5G network construction quality monitoring method based on big data analysis

Abstract

The invention discloses a 5G network construction quality monitoring method based on big data analysis, and the method comprises the steps: carrying out the multi-modal data collection of a 5G base station process through a mobile terminal collection device, carrying out the offline image recognition based on a lightweight model, and generating a supervision data package; constructing a hierarchical supervision rule knowledge base including a specification standard layer, a historical experience layer and a manufacturer guide layer, and establishing a probability association model of construction parameter deviation and acceptance results in combination with an improved weighted time sequence association mining algorithm; a double-threshold risk assessment mechanism is adopted, when a rule threshold and a probability risk threshold are triggered at the same time, a risk level is assessed through a multi-dimensional risk scoring model, and a graded early warning report is generated; a structured rectification notice is generated for the quality problem, a rectification process time sequence verification chain is constructed based on a timestamp and a digital signature, and closed-loop management is realized through comparison between design parameters and an actual state; and establishing a decision feedback optimization ring, and continuously optimizing probability correlation model parameters and a supervision rule knowledge base.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method for monitoring the quality of 5G network construction based on big data analysis. Background Technology

[0002] With the large-scale deployment of 5G network construction and the surge in the number of base stations, traditional supervision models are facing severe challenges. Currently, 5G base station construction supervision mainly relies on manual inspections and paper records, which suffers from systemic defects such as inconsistent data collection standards, delayed discovery of quality problems, and difficulty in tracing the rectification process.

[0003] In existing technologies, some engineering management software only achieves simple electronic recording and lacks the analytical capabilities to deeply integrate with supervision business; while most mainstream quality monitoring systems on the market are developed by equipment manufacturers, focusing on monitoring equipment performance parameters and failing to meet the workflow and decision-making needs of supervision units. Especially in field construction environments, where network conditions are unstable, existing systems are difficult to support offline working modes.

[0004] Supervision decisions heavily rely on personal experience, new employee training is lengthy, and historical project data has not been effectively transformed into structured knowledge, leading to recurring quality problems. While some research has attempted to apply big data analytics to engineering quality monitoring, it largely remains theoretical, failing to address practical challenges such as high algorithm complexity, high implementation costs, and mismatch with actual supervisory capabilities. The industry urgently needs a quality monitoring method for 5G base station construction that can fully utilize the value of historical data, align with the technical capabilities of supervision units, and adapt to field working environments. Summary of the Invention

[0005] The purpose of this invention is to provide a method for monitoring the quality of 5G network construction based on big data analysis.

[0006] The problems that this invention aims to solve are: non-standard data collection, delayed discovery of quality problems, difficulty in tracing the rectification process, and excessive reliance on personal experience in the construction supervision of 5G base stations.

[0007] A method for monitoring the quality of 5G network construction based on big data analysis, the technical solution of which is as follows: Multimodal data is collected from 5G base station operations using a data acquisition device deployed on a mobile terminal. The data acquisition device performs offline image recognition on the collected images based on a lightweight model to generate a supervision data package. Measurement parameters are acquired and structured supervision records are formed based on an auxiliary module integrated into the mobile terminal. A hierarchical supervision rule knowledge base is constructed, and the three-layer rules are semantically associated based on the supervision knowledge base of the rule engine. Based on the improved weighted time-series association mining algorithm, combined with the historical project data accumulated by the supervision unit, a probabilistic association model between construction parameter deviations and acceptance results is constructed, and a time decay factor is introduced to calculate the probability of the impact of each construction parameter deviation on the final acceptance result. A dual-threshold risk assessment mechanism is adopted. When the collected construction data simultaneously triggers the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed based on the risk scoring model with weighted coefficients. A supervision early warning report is generated and pushed to the mobile terminal of the supervision engineer through a hierarchical push strategy to provide data support for supervision decision-making. For confirmed quality issues, a structured rectification notice is generated. The entire rectification process is recorded based on a chain of evidence with timestamps and digital signatures. The construction party is required to upload rectification evidence according to the time nodes, and a time-series verification chain for the rectification process is constructed. The design parameters are compared with the actual installation status for on-site verification, forming a traceable closed-loop management.

[0008] Furthermore, the process of collecting multimodal data from the 5G base station operation, generating a supervision data package, acquiring measurement parameters, and forming a structured supervision record includes: The acquisition device guides the supervisor to acquire images according to the shooting angle and range through a visual guide line on the screen. It then aligns the image data, location information, time series and process type in time and space to form the original acquisition dataset. The offline image recognition performs process integrity judgment, which identifies whether the specified shooting angle and area have been completed by comparing with a preset process image template; the recognition result and the original data are bound to the device fingerprint through digital signature to generate a structured supervision data package with tamper-proof characteristics. The acquired measurement parameters are associated with and stored with the corresponding identification results and environmental parameters. They are organized into structured supervision records according to the field structure required by the supervision specifications. The structured supervision records are uploaded to the supervision cloud platform through a differential synchronization mechanism when network conditions permit, to ensure data continuity and integrity.

[0009] Furthermore, the construction of the hierarchical supervision rule knowledge base includes: The hierarchical supervision rule knowledge base includes a standard layer, a historical experience layer, and a manufacturer's guide layer. The standard layer integrates mandatory regulatory clauses in the communications industry, extracts parameter thresholds and judgment conditions, and constructs a structured rule set. The historical experience layer is based on the supervision unit's historical project database, extracts high-frequency quality problem patterns and their handling solutions, and forms an experience rule set. The manufacturer's guide layer includes the installation technical requirements of mainstream equipment manufacturers, which are standardized and converted into equipment-specific rule sets. The rule engine realizes the association of the three layers of rules. The rule engine adopts a condition-action paradigm, and activates the associated rule when the input data meets the rule's preconditions.

