Communication engineering supervision auxiliary analysis method based on artificial intelligence

Abstract

The invention relates to the technical field of intelligent buildings, in particular to a communication engineering supervision auxiliary analysis method based on artificial intelligence, and the method comprises the following steps: S1, collecting multi-source communication engineering data of a construction site; s2, preprocessing the acquired multi-source communication engineering data; s3, performing intelligent analysis on the construction quality, the construction progress and the safety risk, and generating a visual analysis result; s4, predicting types and time of faults which may occur in the communication engineering; s5, when it is detected that the multi-source communication engineering data is abnormal or the fault prediction module predicts a fault, early warning information is generated; and S6, displaying the data analysis result, the fault prediction result and the early warning information in the form of charts, three-dimensional models or augmented reality through a visual display module. According to the method, the problems of data islands, one-sided analysis, early warning lag, low decision efficiency and the like in traditional engineering supervision are solved, comprehensive and intelligent upgrading of the supervision process is realized, and the requirement of intelligent supervision in a complex engineering scene is met.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent buildings, and in particular to an artificial intelligence-based communication engineering supervision auxiliary analysis method. Background Art

[0002] The current construction supervision field generally adopts the traditional model of manual inspections and paper records. This model relies on discrete sensor data collection and static analysis based on empirical rules. This model lacks the ability to integrate multi-source heterogeneous data (including high-dimensional image streams, time-series sensor signals, and unstructured construction logs) in real time, resulting in construction quality inspections lagging behind actual project progress. Traditional time series models (such as ARIMA) are poorly adaptable to nonlinear project progress forecasting and are unable to model the spatiotemporal coupling relationships between multiple processes. Furthermore, existing safety risk warning systems often use fixed threshold rules, making it difficult to dynamically adapt to changes in the construction site environment (such as sudden weather changes and equipment anomalies). Warning information is pushed one-way in text format and lacks visualization and linkage with three-dimensional geographic information and equipment status data. Furthermore, supervision decisions rely on manual experience and lack a data-driven closed-loop feedback mechanism, leading to inefficient exception response and irrational resource allocation. Although some existing technologies have adopted machine learning algorithms (such as LSTM and SVM) for single-dimensional analysis, they have defects such as weak model generalization ability (unable to migrate across engineering scenarios), poor multi-task collaboration (quality, progress, and safety analysis are isolated from each other), and lack of uncertainty quantification (prediction results lack credibility assessment), making it difficult to meet the needs of intelligent supervision in complex engineering scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide an artificial intelligence-based auxiliary analysis method for communication engineering supervision, which solves the pain points of traditional engineering supervision such as data silos, one-sided analysis, delayed warning, and inefficient decision-making, realizes a comprehensive intelligent upgrade of the supervision process, and meets the needs of intelligent supervision in complex engineering scenarios.

[0004] In order to achieve the above objectives, an artificial intelligence-based communication engineering supervision auxiliary analysis method is provided, which includes the following steps:

[0005] S1, collect multi-source communication engineering data of the construction site in real time through the data acquisition module;

[0006] S2. Preprocessing the collected multi-source communication engineering data through a data processing module, wherein the preprocessing includes cleaning, normalization, and feature extraction;

[0007] S3. Perform intelligent analysis of construction quality, construction progress, and safety risks on the pre-processed multi-source communication engineering data through the data analysis module, and generate visual analysis results;

[0008] S4, predicting the type and time of faults that may occur in the communication project through the analysis results of the data analysis module by the fault prediction module;

[0009] S5. When an anomaly in the multi-source communication engineering data is detected by the early warning notification module or a fault is predicted by the fault prediction module, early warning information is generated;

[0010] S6. Use the visualization module to display data analysis results, fault prediction results, and warning information in the form of charts, three-dimensional models, or augmented reality.

