Finally, the time series data is segmented by the sliding window mechanism, combined with the construction process network, and the recursive neural network is used to optimize the initial candidate process set, outputting the final process sequence and progress prediction value. Assume that the time series data is T = [t1, t2,…, t N ], the sliding window size is w, then each time a subsequence of length w is taken from the time series data [t i ,t i+1 ,…,t i+w-1] for processing. When processing these subsequences, the recurrent neural network combines information from the construction process network to continuously adjust the initial set of candidate processes. During this process, the recurrent neural network predicts the next process based on the current input and the previous state. By comparing the predictions with the actual situation and calculating the error, the network weights are continuously adjusted, ultimately outputting a final process sequence and progress forecast that aligns with construction logic and actual conditions.

[0068] Example 4:

[0069] In this method's time series prediction algorithm, a progress optimization model based on a generative adversarial network is introduced to improve the accuracy and rationality of predictions. A generative adversarial network primarily consists of a generator and a discriminator, which collaborate and compete with each other.

[0070] The generator's task is to generate a predicted sequence of construction steps and corresponding progress estimates based on the input global fusion features and information from the construction progress knowledge base. This is like a "forecasting assistant," drawing on existing construction experience (knowledge base information) and comprehensive information collected from the current construction site (global fusion features) to infer the schedule for the next construction steps and the approximate progress of each step.

[0071] The discriminator acts like a quality inspector, determining whether the generator's results are reasonable. Its criteria for this assessment include process timing continuity, resource utilization, and compliance with construction regulations.

[0072] Process temporal continuity refers to whether the generated process sequence conforms to the normal logical order of construction. For example, in prefabricated pipeline construction, the pipeline foundation laying process must be completed before the subsequent connection and fixing processes can proceed. If the process sequence generated by the generator places the connection process before the foundation laying, this is considered a process temporal discontinuity and does not meet actual construction requirements. The discriminator will identify this unreasonable situation.

[0073] Resource utilization is also a key factor considered by the discriminator. Construction site resources, such as manpower, machinery, and materials, are limited. Each process consumes a certain amount of resources during the construction process. If the generated process sequence is not arranged properly, the demand for a certain resource within a certain time period will be excessive, exceeding the actual supply range, which will affect the normal progress of construction. The discriminator checks whether the resource utilization rate is within a reasonable range to ensure that the generated process sequence is feasible in terms of resource utilization.

[0074] Compliance with construction regulations is also crucial. Construction has strict regulations and standards, encompassing aspects such as construction safety and quality. The discriminator compares the generated process sequence with the construction specification knowledge graph to determine whether each process complies with the corresponding construction standard clauses. For example, in pipeline construction, there are clear regulations regarding installation angles and fixing methods. If a process in the generated process sequence does not comply with these regulations, the discriminator will identify it.

[0075] Throughout the optimization process, the generator continuously attempts to generate more reasonable process sequences and progress forecasts to pass the discriminator's "inspection." The discriminator, in turn, continuously improves its judgment, more accurately identifying unreasonable results. Through this "competitive" process, their mutual progress ensures that the resulting process sequences and progress forecasts increasingly align with actual construction conditions, significantly improving the reliability of construction progress forecasts.

[0076] Example 5:

[0077] In the actual application of this method, an improved long short-term memory network (LSTM) is used as a recurrent neural network to optimize the process sequence prediction and improve the accuracy of progress prediction.

[0078] The construction process network is encoded as a memory cell state. This memory cell state acts like an information repository, storing the timestamps of each process node and the dependencies between them. For example, a construction process network consists of multiple process nodes, each representing a task. The timestamp indicates the expected start or completion time of the task, while the dependencies indicate the logical connections between tasks, such as which tasks precede or follow each other, and which tasks depend on each other for execution. By integrating this critical information into a memory cell state, the network can clearly "remember" the sequence and timing of construction processes.

[0079] In the improved LSTM, a gating mechanism controls the forgetting and updating of temporal information. This works like an intelligent information filter. The forget gate determines which previous information should be discarded based on the actual situation. The input gate determines whether new information should be stored in the memory cell. The output gate controls the final output information. For example, if the progress of a certain link in the construction process changes, the forget gate may choose to forget some old progress-related information. The input gate adds the new progress data, and the output gate transmits the integrated information, allowing the network to make more accurate judgments based on the latest situation.