[0010] Furthermore, the probabilistic correlation model between construction parameter deviations and acceptance results is constructed, and a time decay factor is introduced to calculate the probability of the impact of each construction parameter deviation on the final acceptance result, including: A construction quality probabilistic prediction model is constructed based on an improved weighted temporal association mining algorithm. This improved algorithm, building upon traditional association rule mining, introduces a construction environment correction coefficient and a process dependency weight. The construction environment correction coefficient adjusts the sensitivity of parameter deviations based on regional climate characteristics and construction season. The process dependency weight characterizes the influence of the quality of preceding processes on the acceptance results of subsequent processes. The algorithm's processing flow includes four stages: historical project data preprocessing, temporal feature extraction, association rule generation, and rule weight optimization. The preprocessing stage performs missing value imputation and outlier filtering on historical data. The temporal feature extraction stage identifies the duration and trend of parameter deviations. The association rule generation stage uses a support-confidence dual-threshold mechanism to screen effective rules. The rule weight optimization stage adjusts weights based on the historical performance of rule prediction accuracy. A time decay mechanism is introduced into the probabilistic association model. The time decay mechanism is designed based on the exponential decay function, which gives recent project data a higher influence. The model calculation process includes: standardizing the deviation of input construction parameters and mapping it to the historical data distribution space; matching the corresponding association rule set according to the parameter type; applying time decay weights to adjust the contribution of historical rules; aggregating the prediction results of multiple rules through a weighted voting mechanism; and outputting the probability value of each construction parameter deviation leading to acceptance failure.

[0011] Furthermore, when the collected construction data simultaneously triggers both the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed based on a risk scoring model with weighted coefficients, including: A dual-threshold collaborative risk assessment mechanism is implemented, which includes two independent assessment dimensions: a rule-triggered threshold and a probabilistic risk threshold. The rule-triggered threshold is determined by the structured rule set in the hierarchical supervision rule knowledge base. The probabilistic risk threshold is set by the acceptance failure probability output by the probabilistic correlation model, and the risk threshold is determined by the percentile method. When the collected construction data exceeds both thresholds simultaneously, the in-depth risk assessment process is activated; otherwise, only the data is recorded without triggering an early warning. A multi-dimensional risk scoring model is constructed, which integrates four assessment dimensions: parameter deviation degree, failure frequency of similar historical cases, process criticality coefficient, and rectification difficulty coefficient. The parameter deviation degree represents the deviation ratio between the measured value and the standard value. The failure frequency of similar historical cases is calculated based on historical experience layer rules to determine the historical probability of acceptance failure caused by similar problems. The process criticality coefficient reflects the impact weight of the process on the overall project quality and is pre-calibrated by industry experts. The rectification difficulty coefficient is determined comprehensively based on the expected rectification resource input and the impact on the construction period. The scores of each dimension are weighted and aggregated through a weight allocation mechanism to generate a comprehensive risk score of 0-100.

[0012] Furthermore, a supervision early warning report is generated and pushed to the mobile terminal of the supervision engineer through a tiered push strategy, including: A three-tiered risk classification system is implemented based on a comprehensive risk score. The three-tiered risk classification includes low risk (score below threshold A), medium risk (score between threshold A and threshold B), and high risk (score above threshold B). Differentiated supervision and early warning reports are generated for different risk levels. For low-risk issues, the notification is sent only to the on-site supervising engineer; for medium-risk issues, the notification is sent to both the on-site supervising engineer and the project director; for high-risk issues, the notification is sent to the on-site supervising engineer, the project director, and the construction unit representative. All notification records and confirmation status are synchronized to establish an early warning response tracking mechanism. If no confirmation is received within the specified time, the notification level is upgraded to ensure that risk issues are handled in a timely manner.

[0013] Furthermore, a structured rectification notice is generated for confirmed quality issues, and a time-series verification chain for the rectification process is constructed, including: The structured rectification notice includes six structured fields: problem description, standard basis, rectification standard, responsible unit, rectification deadline, and acceptance points. Each node in the rectification process generates a digital evidence package, which contains three sets of time-series data: pre-rectification status images, rectification process records, and post-rectification result verification. Each set of data is accompanied by GPS location information, device fingerprint, timestamp, and digital signature, and a unique data fingerprint is generated through a hash algorithm. Adjacent node data fingerprints are linked through a chain structure to form an irreversible time-series verification chain. When the construction party uploads evidence through a mobile terminal, the consistency between the location information and the geographical scope of the project is verified to prevent false uploads from different locations.

[0014] The beneficial effects of this invention are: the organic integration of multimodal data acquisition, hierarchical rule knowledge base, dual-threshold risk assessment, closed-loop rectification management and decision feedback optimization, fundamentally changes the traditional supervision work mode and realizes the qualitative change of engineering quality monitoring from passive response to proactive prevention.

[0015] The probabilistic correlation model built on the supervision unit's own historical data avoids the problem of general algorithms being out of touch with supervision practice, making the risk assessment results more in line with the actual project; the dual-threshold collaborative assessment mechanism effectively balances the sensitivity and accuracy of early warning, greatly reduces the false alarm rate, and enables the precise allocation of supervision resources.

[0016] This not only improved the quality control level of individual processes, but also restructured the whole life cycle management paradigm of 5G base station construction quality monitoring, enabling supervision work to move from qualitative judgment to quantitative decision-making, from experience-based reliance to data support, and from fragmented records to systematic knowledge accumulation. Attached Figure Description

[0017] Figure 1 This is a flowchart of a 5G network construction quality monitoring method based on big data analysis. Detailed Implementation

[0018] The present invention will be further described clearly and completely below, but the scope of protection of the present invention is not limited thereto.