[0011] Furthermore, the data acquisition module includes:

[0012] An image recognition unit, including an intelligent camera array deployed at the construction site; acquiring image data inside the building or at the construction site through the intelligent camera array;

[0013] Distributed sensor arrays are deployed at communication base stations, optical cable laying paths, and equipment installation nodes; the distributed sensor arrays are used to collect environmental parameters at the construction site, including temperature, humidity, wind speed, and vibration, as well as performance parameters of the engineering structure, including stress data, displacement data, and deformation data;

[0014] Equipment operation monitoring unit, used to collect equipment operation status and performance parameters and obtain equipment operation data;

[0015] Inspection drones, which are equipped with lidar scanners and thermal imagers, can collect construction data from high altitudes and complex areas of the construction site;

[0016] The image data, equipment operation data, environmental parameters and construction data are aggregated into multi-source communication engineering data and uploaded, wherein the multi-source communication engineering data also includes construction tasks.

[0017] Furthermore, the data processing module includes:

[0018] Data cleaning unit, used to remove noise data and outliers from multi-source communication engineering data;

[0019] A data normalization unit, used to convert multi-source communication engineering data into a unified format;

[0020] The feature extraction unit is used to extract key features from multi-source communication engineering data.

[0021] Furthermore, the data analysis module includes: a construction quality analysis unit, a project progress analysis unit and a safety risk analysis unit; and step S3 includes the following steps:

[0022] S301, analyzing the construction image using a convolutional neural network model of a construction quality analysis unit to identify construction defects;

[0023] S302, predicting and evaluating the construction progress through the spatiotemporal Transformer hybrid prediction model of the project progress analysis unit;

[0024] S303. Identify and evaluate safety hazards at the construction site through the graph neural network model of the safety risk analysis unit.

[0025] Furthermore, the construction quality analysis unit uses a convolutional neural network model based on a deep residual network. This model captures local details and global structural features in construction images through a multi-scale feature extraction layer. It then dynamically weights key areas with a self-attention mechanism to locate and classify construction defects at the pixel level and output them. It then integrates a transfer learning strategy to adapt to quality inspection requirements in different engineering scenarios, while continuously optimizing model parameters through an online learning mechanism to improve inspection accuracy.

[0026] Project Progress Analysis Unit: This unit builds a hybrid prediction model based on spatiotemporal Transformers. This model uses a multi-head self-attention mechanism to capture the temporal dependencies between construction tasks. Combined with a graph convolutional network to model the logical constraints between tasks, it provides a multi-step rolling prediction of the project progress. It also introduces an uncertainty quantification module to generate confidence intervals for the project progress prediction results. By comparing the deviation between the actual progress and the planned progress, it analyzes and identifies potential delay risks on the critical path of the construction task.

[0027] Safety Risk Analysis Unit: Adopting a dynamic risk assessment framework based on knowledge graphs, it constructs a risk factor map for the construction site through entity recognition and relationship extraction, and uses graph neural networks to analyze the propagation path of risk events. It combines Monte Carlo simulation to quantify the probability of risk occurrence and the scope of impact. At the same time, it updates the risk factor map status through real-time data-driven updates to enable early identification and dynamic assessment of safety hazards.

[0028] Furthermore, the uncertainty quantification module includes:

[0029] Parameter probability distribution modeling unit: Based on the Bayesian neural network, the posterior distribution of the hybrid prediction model parameters is obtained, and then the confidence interval of the weight parameters is estimated through the Markov chain Monte Carlo sampling method to characterize the uncertainty range of the model parameters;

[0030] Confidence interval calculation unit: connected to the parameter probability distribution modeling unit, samples multiple groups of parameters from the posterior distribution through Monte Carlo Dropout and performs forward propagation, and calculates the upper and lower bounds of the predicted value containing 95% confidence intervals based on quantile regression to characterize the uncertainty range of the predicted value.

[0031] Furthermore, the uncertainty quantification module further includes:

[0032] Sensitivity analysis unit: connected to the confidence interval calculation unit, uses the Sobol index method to perform global sensitivity analysis on input variables, quantifies the contribution of input variables to the prediction results, and generates a sensitivity ranking report to identify key factors affecting the project progress; the input variables include resource input, weather conditions, and equipment status at the construction site;

[0033] Uncertainty propagation modeling unit: connects the confidence interval calculation unit and the sensitivity analysis unit, constructs an error propagation model based on stochastic differential equations, analyzes the cascade propagation path of upstream process prediction errors in the construction task network, and quantifies the risk probability of downstream task progress being affected.