[0080] To prevent the vanishing gradient problem, skip connections were introduced. Vanishing gradients are like information becoming increasingly blurred during transmission, ultimately becoming useless. Skip connections, however, act like a "green channel" for information, allowing it to be transmitted more directly to deeper layers of the network. This prevents information loss even in densely populated networks, ensuring effective training and accurate predictions.

[0081] During network training, network weights are adjusted based on the backpropagation of prediction errors. Prediction error is calculated based on the difference between actual progress and predicted values. For example, if construction has already reached a certain stage, but the predicted progress differs from the actual progress, this difference is the error. By continuously calculating this error and propagating this information back to the network, the network can identify where its predictions were inaccurate and then adjust the weights of the connections between nodes to ensure that the next prediction is closer to the actual situation, thereby improving the accuracy of construction progress predictions.

[0082] Example 6:

[0083] This implementation also includes processing real-time monitoring sensor data, real-time correction of process sequences, and updating of progress prediction results. First, wavelet transform analysis is performed on the real-time monitoring sensor data to extract abnormal signal characteristics during the construction process. Wavelet transform can analyze signals at different time and frequency scales, effectively capturing sudden changes and anomalies in the signal. Let the sensor data be x(t) and the wavelet function be ψ(t), then the wavelet transform coefficients W(a,b) are: Where a is the scale parameter, which controls the expansion and contraction of the wavelet function, b is the translation parameter, which controls the position of the wavelet function, and ψ * By analyzing the wavelet transform coefficients, we can find abnormal fluctuations in the data, such as a sudden increase in temperature or pressure outside the normal range.

[0084] When an abnormal signal is detected, the real-time correction of the process sequence is triggered, and the progress forecast result is updated according to the corrected sequence. The real-time correction adopts the grey system theory, and the specific method is as follows. Define the grey correlation function of the process adjustment threshold, resource reallocation rule and construction interruption tolerance. Let the process adjustment threshold be θ1, the resource reallocation rule be a set of conditions and operations, and the construction interruption tolerance be θ2. Grey correlation function r(X0,X i ) is used to measure the difference between the reference sequence X0 and the comparison sequence X i The degree of correlation between them is calculated as follows: where x0(k) and x i (k) are the values of the reference sequence and the comparison sequence at the kth moment, respectively, and ρ is the resolution coefficient, which is usually 0<ρ<1.

[0085] A gray prediction model generates candidate correction options, and the optimal adjustment strategy is selected based on a whitening weight function. The gray prediction model predicts future development trends based on existing data and generates multiple possible process adjustment options. The whitening weight function is used to evaluate and select these options, selecting the one that best matches the current construction situation and has the least impact on the construction progress as the optimal adjustment strategy. The process sequence is then revised based on the optimal adjustment strategy, and the progress forecast is recalculated to ensure that the construction progress forecast promptly reflects changes in actual construction.

[0086] In addition, a construction specification knowledge graph is constructed, in which nodes represent construction standard clauses and edges represent logical relationships or conflicting constraints between clauses. During the progress forecasting process, a graph embedding algorithm is used to verify the consistency of the process sequence with the knowledge graph. The graph embedding algorithm uses a representation learning method based on a hierarchical graph neural network. Specifically, it involves extracting subgraph structural features from the current process sequence, hierarchically aggregating node neighbor information in the knowledge graph, and generating a global consistency vector through graph pooling operations. In this way, it can ensure that the predicted process sequence meets the requirements of the construction specification, avoid construction violations, and improve construction quality and safety.