[0019] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0020] Example 1 A method for monitoring the quality of 5G network construction based on big data analysis, the technical solution of which is as follows: Multimodal data is collected from 5G base station operations using a data acquisition device deployed on a mobile terminal. The data acquisition device performs offline image recognition on the collected images based on a lightweight model to generate a supervision data package. Measurement parameters are acquired and structured supervision records are formed based on an auxiliary module integrated into the mobile terminal. A hierarchical supervision rule knowledge base is constructed, and the three-layer rules are semantically associated based on the supervision knowledge base of the rule engine. Based on the improved weighted time-series association mining algorithm, combined with the historical project data accumulated by the supervision unit, a probabilistic association model between construction parameter deviations and acceptance results is constructed, and a time decay factor is introduced to calculate the probability of the impact of each construction parameter deviation on the final acceptance result. A dual-threshold risk assessment mechanism is adopted. When the collected construction data simultaneously triggers the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed based on the risk scoring model with weighted coefficients. A supervision early warning report is generated and pushed to the mobile terminal of the supervision engineer through a hierarchical push strategy to provide data support for supervision decision-making. For confirmed quality issues, a structured rectification notice is generated. The entire rectification process is recorded based on a chain of evidence with timestamps and digital signatures. The construction party is required to upload rectification evidence according to the time nodes, and a time-series verification chain for the rectification process is constructed. The design parameters are compared with the actual installation status for on-site verification, forming a traceable closed-loop management.

[0021] refer to Figure 1 The diagram shown is a flowchart of a 5G network construction quality monitoring method based on big data analysis.

[0022] Furthermore, the process of collecting multimodal data from the 5G base station operation, generating a supervision data package, acquiring measurement parameters, and forming a structured supervision record includes: The acquisition device displays a semi-transparent visual guide line on the mobile terminal screen, adjusting the shape and position of the guide line according to the current process type. For antenna installation, the guide line appears as a fan-shaped area, indicating the direction of the antenna's main lobe and the coverage angle. For cabinet installation, the guide line appears as a rectangular frame, indicating the four corner positioning points. When the deviation is less than 15 degrees and the coverage area reaches 90% of the preset range, a green prompt is given; otherwise, a red prompt is displayed to guide the supervisor to adjust the shooting position. During the construction of the original data collection dataset, GPS coordinates with an accuracy of no less than 5 meters, equipment timestamps, and process type identifiers are obtained. By establishing a unified spatiotemporal coordinate system, the above information is bound with the image data metadata to form a four-tuple data structure {image frame, location coordinates, timestamp, process type}, ensuring that each collection point has complete spatiotemporal context information.

[0023] Offline image recognition uses the lightweight MobileNetV2 model, deployed on mobile terminals, requiring no network connection. The preset process image template library contains multi-angle reference images of each process under standard conditions, with each process containing at least four viewpoint templates. Integrity judgment is achieved by calculating the structural similarity index (SSIM) between the acquired image and each template. When the SSIM values ​​of all specified angles are greater than 0.75 and the coverage area ratio meets the requirements, the process integrity is deemed to meet the requirements. The recognition result includes the process status, confidence score, and parameter measurement values. A data digest is generated by hashing the original data using the SHA-256 hash algorithm and combined with the device's unique identifier for RSA-2048 digital signature, forming an immutable data packet structure.

[0024] Measurement parameters include engineering parameters such as antenna tilt angle, installation height, and equipment spacing, which are obtained through mobile terminal sensors or external Bluetooth measurement devices; environmental parameters include temperature, humidity, wind speed, and light intensity, which are obtained through mobile terminal built-in sensors or meteorological APIs; all parameters are organized according to the field structure required for communication base station engineering construction supervision, and a structured supervision record is constructed using JSON format, including basic project information, process details, measurement data, environmental conditions, image index, and signature verification fields; The differential synchronization mechanism maintains a local data status table, recording the synchronization status flag for each record: 0 - not synchronized, 1 - synchronizing, 2 - synchronized. When a network signal strength greater than -90dBm and a network latency less than 500ms are detected, the synchronization process is triggered. After each synchronization is completed, the local status table is updated, and a synchronization log is generated to record the transmission time, data volume, and verification results. When the network is interrupted, synchronization is paused but local data collection continues until the network is restored, ensuring data integrity and continuity.

[0025] Furthermore, the construction of the hierarchical supervision rule knowledge base includes: The hierarchical supervision rule knowledge base includes a standard layer, a historical experience layer, and a manufacturer's guide layer. The standard layer mainly includes mandatory industry standards. When converting textual clauses into structured rules, a combination of keyword extraction and semantic parsing is used. For example, for the rule that the antenna installation tilt angle error must not exceed ±0.5°, the parameter names antenna tilt angle, threshold 0.5, unit degree, and comparison operator ≤ are extracted to construct the rule structure {parameter, threshold, comparison operator, consequence}. All rules are categorized and indexed according to process type, and rule priority identifiers are set, with mandatory standard rules set to the highest priority (level 1).

[0026] The historical experience layer data comes from the supervision unit's database of 5G base station construction projects over the past five years; it identifies high-frequency problem patterns such as insufficient bolt tightening torque in rainy weather; each pattern is associated with multiple handling solutions, sorted by historical success rate, forming rule pairs: IF(condition set){problem phenomenon, environmental factors, deviation degree}THEN(action set){handling measures, responsible party, time limit requirements}.

[0027] The manufacturer's guide layer integrates the technical manuals of mainstream equipment manufacturers and achieves standardized conversion by establishing a mapping table between equipment models and technical parameters. For different manufacturers' different descriptions of the same parameters, a synonym mapping relationship library is built to uniformly map downtilt angle, pitch angle, and tilt angle to standard antenna tilt angle terminology. Equipment-specific rules are set with applicable scope identifiers, including equipment model prefixes, frequency band ranges, and site type restrictions to avoid rule misuse.

[0028] The three-tiered rules are stored in the engine as a decision table. When a standard rule is triggered, the engine queries the associated rules with the same parameters in the historical experience layer and retrieves the technical requirements of the corresponding equipment model in the manufacturer's guide layer. The condition-action paradigm is specifically represented as a when condition then action. The condition part supports logical combinations (AND / OR / NOT), and the action part includes risk scoring, warning level setting, and handling suggestions.