[0034] Furthermore, the uncertainty quantification module further includes:

[0035] Dynamic threshold adjustment unit: connected to the confidence interval calculation unit, dynamically correcting the adaptive warning threshold according to the confidence interval width;

[0036] Among them: The initial warning threshold is obtained from historical engineering data statistics and is defined as T init =μ historical +k·σ historical , where μ is the mean of the historical confidence interval width, σ is the standard deviation, and k is the safety factor;

[0037] The dynamic adjustment rules are:

[0038] When the real-time confidence interval width W current >T init When , the manual review mechanism is triggered and threshold update suggestions are generated;

[0039] When W current ≤T init When T active =α·T init +(1-α)W current , where α is the smoothing factor, T active is the dynamic warning threshold that takes effect;

[0040] The final decision logic of the adaptive warning threshold is: if the deviation of the predicted value from the planned value exceeds T active , then trigger the graded warning signal.

[0041] Furthermore, the uncertainty quantification module further includes:

[0042] A real-time feedback calibration unit is connected to the parameter probability distribution modeling unit and the confidence interval calculation unit, and compares the actual progress data of the project with the predicted value of the hybrid prediction model through an online Bayesian update algorithm, corrects the posterior distribution of the model parameters of the hybrid prediction model in real time, and uses KL divergence to monitor the cumulative effect of the prediction deviation;

[0043] Model retraining trigger unit: connected to the real-time feedback calibration unit, when the KL divergence value exceeds the preset tolerance threshold, it automatically triggers the incremental learning process to update the model parameters to adapt to changes in construction site conditions.

[0044] Furthermore, the fault prediction module includes:

[0045] Time series prediction unit: predicts the equipment operating status based on long short-term memory network;

[0046] Fault classification unit: classifies possible fault types based on support vector machine;

[0047] Failure time prediction unit: predicts the time when a failure will occur based on the regression model.

[0048] Principles and advantages:

[0049] 1. This solution achieves full-factor coverage of multi-source heterogeneous data. Specifically, through the coordinated collection of distributed sensor arrays and drone inspection devices, it realizes all-weather collection of various data such as temperature, vibration, images, and environmental parameters in the construction scene. Compared with traditional methods, the data coverage rate is significantly improved, eliminating blind spots in manual inspections and avoiding the problem of incomplete supervision results.

[0050] 2. The data processing module uses a multimodal joint denoising algorithm through the data cleaning unit to effectively remove noise interference and outliers in multi-source communication engineering data, significantly improving data quality and ensuring the reliability of subsequent analysis; the data normalization unit uses multi-stage standardization processing to uniformly map the sensor time series signals, construction image pixel matrices and equipment log text feature vectors output by heterogeneous data sources to a standardized space, and constructs a multidimensional index matrix to achieve spatiotemporal alignment of multi-source data, providing highly consistent input for subsequent analysis; the feature extraction unit uses a parallelized deep feature extraction engine to extract local texture and global structural features from construction images, separate the time-frequency domain features representing the health status of equipment from sensor signals, and extract key entity-relationship triples from unstructured log texts, and finally output a multi-dimensional feature tensor that integrates spatiotemporal semantics, significantly enhancing the feature expression ability and model generalization performance of the downstream analysis module, providing a high-quality data foundation for intelligent decision-making in communication engineering supervision.