[0087] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0088] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the construction progress of assembled pipelines based on deep learning, characterized in that: include: Collect multi-source engineering data during the construction of prefabricated pipelines, including structured construction parameters, unstructured construction logs, pipeline 3D layout models, and real-time monitoring sensor data; Preprocessing the multi-source engineering data to generate a standardized feature set, wherein the standardized feature set includes pipeline layout geometric features, construction parameter time series features, log text semantic features, and sensor monitoring dynamic features; Extracting feature vectors from each data source based on a deep feature extraction model, wherein the deep feature extraction model includes a three-dimensional convolutional network, a time series encoder, and a text embedding network; Building a construction progress knowledge base, which stores historical construction case data, including pipeline construction process sequences, logical relationships between processes, and corresponding historical progress records; Mapping the feature vectors to a unified semantic space through a multi-source feature fusion method to generate a global fusion feature; Based on the global fusion features and the historical data in the construction progress knowledge base, a time series prediction algorithm is used to generate a process sequence and progress prediction result of the current pipeline construction.

2. The method for predicting the construction progress of an assembled pipeline according to claim 1, wherein: The method of using a time series prediction algorithm to generate a process sequence and progress prediction result for the current pipeline construction includes: generating an initial candidate process set according to the global fusion feature and the correlation between the processes in the knowledge base; Building a construction process network based on the logical relationship between the processes, wherein nodes in the network represent processes and edges represent logical dependencies or resource constraints between processes; The time series data is segmented by a sliding window mechanism, combined with the construction process network, and the initial candidate process set is optimized using a recursive neural network to output the final process sequence and progress prediction value.

3. The method for predicting the construction progress of an assembled pipeline according to claim 1, wherein: The preprocessing of the multi-source engineering data includes: A gridding algorithm is used to generate spatial topological features of the pipeline three-dimensional layout model, a one-dimensional convolutional network is used to extract local fluctuation features of the construction parameter time series features, and a bidirectional gated recurrent unit is used to extract context-related features of the log text semantic features.

4. The method for predicting the construction progress of an assembled pipeline according to claim 3, wherein: Mapping the feature vector to a unified semantic space by a multi-source feature fusion method includes: Calculate the weight distribution of features from different data sources through a multi-head attention mechanism; The weighted features are input into the residual connection network to fuse multi-scale features and eliminate redundant information.

5. The method for predicting the construction progress of an assembled pipeline according to claim 1, wherein: The time series prediction algorithm also includes: A progress optimization model is constructed based on a generative adversarial network. The model dynamically adjusts the rationality of the prediction sequence through a discriminator. The discriminator input includes process timing continuity, resource occupancy rate and construction specification compliance.

6. The method for predicting the construction progress of an assembled pipeline according to claim 5, characterized in that: The recursive neural network is an improved long short-term memory network, comprising: Encoding the construction process network into a memory unit state, wherein the state includes timestamps and dependency relationships of process nodes; The gating mechanism is used to control the forgetting and updating of temporal information, and skip connections are introduced to alleviate the gradient vanishing problem. The network weights are adjusted by backpropagating the prediction error, which is calculated as the root mean square deviation of the actual progress from the predicted value.

7. The method for predicting the construction progress of an assembled pipeline according to claim 1, wherein: The method further includes: using wavelet transform analysis on the real-time monitoring sensor data to extract abnormal signal characteristics during the construction process; when an abnormal signal is detected, triggering real-time correction of the process sequence, and updating the progress prediction result according to the corrected sequence.

8. The method for predicting the construction progress of an assembled pipeline according to claim 7, characterized in that: The real-time correction adopts the grey system theory, and the specific method includes: Define the grey correlation function of process adjustment threshold, resource reallocation rules and construction interruption tolerance; Correction candidate plans are generated through the grey prediction model, and the optimal adjustment strategy is selected based on the whitening weight function.

9. The method for predicting the construction progress of an assembled pipeline according to claim 1, wherein: The method further comprises: Constructing a construction specification knowledge graph, in which nodes represent construction standard clauses and edges represent logical relationships or conflict constraints between clauses; During the progress prediction process, the consistency between the process sequence and the knowledge graph is verified through a graph embedding algorithm.

10. The method for predicting the construction progress of an assembled pipeline according to claim 9, characterized in that: The graph embedding algorithm is a representation learning method based on a hierarchical graph neural network, which includes: extracting subgraph structural features from the current process sequence, hierarchically aggregating node neighbor information in the knowledge graph, and generating a global consistency vector through graph pooling operations.

Images