[0029] Furthermore, the probabilistic correlation model between construction parameter deviations and acceptance results is constructed, and a time decay factor is introduced to calculate the probability of the impact of each construction parameter deviation on the final acceptance result, including: A construction quality probabilistic prediction model is constructed based on an improved weighted temporal correlation mining algorithm. The improved weighted temporal correlation mining algorithm is based on the FP-Growth algorithm and introduces construction environment correction coefficients and process dependency weights for the characteristics of the construction quality domain. The construction environment correction coefficient is adjusted based on the regional climate characteristics and the sensitivity of parameter deviations during the construction season. In rainy areas (annual precipitation > 800 mm) during the rainy season, the sensitivity of parameter deviations for bolt tightening processes increases, and α is set to 1.2. In dry areas, α is set to 0.8. The coefficient is determined by correlation analysis between historical project acceptance results and the construction environment. The correlation coefficient is used for verification, and a significant correlation is considered when the absolute value of the correlation coefficient is greater than 0.6. The process dependency weight β represents the degree of influence of the quality of the preceding process i on the acceptance result of the subsequent process j, and is calculated through the process network diagram. =0.7×(partial derivative of the acceptance rate of process i with the acceptance rate of process j)+0.3×(expert evaluation influence coefficient of process), where the expert evaluation coefficient is the average of scores given independently by more than 10 senior supervision engineers to ensure that the weighting is both data-driven and based on professional experience.

[0030] The algorithm processing flow includes four stages: historical project data preprocessing, temporal feature extraction, association rule generation, and rule weight optimization. The preprocessing stage performs missing value imputation and outlier filtering on historical data. The temporal feature extraction stage identifies the duration and trend of parameter deviation. The association rule generation stage uses a support-confidence dual threshold mechanism to screen effective rules. The rule weight optimization stage adjusts the weights based on the historical performance of the rule prediction accuracy. In the historical project data preprocessing stage, missing values ​​are filled using the K-nearest neighbor imputation method (K=5) by using project data under similar processes and environmental conditions; outlier filtering adopts the improved 3σ principle, and sets differentiated thresholds considering parameter types: for angle parameters such as tilt angle, a deviation exceeding ±2° is considered an anomaly; for distance parameters, a relative deviation exceeding 15% is considered an anomaly; for discrete parameters, the chi-square test is used to identify distribution anomalies. In the time series feature extraction stage, a time series is constructed for each construction parameter. For the parameter t=1,2,...,n}, calculate two core features: the duration of deviation T_d and the slope of the trend k; T_d is defined as the continuous time length (in hours) during which the parameter exceeds the standard range; the slope of the trend k is calculated using moving window linear regression, with the window size set to 3 measurement points, k>0.05 indicates a deteriorating trend, and k<-0.05 indicates an improving trend; In the association rule generation phase, a dual threshold mechanism of support and confidence is adopted. For the 5G base station construction field, the minimum support min_sup=0.25 and the minimum confidence min_conf=0.75 are determined through historical data analysis. The generated rule form is: {parameter A bias > threshold a, duration > ... }⇒{Acceptance Failure}, the rule strength is calculated as: strength=confidence×(1+α×β), which reflects the combined impact of environmental and process dependence, where confidence is the confidence level; During the rule weight optimization phase, an initial weight is set for each rule r. After the new project is accepted, the weights are updated based on the accuracy of the rule predictions. δ is the prediction impact coefficient, where δ=0.1 indicates a correct prediction and δ=-0.15 indicates an incorrect prediction; the upper limit of the weight is set to 1.0 and the lower limit is set to 0.2 to avoid extreme values.

[0031] A time decay mechanism is introduced into the probabilistic association model. This time decay mechanism is designed based on the exponential decay function, giving recent project data a higher influence. The model calculation process includes: standardizing the deviation of the input construction parameters and mapping it to the historical data distribution space; matching the corresponding association rule set according to the parameter type; applying time decay weights to adjust the contribution of historical rules; aggregating the prediction results of multiple rules through a weighted voting mechanism; and outputting the probability value of each construction parameter deviation leading to acceptance failure. The time decay mechanism employs an exponential decay function. , where Δt is the interval between historical projects and the current time (unit: month), the decay coefficient λ is determined by grid search method, the value range is [0.01, 0.2], the step size is 0.01, based on the principle of maximizing cross-validation accuracy, λ=0.08 is the best in this field, which means that the influence of project data is halved in about 12 months; During model calculation, the deviation of input construction parameters is first standardized using Z-score: z = (x - μ) / σ, where μ and σ are the mean and standard deviation of the parameter in historical data, respectively. The standardized parameters are then mapped to the historical rule condition space to match all applicable rules. The contribution of each matching rule is calculated as follows: Where strength is the rule strength. Here, represents the time decay weight corresponding to rule r, and similarity is the similarity between the input data and the rule conditions, calculated using a Gaussian kernel function; the final failure probability P is calculated through weighted voting. ,in The result is the rule result (1 indicates failure, 0 indicates success); the probability output range is 0-1, rounded to four decimal places, and values ​​greater than 0.65 are considered high risk.

[0032] Furthermore, when the collected construction data simultaneously triggers both the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed based on a risk scoring model with weighted coefficients, including: A dual-threshold collaborative risk assessment mechanism is implemented, which includes two independent assessment dimensions: a rule trigger threshold and a probability risk threshold. The rule trigger threshold is determined by the structured rule set in the hierarchical supervision rule knowledge base. The probability risk threshold is set by the acceptance failure probability output by the probability association model, and the risk threshold is determined by the percentile method. In the aforementioned dual-threshold collaborative risk assessment mechanism, the rule trigger threshold is directly extracted from the structured rule set of the hierarchical supervision rule knowledge base. For numerical parameters (such as antenna tilt angle), the threshold is the maximum allowable deviation value specified by the standard layer (such as ±0.5°). For compound condition rules (such as "rainy weather + bolt torque < 80% of standard value"), all conditions must be met simultaneously to be considered as triggered. The probability risk threshold uses the 75th percentile of the probability of historical project acceptance failure as the critical point. This percentile is determined through ROC curve analysis of historical data. Based on the principle of maximizing the Youden index (sensitivity + specificity - 1), and verified by backtesting of 500+ historical projects, the 75th percentile performs best in this field, with a corresponding probability value of 0.62.