[0051] 3. This solution's construction quality analysis unit, by integrating the multi-level feature extraction capabilities of a deep residual network with the spatial semantic focus of a self-attention mechanism, can achieve pixel-level semantic segmentation of quality defects such as cracks, misalignments, and deformations in construction site images. This significantly improves recognition accuracy in complex lighting and occlusion scenarios, significantly reducing missed detection rates. Incorporating a transfer learning framework, it enables rapid model adaptation across project scenarios, reducing the need for annotated data for new scenarios. Furthermore, through an online incremental learning mechanism, it dynamically optimizes model parameters to ensure high detection accuracy even after long-term operation. The project progress analysis unit, on the other hand, constructs a hybrid prediction model based on a spatiotemporal Transformer and a graph convolutional network. This model captures the temporal dependencies between multiple processes through a multi-head self-attention mechanism and utilizes a graph convolutional network to model the impact of non-temporal factors such as resource allocation and environmental constraints on the critical path. This significantly reduces the error in total project duration prediction and improves the accuracy of critical path identification. Furthermore, an integrated uncertainty quantification module outputs confidence intervals for prediction results, supporting dynamic risk assessment and intelligent optimization of buffer time, significantly improving resource utilization and reducing the probability of project delays. The safety risk analysis unit adopts a dynamic risk assessment framework driven by knowledge graphs. It extracts multi-dimensional risk factors from construction logs and sensor data through entity recognition technology, and uses graph neural networks to model risk propagation paths. Combined with Monte Carlo simulation, it quantifies the probability of cascading impact of risk events, thereby achieving early warning of safety hazards and significantly reducing the false alarm rate. At the same time, it generates multi-dimensional reports including risk heat maps, disposal priority rankings, and resource scheduling recommendations, greatly improving the efficiency of high-risk event response and reducing the accident rate.

[0052] 4. The uncertainty quantification module constructs the probability distribution of model parameters through Bayesian neural networks, and combines the Monte Carlo Dropout method to generate confidence intervals for prediction results, significantly improving the interpretability and reliability of prediction results; at the same time, the Sobol index method is used to perform global sensitivity analysis on input variables, quantify the contribution of each variable to the prediction results and generate a sensitivity ranking report to help identify key influencing factors; through stochastic differential equation modeling of the propagation path of uncertainty in the construction task network, the cascade effect of upstream process prediction errors on downstream task progress is analyzed, supporting dynamic risk assessment and resource optimization; an integrated dynamic threshold adjustment mechanism is used to automatically correct the risk warning threshold according to the width of the confidence interval, significantly reducing the false alarm rate and improving the accuracy of the warning; the online Bayesian update function is used to compare the actual progress data with the predicted value and correct the probability distribution of the model parameters in real time, ensuring the long-term applicability and stability of the system under complex working conditions, and providing scientific and reliable uncertainty quantification support for communication engineering supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flowchart of an artificial intelligence-based communication engineering supervision auxiliary analysis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following is further described in detail through specific implementation methods:

[0055] Example

[0056] An artificial intelligence-based communication engineering supervision auxiliary analysis method, basically as follows Figure 1 As shown, the following steps are included:

[0057] S1. Real-time collection of multi-source communication engineering data at the construction site through a data acquisition module; the data acquisition module includes:

[0058] An image recognition unit, including an intelligent camera array deployed at the construction site; acquiring image data inside the building or at the construction site through the intelligent camera array;

[0059] Distributed sensor arrays are deployed at communication base stations, optical cable laying paths, and equipment installation nodes; the distributed sensor arrays are used to collect environmental parameters at the construction site, including temperature, humidity, wind speed, and vibration, as well as performance parameters of the engineering structure, including stress data, displacement data, and deformation data;

[0060] Equipment operation monitoring unit, used to collect equipment operation status and performance parameters and obtain equipment operation data;

[0061] Inspection drones, which are equipped with lidar scanners and thermal imagers, can collect construction data from high altitudes and complex areas of the construction site;

[0062] Image data, equipment operation data, environmental parameters and construction data are aggregated into multi-source communication engineering data and uploaded. The multi-source communication engineering data also includes construction tasks, construction logs, etc.

[0063] S2. Preprocessing the collected multi-source communication engineering data in the server through a data processing module, wherein the preprocessing includes cleaning, normalization, and feature extraction; the data processing module includes:

[0064] The data cleaning unit removes noise and outliers from multi-source communication engineering data. It eliminates sensor noise using an adaptive Kalman filter. A wavelet transform-median filter combined denoising algorithm is used to filter out Gaussian noise and pulse interference from multi-source communication engineering data. The isolation forest algorithm is used to identify and remove outliers caused by equipment failures or sudden environmental changes (including outliers exceeding the 3σ rule and invalid data that violates physical constraints). A sliding window mechanism is used to dynamically interpolate and repair missing data.

[0065] The data normalization unit converts multi-source communication engineering data into a unified format. Through a two-stage process of Z-score normalization and Min-Max scaling, it maps sensor time series signals (in units of dBm, °C, and MPa), construction image pixel matrices (RGB three-channel values ranging from 0 to 255), and device log text feature vectors from heterogeneous data sources into a normalized space of [-1, 1]. It then constructs a multidimensional index matrix containing timestamps, spatial coordinates, and data types to achieve spatiotemporal alignment of multi-source data.

[0066] The feature extraction unit is used to extract key features from multi-source communication engineering data. Based on a hybrid engine for deep feature extraction, it uses parallelized convolutional neural network (CNN) branches to extract local texture and global structural features from construction images. It then uses a wavelet packet decomposition algorithm to separate time-frequency domain features representing equipment health status from sensor signals. It also uses natural language processing (NLP) techniques to extract key entity-relationship triplets from unstructured log text. Ultimately, it outputs a multi-dimensional feature tensor that integrates spatiotemporal semantics for use by downstream analysis modules.

[0067] S3. In the server, a data analysis module is used to perform intelligent analysis of construction quality, construction progress, and safety risks on the pre-processed multi-source communication engineering data, and generate visual analysis results; the data analysis module includes: a construction quality analysis unit, a project progress analysis unit, and a safety risk analysis unit; step S3 includes the following steps:

[0068] S301. Analyze construction images using a convolutional neural network model of a construction quality analysis unit to identify construction defects. The construction quality analysis unit uses a convolutional neural network model based on a deep residual network to capture local details and global structural features in construction images through a multi-scale feature extraction layer. Combined with a self-attention mechanism to dynamically weight key areas, the unit locates and classifies construction defects at the pixel level and outputs the results. A transfer learning strategy is then integrated to adapt to quality inspection requirements in different engineering scenarios. Model parameters are continuously optimized through an online learning mechanism to improve inspection accuracy.

[0069] S302. Predict and evaluate the construction progress using the spatiotemporal Transformer hybrid prediction model of the project progress analysis unit. The project progress analysis unit constructs a spatiotemporal Transformer-based hybrid prediction model, captures the temporal dependencies between construction tasks through a multi-head self-attention mechanism, and combines the logical constraints between modeling tasks with a graph convolutional network to perform multi-step rolling predictions of the project progress. Furthermore, an uncertainty quantification module is introduced to generate confidence intervals for the project progress prediction results. By comparing the deviation between the actual progress and the planned progress, potential delay risks on the critical path of the construction task are analyzed and identified.

[0070] The uncertainty quantification module includes:

[0071] Parameter probability distribution modeling unit: Based on the Bayesian neural network, the posterior distribution of the hybrid prediction model parameters is obtained, and then the confidence interval of the weight parameters is estimated through the Markov chain Monte Carlo sampling method to characterize the uncertainty range of the model parameters;

[0072] Confidence interval calculation unit: connected to the parameter probability distribution modeling unit, samples multiple groups of parameters from the posterior distribution through Monte Carlo Dropout and performs forward propagation, and calculates the upper and lower bounds of the predicted value containing 95% confidence intervals based on quantile regression to characterize the uncertainty range of the predicted value.

[0073] Sensitivity analysis unit: connected to the confidence interval calculation unit, uses the Sobol index method to perform global sensitivity analysis on input variables, quantifies the contribution of input variables to the prediction results, and generates a sensitivity ranking report to identify key factors affecting the project progress; the input variables include resource input, weather conditions, and equipment status at the construction site;

[0074] Uncertainty propagation modeling unit: connects the confidence interval calculation unit and the sensitivity analysis unit, constructs an error propagation model based on stochastic differential equations, analyzes the cascade propagation path of upstream process prediction errors in the construction task network, and quantifies the risk probability of downstream task progress being affected.