[0033] When the collected construction data simultaneously exceeds two thresholds, the in-depth risk assessment process is activated; otherwise, only the data is recorded without triggering an alert. The simultaneous exceedance of the two thresholds is defined as follows: within the same process, the same measurement point, and a time interval of no more than 15 minutes, the collected data simultaneously exceeds both the rule trigger threshold and the probability risk threshold. A 10-minute observation window is set. If the conditions are met continuously within 10 minutes after the first simultaneous exceedance of the thresholds, or if the cumulative time for meeting the conditions exceeds 70%, the in-depth risk assessment process is activated; otherwise, only the abnormal data points are recorded without triggering an alert.

[0034] A multi-dimensional risk scoring model is constructed, which integrates four evaluation dimensions: parameter deviation degree, failure frequency of similar historical cases, process criticality coefficient, and rectification difficulty coefficient. The degree of parameter deviation characterizes the proportion of deviation between the measured value and the standard value; in the multi-dimensional risk scoring model, the degree of parameter deviation... The calculation formula is: =min(100, |Measured value-Standard value| / Maximum allowable deviation×120), where the coefficient 120 ensures that the score reaches the upper limit when the standard is seriously exceeded, and the maximum allowable deviation is taken from the standard layer rules of the specification. The failure frequency of historical similar cases is calculated based on historical experience layer rules, representing the historical probability of acceptance failure caused by similar problems. By retrieving cases with similar conditions (parameter type, deviation range, environmental condition matching degree > 85%) from the historical experience layer rule base, the acceptance failure rate is calculated and smoothed out. =100×(n_f+1) / (n_t+2), where n_f is the number of failed cases, n_t is the total number of cases, and +1 and +2 are Laplace smoothing terms to avoid the zero probability problem.

[0035] The criticality coefficient of the process reflects the weight of the process's impact on the overall project quality and is pre-calibrated by industry experts. The criticality coefficient K_c is determined using the Delphi method: 15 communication engineering supervision experts with more than 10 years of experience are invited to independently score each process of the 5G base station on a scale of 1-5 (5 points represents the most critical). After three rounds of anonymous feedback, the scores converge and are mapped to a coefficient of 0.5-1.0. The calculation formula is K_c = 0.5 + 0.125 × average score. For example, the average score of the antenna calibration process is 4.6, so K_c = 1.075, and the upper limit of 1.0 is taken; the average score of the foundation installation process is 2.8, so K_c = 0.85. The rectification difficulty coefficient is determined based on the expected rectification resource input and the impact on the construction period. The rectification difficulty coefficient D_r is calculated by combining the comprehensive resource input R_s (1-5 points, 1=low input, 5=high input) and the impact on the construction period T_i (1-5 points, 1=no impact, 5=serious delay). The formula is D_r=100×(0.6×R_s+0.4×T_i) / 5. The weights of 0.6 and 0.4 are determined by the AHP analytic hierarchy process and are verified to be effective by the consistency test (CR=0.03<0.1).

[0036] The scores from each dimension are weighted and aggregated using a weighted allocation mechanism to generate a comprehensive risk score ranging from 0 to 100. =0.35、 =0.25、 =0.25、 =0.15, determined by the AHP method and verified by the accuracy of historical project predictions; the comprehensive risk score R is calculated using the formula: R= × + × + ×K_c×100+ ×D_r, where K_c×100 converts the coefficients to a 0-100 score scale; the scoring results are rounded to the nearest integer, 0-35 points for low risk, 36-70 points for medium risk, and 71-100 points for high risk; the weighting and scoring thresholds have been optimized through backtesting of 200+ historical projects to ensure that the correlation coefficient between the risk level classification and the actual acceptance results reaches 0.82 (p<0.01), where , , , These are the weighting coefficients corresponding to the scores of each dimension.

[0037] Furthermore, a supervision early warning report is generated and pushed to the mobile terminal of the supervision engineer through a tiered push strategy, including: A three-level risk classification is implemented based on a comprehensive risk score. The three-level risk classification includes low risk (score below threshold A), medium risk (score between threshold A and threshold B), and high risk (score above threshold B). In the three-level risk classification, threshold A is set at 35 points and threshold B is set at 70 points. These thresholds are determined through ROC curve analysis of historical project risk scores and actual acceptance results, selecting the cutoff point that maximizes the Youden index (sensitivity + specificity - 1). Validated by over 500 historical projects, in the field of communication base station construction, 98.5% of risk issues below 35 points will not lead to acceptance failure, while 87.3% of risk issues above 70 points will lead to acceptance failure. The intermediate range is an uncertain area.

[0038] Differentiated supervision and early warning reports are generated for different risk levels. Low-risk reports include a brief description of the problem, standard requirements, explanations of minor deviations, and suggested measures, with a length of no more than 200 words. Medium-risk reports add comparisons with similar historical cases (showing the handling results of the three most similar historical cases), impact scope analysis, estimated impact on the project schedule, and priority handling suggestions, with a length of approximately 400 words. High-risk reports further include multi-angle on-site photos (no less than four), a list of emergency contacts, analysis of potential legal and economic consequences, emergency plan options, and a 24-hour expert support channel, with a length of 600-800 words. All reports include a unique identifier, generation timestamp, risk score details, and digital signature verification information.

[0039] For low-risk issues, the notification is sent only to the on-site supervising engineer; for medium-risk issues, it is sent to both the on-site supervising engineer and the project director; for high-risk issues, it is sent to the on-site supervising engineer, the project director, and the construction unit representative. All notification records are synchronized with the receipt confirmation status to establish an early warning response tracking mechanism. If no confirmation is received within the specified time, the notification level is upgraded to ensure that risk issues are handled in a timely manner. The early warning response tracking mechanism sets differentiated confirmation time limits: 24 hours for low-risk issues, 4 hours for medium-risk issues, and 60 minutes for high-risk issues. The confirmation method is for the recipient to click the "confirmed" button and fill in preliminary handling opinions (no less than 20 words). The confirmation status is checked every 15 minutes, and the level is upgraded if no confirmation is received within the time limit.

[0040] Furthermore, a structured rectification notice is generated for confirmed quality issues, and a time-series verification chain for the rectification process is constructed, including: The structured rectification notice includes six structured fields: problem description, standard basis, rectification standard, responsible unit, rectification deadline, and acceptance points. It is defined using JSON format, and the filling specifications for each field are as follows: The problem description field includes the problem location coordinates (accuracy 0.1 meters), problem type classification code (referencing YD / T 5160 standard), problem details text (no more than 300 characters), and problem severity rating; the standard basis field cites the specific standard clause number and content summary, such as "YD5191-2018 Clause 5.2.3"; the rectification standard field quantifies the specific indicators for acceptance, including the allowable deviation range and measurement method; the responsible unit field includes the full name of the unit, the name of the project leader, contact number, and an electronic signature area; the rectification deadline field adopts ISO... The 8601 time format allows setting start and end times and calculating the number of working days. The acceptance criteria field lists 3-5 key inspection items, each including the inspection method, acceptance criteria, and required tools. After the notification is generated, the supervising engineer confirms it via electronic signature through the APP, records the signature timestamp and equipment information, and stores it after AES-256 encryption.

[0041] Each stage of the rectification process generates a digital evidence package, which includes three sets of time-series data: pre-rectification status images, rectification process records, and post-rectification result verification. These include JPG format images of at least 2 megapixels each (at least 2 before rectification, at least 3 during rectification, and at least 2 after rectification, including both panoramic and detailed views), GPS location information (accuracy of at least 5 meters, WGS-84 coordinate system), device fingerprint (a 32-bit string generated by MD5 hashing the device's IMEI number, MAC address, and application installation ID), ISO 8601 format timestamp (accurate to the second, calibrated with the National Time Service Center network, with an error of no more than 1 second), and a digital signature based on the RSA-2048 algorithm. Each image is accompanied by metadata for shooting parameters, including focal length, aperture, exposure time, and ambient light intensity (unit: lux).

[0042] Each set of data is appended with GPS location information, device fingerprint, timestamp, and digital signature, and a unique data fingerprint is generated through a hash algorithm. Adjacent node data fingerprints are linked through a chain structure to form an irreversible time-series verification chain. When the construction party uploads evidence through a mobile terminal, the consistency between the location information and the geographical scope of the project is verified to prevent false uploads from different locations. The geographical scope of the project is predefined as a circular area with a radius of 50 meters from the center point of the base station (5G macro station) or a circular area with a radius of 20 meters (micro station). This radius value is determined through statistical analysis of the construction activity scope of historical projects, covering 98% of the normal construction activity area.

[0043] Example 2 This example selects a 5G macro base station construction project in the suburbs of a city, site number JS-2025-0458, construction date September 15, 2025, during the autumn rainy season, in a hilly area with an annual rainfall of approximately 950mm. Supervision personnel used a mobile terminal for quality monitoring; the terminal has a built-in lightweight MobileNetV2 model that runs offline without a network connection.

[0044] The supervisors collected data on the antenna installation process. A semi-transparent fan-shaped guide line was displayed on the mobile terminal screen, indicating the main lobe direction as true north and the coverage angle as 65°±5°. The supervisors then took photos from four key angles: a panoramic view of the mounting bracket from the front, a detailed view of the tilt angle scale from the left side, a detailed view of the fixing bolts from the right side, and a top view showing the azimuth angle.

[0045] The collected data is as follows: Image data: 4 photos, resolution 2340×1080; Location information: GPS coordinates 118.786523°E, 32.054218°N, accuracy 4.8 meters; Time series: 2023-09-15T14:23:45.678; Process type: Antenna installation - tilt calibration; Offline image recognition results: SSIM values ​​compared to the preset template: front view 0.83, left side 0.78, right side 0.81, top view 0.76, all greater than the 0.75 threshold; antenna tilt angle identified as 7.8°, standard requirement is 8.0°±0.5°, tilt angle deviation: 0.2° within the allowable range; coverage area ratio: 95.3%, greater than the 90% threshold; Generate a monitoring data packet, calculate the SHA-256 hash value as 3a7d4e...1c29, combine it with the device fingerprint IMEI: 864592047583621 to perform RSA-2048 signature, and form a tamper-proof data packet.

[0046] The hierarchical supervision rule knowledge base was queried: Standard layer: Antenna tilt angle allowable deviation ±0.5°, current deviation 0.2° is within limits; Historical experience layer: A search of the historical database revealed that insufficient antenna fixing bolt torque is a high-frequency problem during construction in September in rainy areas with annual precipitation >800mm. Rust was detected in some bolt photos, triggering the experience rule: IF (Rainy season construction AND Rainy area AND Bolt rust) THEN Risk increased; Manufacturer's guide layer: Matching the installation manual, the bolt torque is required to be 25±2 N·m, and a re-inspection is required every 24 hours. Construction environment correction factor α was calculated: Based on regional climate (rainy area) and construction season (September is the rainy season), α=1.3; based on historical correlation analysis, the Spearman coefficient is 0.68.

[0047] By accessing historical project data, a correlation was established between construction parameters and acceptance results: Historical data: Acceptance records of 500 5G base station projects in the past 3 years; Current parameters: Bolt preload test value 21 N·m (standard 25±2 N·m), deviation 16%; Environmental conditions: Relative humidity 85%, probability of rainfall 70%; Apply the time decay mechanism to calculate the time decay weight: The project weighting for the most recent 3 months is as follows: =1.0, project weight 6 months ago: =0.619; Based on the weighted time-series association mining algorithm, the historical rule "rainy season + bolt torque < standard value 85% + humidity > 80%" was matched, resulting in "acceptance unqualified". The support score was 0.31 and the confidence score was 0.83. After adjustment by the environmental correction coefficient, the rule strength was 0.83 × (1 + 1.3 × 0.85 × 0.7) = 1.37. Calculate the probability of final acceptance failure =(1.37×1+0.65×0) / 1.37+0.65=0.676; the probability risk threshold is 0.62 (75th percentile). 0.676 > 0.62, triggering the probability risk threshold.