[0075] Dynamic threshold adjustment unit: connected to the confidence interval calculation unit, dynamically correcting the adaptive warning threshold according to the confidence interval width;

[0076] Among them: The initial warning threshold is obtained from historical engineering data statistics and is defined as T init =μ historical +k·σ historical , where μ is the mean of the historical confidence interval width, σ is the standard deviation, and k is the safety factor;

[0077] The dynamic adjustment rules are:

[0078] When the real-time confidence interval width W current >T init When , the manual review mechanism is triggered and threshold update suggestions are generated;

[0079] When W current ≤T init When T active =α·T init +(1-α)W current , where α is the smoothing factor, T active is the dynamic warning threshold that takes effect;

[0080] The final decision logic of the adaptive warning threshold is: if the deviation of the predicted value from the planned value exceeds T active , then trigger the graded warning signal.

[0081] A real-time feedback calibration unit is connected to the parameter probability distribution modeling unit and the confidence interval calculation unit, and compares the actual progress data of the project with the predicted value of the hybrid prediction model through an online Bayesian update algorithm, corrects the posterior distribution of the model parameters of the hybrid prediction model in real time, and uses KL divergence to monitor the cumulative effect of the prediction deviation;

[0082] Model retraining trigger unit: connected to the real-time feedback calibration unit, when the KL divergence value exceeds the preset tolerance threshold, it automatically triggers the incremental learning process to update the model parameters to adapt to changes in construction site conditions.

[0083] S303. Identify and assess safety hazards at the construction site using the graph neural network model of the safety risk analysis unit. The safety risk analysis unit uses a dynamic risk assessment framework based on a knowledge graph to construct a risk factor map for the construction site through entity recognition and relationship extraction. It also uses a graph neural network to analyze the propagation path of risk events and combines Monte Carlo simulation to quantify the probability of risk occurrence and the scope of impact. It also uses real-time data to drive updates to the risk factor map, enabling early identification and dynamic assessment of safety hazards.

[0084] This uncertainty quantification module significantly improves the interpretability and reliability of prediction results through the technical chain of Bayesian inference-sensitivity analysis-error propagation modeling-dynamic threshold adjustment-online feedback calibration, supports dynamic risk assessment and resource optimization, provides scientific and reliable uncertainty quantification support for communication engineering supervision, and promotes the transformation and upgrading of supervision decisions from "experience-driven" to "data-driven".

[0085] S4. In the server, the fault prediction module analyzes the analysis results of the data analysis module to predict the type and time of the fault that may occur in the communication project; the fault prediction module includes:

[0086] Time Series Prediction Unit: This unit predicts device operating status based on a long short-term memory network. It builds a bidirectional long short-term memory network (BiLSTM-AM) enhanced with an attention mechanism. It processes time series data on device operating status (including current, temperature, and vibration frequency) through a sliding window mechanism, extracts multi-scale time domain features, and predicts device performance degradation trends within the next 72 hours, while also outputting a prediction curve with confidence intervals.

[0087] Fault Classification Unit: This solution classifies possible fault types based on a support vector machine. This solution, connected to the time series prediction unit, uses a multi-kernel support vector machine (MK-SVM) framework, integrating a weighted combination of Gaussian and polynomial kernels to classify fault feature vectors. These feature vectors are composed of time series prediction results, real-time operating parameters (ambient humidity, load factor), and text descriptions from historical maintenance records, using word embedding encoding. This allows for accurate identification of nine typical communication engineering faults, including equipment overload, cable aging, and loose connectors, with a classification accuracy of ≥94%.

[0088] Fault time prediction unit: Predicts the time of fault occurrence based on the regression model. This solution specifically connects the fault classification unit with the time series prediction unit to build a LightGBM integrated model based on quantile regression. It uses the current health index of the equipment, the severity level of the fault type, and the availability of maintenance resources as input to predict the probability density distribution of the fault occurrence time and generate hierarchical time labels including the early warning time window (T1±Δt1) and the critical handling time window (T2±Δt2). Δt1 and Δt2 are dynamically adjusted according to the propagation characteristics of the fault type.