[0048] Parameter deviation = |21-25| / 2×120 = 240, take the upper limit of 100; frequency of failure in similar historical cases = 100×(15+1) / (18+2) = 80, 15 out of 18 similar cases failed; Process criticality coefficient K_c = 0.5+0.125×4.6 = 1.075, take the upper limit of 1.0, convert to 100 points; Rectification difficulty coefficient D_r = 100×(0.6×3+0.4×4) / 5 = 68, resource input 3 points, construction period impact 4 points; Comprehensive risk score R = 0.35×100 + 0.25×80 + 0.25×100 + 0.15×68 = 35 + 20 + 25 + 10.2 = 90.2 points.

[0049] Risk rating: 90.2 > threshold B70, classified as high risk. A high-risk warning report is generated, containing 4 site photos, 3 similar historical cases, and a list of emergency contacts (Project Manager Wang, 1385678; Technical Expert Li, 1391234). This report is sent to site supervisor Zhang, project director Liu, and construction unit representative Wang. The report is sent at 14:28 and requires confirmation within 60 minutes.

[0050] At 14:45, the project director confirmed the issue and generated a structured rectification notice, which included: Issue description: Insufficient torque of antenna fixing bolts, measured value 21 N·m, standard 25±2 N·m; Standard basis: YD 5191-2018, Article 5.4.2, Huawei AAU Installation Manual V3.2, Chapter 8; Rectification standard: All bolt torque values ​​25±2 N·m, 100% re-inspection pass rate; Responsible unit: XX Communication Engineering Company, Person in charge: Mr. Zhao (136****4321); Rectification deadline: Before 18:00 on September 15, 2023; Acceptance points: ① Torque wrench calibration certificate ② Torque test records of all bolts ③ Proof of completion of rust prevention treatment.

[0051] The construction team uploaded the following rectification evidence: Before rectification: 4 photos showing bolt rust, GPS location 118.786531°E, 32.054203°N, distance from the project center 8.3 meters < 50 meters × 1.1, location verification passed; During rectification: 5 photos showing rust removal, application of rust inhibitor, and re-tightening process; After rectification: 3 photos, torque test record sheet, all bolts 25-26 N·m, GPS location 118.786497°E, 32.054225°N.

[0052] Calculate data fingerprint: Fingerprint before rectification =SHA256(file set)=a1b2c3...789; Fingerprint under rectification =SHA256(file set)=d4e5f6...012; Chained fingerprint =SHA256( || (3600) = g7h8i9...345 (time interval 3600 seconds); At 17:50, the supervisor conducted an on-site review, confirmed the rectification was qualified, and recorded the acceptance results. The rectification process formed a complete time-series verification chain, with all data encrypted and stored, forming a traceable closed-loop management system. After 30 days of follow-up, the base station passed acceptance on the first attempt, with no recurrence of quality issues.

[0053] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0054] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0055] This invention discloses a method for quality monitoring of 5G network construction based on big data analysis. It collects multimodal data on 5G base station construction processes using a mobile terminal acquisition device, performs offline image recognition based on a lightweight model, and generates a supervision data package. A hierarchical supervision rule knowledge base is constructed, comprising a standard layer, a historical experience layer, and a manufacturer's guideline layer. An improved weighted temporal correlation mining algorithm is used to establish a probabilistic correlation model between construction parameter deviations and acceptance results. A dual-threshold risk assessment mechanism is adopted; when both rule thresholds and probabilistic risk thresholds are triggered simultaneously, a multi-dimensional risk scoring model is used to assess the risk level and generate a graded early warning report. Structured rectification notices are generated for quality issues, and a temporal verification chain for the rectification process is constructed based on timestamps and digital signatures. Closed-loop management is achieved by comparing design parameters with actual conditions. A decision feedback optimization loop is established to continuously optimize the probabilistic correlation model parameters and the supervision rule knowledge base.

Claims

1. A method for monitoring the quality of 5G network construction based on big data analysis, characterized in that, include: S1: Multimodal data collection is performed on the 5G base station process using a collection device deployed on a mobile terminal. The collected images are then offline identified using a lightweight model to generate a supervision data package. Measurement parameters are obtained through the auxiliary module integrated into the mobile terminal to form a structured supervision record. S2: Construct a hierarchical supervision rule knowledge base including a standard layer, a historical experience layer, and a manufacturer's guide layer, and realize the semantic association of the three layers of rules based on the rule engine; S3: Based on the improved weighted time series association mining algorithm, a probabilistic association model between construction parameter deviations and acceptance results is constructed by combining historical project data. A time decay factor is introduced to calculate the probability of the impact of each construction parameter deviation on the acceptance results. S4: A dual-threshold risk assessment mechanism is adopted. When the collected construction data simultaneously triggers the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed through a risk scoring model with weighted coefficients. Generate a supervision early warning report and push the report to the mobile terminal of the corresponding supervisor based on the risk level using a tiered push strategy; S5: Generate a structured rectification notice for confirmed quality issues, construct an evidence chain for the entire rectification process based on timestamps and digital signatures, require the construction party to upload rectification evidence according to time nodes to form a time-series verification chain, and conduct on-site verification by comparing design parameters with actual installation status.

2. The 5G network construction quality monitoring method based on big data analysis as described in claim 1, characterized in that, S1 also includes: The screen provides visual guidance lines to prompt supervisors to collect image data according to preset shooting angles and ranges, and to align the collected image data, location information, time series and process types in time and space. Based on preset process image templates, the integrity of the process is judged to identify whether the specified shooting angle and area have been completed; the recognition results and raw data are bound to the equipment fingerprint through digital signature to generate a structured supervision data package; The measurement parameters are associated with the corresponding identification results and environmental parameters according to the field structure required by the supervision specifications to form a structured supervision record, and the structured supervision record is uploaded to the supervision cloud platform through a differential synchronization mechanism.

3. The 5G network construction quality monitoring method based on big data analysis as described in claim 1, characterized in that, S2 also includes: A hierarchical supervision rule knowledge base is constructed, comprising a standard layer, a historical experience layer, and a manufacturer's guide layer. The standard layer integrates telecommunications industry standard clauses, extracts parameter thresholds and judgment conditions to form a structured rule set. The historical experience layer is based on a historical project database, extracting quality problem patterns and their handling solutions to form an experience rule set. The manufacturer's guide layer includes equipment manufacturer installation technical requirements, which are standardized to form a device-specific rule set. A rule engine using a condition-action paradigm is used to associate the three layers of rules; when the input data meets the rule's preconditions, the corresponding associated rule is activated.

4. The 5G network construction quality monitoring method based on big data analysis as described in claim 1, characterized in that, S3 also includes: A construction quality probabilistic prediction model is constructed based on an improved weighted temporal association mining algorithm. This algorithm introduces a construction environment correction coefficient α and a process dependency weight β. The construction environment correction coefficient is determined based on regional climate characteristics and the construction season. The process dependency weight represents the degree of influence of the quality of the preceding process i on the acceptance result of the subsequent process j. This is calculated using a process network diagram. =0.7×(partial derivative of the acceptance rate of process i with the acceptance rate of process j)+0.3×(influence coefficient of process expert evaluation); The algorithm processing flow includes four stages: historical project data preprocessing, temporal feature extraction, association rule generation, and rule weight optimization. The preprocessing stage performs missing value imputation and outlier filtering on historical data. The temporal feature extraction stage identifies the duration and trend of parameter deviations. The association rule generation stage uses a support-confidence dual-threshold mechanism to filter effective rules, with rule strength = confidence × (1 + α × β), where confidence is the confidence level. The rule weight optimization stage adjusts the weights based on the historical performance of rule prediction accuracy, with initial weights... Where support represents the degree of support, and the new weights ,in The previous weight is δ, and the predicted impact coefficient is δ. A time decay mechanism based on an exponential decay function is introduced into the probabilistic correlation model. Where W(t) is the time decay weight, Δt is the interval between historical projects and the current time, and λ is the decay coefficient; the model calculation process includes: standardizing the deviation of the input construction parameters; matching the corresponding association rule set according to the parameter type; applying the time decay weight to adjust the contribution of historical rules, and the contribution of each matching rule is calculated as follows: Where strength is the rule strength. Let r be the time decay weight corresponding to rule r, and similarity be the similarity between the input data and the rule conditions. A weighted voting mechanism is used to aggregate the prediction results of multiple rules, and the probability value P of each construction parameter deviation leading to acceptance failure is output. ,in This is the result of the rule.

5. The method for monitoring the quality of 5G network construction based on big data analysis as described in claim 1, characterized in that, In step S4, when the collected construction data simultaneously triggers both the rule threshold in the hierarchical supervision rule knowledge base and the risk threshold of the probability association model, the risk level is assessed using a risk scoring model with weighted coefficients, including: A dual-threshold collaborative risk assessment mechanism is implemented, which includes a rule trigger threshold determined by the structured rule set in the hierarchical supervision rule knowledge base and a probability risk threshold set by the probability of acceptance failure output by the probability association model. The probability risk threshold is determined by the percentile method. When the collected construction data exceeds both thresholds at the same time, the in-depth risk assessment process is activated. A risk scoring model is constructed that integrates four assessment dimensions: parameter deviation degree, failure frequency of similar historical cases, process criticality coefficient, and rectification difficulty coefficient. The parameter deviation degree characterizes the proportion of deviation between the measured value and the standard value. =min(100, |Measured value - Standard value| / Maximum allowable deviation × 120); Failure frequency of historical similar cases Calculate the probability of acceptance failure due to similar issues based on historical experience data. =100×(n_f+1) / (n_t+2), where n_f is the number of failed cases, n_t is the total number of cases, and +1 and +2 are Laplace smoothing terms; the process criticality coefficient K_c reflects the weight of the process's impact on the overall project quality; the rectification difficulty coefficient D_r reflects the rectification resource requirement R_s and the impact on the construction period T_i, D_r=100×(0.6×R_s+0.4×T_i) / 5; the scores of each dimension are weighted and aggregated through a weight allocation mechanism to generate a comprehensive risk score R, R= × + × + ×K_c×100+ ×D_r, where , , , These are the weighting coefficients corresponding to the scores of each dimension.

6. The method for monitoring the construction quality of 5G networks based on big data analysis as described in claim 1, characterized in that, In step S4, a supervision early warning report is generated, and a tiered push strategy is used to push the report to the mobile terminals of the corresponding supervisors based on the risk level, including: A three-level risk classification system is implemented based on a comprehensive risk score. The three-level risk classification includes three levels: low risk, medium risk, and high risk. Differentiated supervision and early warning reports are generated for different risk levels. Issues are tiered and pushed out based on risk level: low-risk issues are pushed to the on-site supervising engineer; medium-risk issues are pushed to the on-site supervising engineer and project manager; high-risk issues are pushed to the on-site supervising engineer, project manager, and representative of the construction unit. Synchronize push notification records and confirmation status, establish an early warning response tracking mechanism, and upgrade the push notification level when no confirmation is received within a preset time limit.

7. The method for monitoring the quality of 5G network construction based on big data analysis as described in claim 1, characterized in that, S5 also includes: Generate a rectification notice containing six structured fields: problem description, standard basis, rectification criteria, responsible unit, rectification deadline, and acceptance points; Digital evidence packages are generated at each node of the rectification process. Each digital evidence package contains three sets of time-series data: pre-rectification status images, rectification process records, and post-rectification result verification. Each set of data is accompanied by GPS location information, device fingerprint, timestamp, and digital signature. A unique data fingerprint is generated through a hash algorithm. The data fingerprints of adjacent nodes are linked through a chain structure to form a time-series verification chain. When the construction party uploads evidence through a mobile terminal, the consistency between the location information and the preset engineering geographical range is verified.

Images