[0089] S5. In the server, when an abnormality of the multi-source communication engineering data is detected or a fault prediction module predicts a fault, a warning message is generated by the warning notification module; the warning notification module includes:

[0090] Dynamic threshold adjustment unit, used to adjust the warning threshold according to environmental conditions and construction stage;

[0091] Multi-channel notification unit, supporting sending warning information via SMS, email, mobile application push, etc.

[0092] The emergency plan recommendation unit is used to recommend corresponding processing plans based on the fault type.

[0093] S6. Display the data analysis results, fault prediction results, and warning information in the form of charts, three-dimensional models, or augmented reality through a visualization display module in the server. The visualization display module includes:

[0094] Data dashboards for real-time display of statistical analysis results of construction quality, progress, and safety risks;

[0095] A 3D model display unit, used to build a digital twin model of the construction site and mark abnormal areas;

[0096] The augmented reality (AR) display unit supports supervisors to view virtual annotation information through smart terminals.

[0097] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme will not be described in detail here. Those of ordinary skill in the art are aware of all common technical knowledge in the technical field to which the invention belongs before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Those of ordinary skill in the art can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for those of ordinary skill in the art to implement this application. It should be pointed out that for those skilled in the art, several variations and improvements can be made without departing from the structure of the present invention. These should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. An artificial intelligence-based communication engineering supervision auxiliary analysis method, characterized by: The following steps are involved: S1, collect multi-source communication engineering data of the construction site in real time through the data acquisition module; S2. Preprocessing the collected multi-source communication engineering data through a data processing module, wherein the preprocessing includes cleaning, normalization, and feature extraction; S3. Perform intelligent analysis of construction quality, construction progress, and safety risks on the pre-processed multi-source communication engineering data through the data analysis module, and generate visual analysis results; S4, predicting the type and time of faults that may occur in the communication project through the analysis results of the data analysis module by the fault prediction module; S5. When an anomaly in the multi-source communication engineering data is detected by the early warning notification module or a fault is predicted by the fault prediction module, early warning information is generated; S6. Use the visualization module to display data analysis results, fault prediction results, and warning information in the form of charts, three-dimensional models, or augmented reality.

2. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 1 is characterized in that: The data acquisition module includes: An image recognition unit, including an intelligent camera array deployed at the construction site; acquiring image data inside the building or at the construction site through the intelligent camera array; Distributed sensor arrays are deployed at communication base stations, optical cable laying paths, and equipment installation nodes; the distributed sensor arrays are used to collect environmental parameters at the construction site, including temperature, humidity, wind speed, and vibration, as well as performance parameters of the engineering structure, including stress data, displacement data, and deformation data; Equipment operation monitoring unit, used to collect equipment operation status and performance parameters and obtain equipment operation data; Inspection drones, which are equipped with lidar scanners and thermal imagers, can collect construction data from high altitudes and complex areas of the construction site; The image data, equipment operation data, environmental parameters and construction data are aggregated into multi-source communication engineering data and uploaded, wherein the multi-source communication engineering data also includes construction tasks.

3. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 2 is characterized in that: The data processing module includes: Data cleaning unit, used to remove noise data and outliers from multi-source communication engineering data; A data normalization unit, used to convert multi-source communication engineering data into a unified format; The feature extraction unit is used to extract key features from multi-source communication engineering data.

4. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 3 is characterized in that: The data analysis module includes: a construction quality analysis unit, a project progress analysis unit and a safety risk analysis unit; the step S3 includes the following steps: S301, analyzing the construction image using a convolutional neural network model of a construction quality analysis unit to identify construction defects; S302, predicting and evaluating the construction progress through the spatiotemporal Transformer hybrid prediction model of the project progress analysis unit; S303. Identify and evaluate safety hazards at the construction site through the graph neural network model of the safety risk analysis unit.

5. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 4 is characterized in that: The construction quality analysis unit uses a convolutional neural network model based on a deep residual network. It uses a multi-scale feature extraction layer to capture local details and global structural features in construction images. It then dynamically weights key areas with a self-attention mechanism to locate and classify construction defects at the pixel level and output them. It then integrates a transfer learning strategy to adapt to quality inspection requirements in different engineering scenarios, while continuously optimizing model parameters through online learning to improve inspection accuracy. Project Progress Analysis Unit: This unit builds a hybrid prediction model based on spatiotemporal Transformers. This model uses a multi-head self-attention mechanism to capture the temporal dependencies between construction tasks. Combined with a graph convolutional network to model the logical constraints between tasks, it provides a multi-step rolling prediction of the project progress. It also introduces an uncertainty quantification module to generate confidence intervals for the project progress prediction results. By comparing the deviation between the actual progress and the planned progress, it analyzes and identifies potential delay risks on the critical path of the construction task. Safety Risk Analysis Unit: Adopting a dynamic risk assessment framework based on knowledge graphs, it constructs a risk factor map for the construction site through entity recognition and relationship extraction, and uses graph neural networks to analyze the propagation path of risk events. It combines Monte Carlo simulation to quantify the probability of risk occurrence and the scope of impact. At the same time, it updates the risk factor map status through real-time data-driven updates to enable early identification and dynamic assessment of safety hazards.

6. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 5 is characterized in that: The uncertainty quantification module includes: Parameter probability distribution modeling unit: Based on the Bayesian neural network, the posterior distribution of the hybrid prediction model parameters is obtained, and then the confidence interval of the weight parameters is estimated through the Markov chain Monte Carlo sampling method to characterize the uncertainty range of the model parameters; Confidence interval calculation unit: connected to the parameter probability distribution modeling unit, samples multiple groups of parameters from the posterior distribution through Monte Carlo Dropout and performs forward propagation, and calculates the upper and lower bounds of the predicted value containing 95% confidence intervals based on quantile regression to characterize the uncertainty range of the predicted value.

7. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 6 is characterized in that: The uncertainty quantification module further includes: Sensitivity analysis unit: connected to the confidence interval calculation unit, uses the Sobol index method to perform global sensitivity analysis on input variables, quantifies the contribution of input variables to the prediction results, and generates a sensitivity ranking report to identify key factors affecting the project progress; the input variables include resource input, weather conditions, and equipment status at the construction site; Uncertainty propagation modeling unit: connects the confidence interval calculation unit and the sensitivity analysis unit, constructs an error propagation model based on stochastic differential equations, analyzes the cascade propagation path of upstream process prediction errors in the construction task network, and quantifies the risk probability of downstream task progress being affected.

8. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 7 is characterized in that: The uncertainty quantification module further includes: Dynamic threshold adjustment unit: connected to the confidence interval calculation unit, dynamically correcting the adaptive warning threshold according to the confidence interval width; Among them: The initial warning threshold is obtained from historical engineering data statistics and is defined as T init =μ historical +k·σ historical , where μ is the mean of the historical confidence interval width, σ is the standard deviation, and k is the safety factor; The dynamic adjustment rules are: When the real-time confidence interval width W current >T init When , the manual review mechanism is triggered and threshold update suggestions are generated; When W current ≤T init When T active =α·T init +(1-α)W current , where α is the smoothing factor, T active is the dynamic warning threshold that takes effect; The final decision logic of the adaptive warning threshold is: if the deviation of the predicted value from the planned value exceeds T active , then trigger the graded warning signal.

9. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 8 is characterized in that: The uncertainty quantification module further includes: A real-time feedback calibration unit is connected to the parameter probability distribution modeling unit and the confidence interval calculation unit, and compares the actual progress data of the project with the predicted value of the hybrid prediction model through an online Bayesian update algorithm, corrects the posterior distribution of the model parameters of the hybrid prediction model in real time, and uses KL divergence to monitor the cumulative effect of the prediction deviation; Model retraining trigger unit: connected to the real-time feedback calibration unit, when the KL divergence value exceeds the preset tolerance threshold, it automatically triggers the incremental learning process to update the model parameters to adapt to changes in construction site conditions.

10. The communication engineering supervision auxiliary analysis method based on artificial intelligence according to claim 6, characterized in that: The fault prediction module includes: Time series prediction unit: predicts the equipment operating status based on long short-term memory network; Fault classification unit: classifies possible fault types based on support vector machine; Failure time prediction unit: predicts the time when a failure will occur based on the regression model.

Images