A construction engineering construction noise data anomaly detection method, system, device and storage medium
By using a stacking ensemble model and feature engineering, abnormal patterns in construction noise data are identified, solving the problem of a lack of existing methods for detecting construction noise data. This enables efficient detection of noise data anomalies and diagnosis of equipment faults, reducing regulatory costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HOHAI UNIV
- Filing Date
- 2022-08-17
- Publication Date
- 2026-06-02
AI Technical Summary
Existing technologies lack methods for detecting anomalies in construction noise data applicable to the construction engineering field, which cannot effectively address the characteristics of high-dimensional, multi-modal, and multi-correlated data at construction sites. Furthermore, traditional methods cannot be combined with domain knowledge and contextual data mining, resulting in high regulatory costs.
A Stacking ensemble model, including weak classifiers and secondary classifiers, is adopted. Through feature engineering and 5-fold cross-validation training, abnormal patterns in construction noise data are identified. Combined with construction noise emission standards and noise time series data feature extraction methods, anomaly detection of noise data is achieved.
It has improved the accuracy and efficiency of noise data detection at construction sites, reduced regulatory costs, promoted the transformation of traditional manual inspections to data-driven remote supervision, and enabled online inspection and fault diagnosis of monitoring equipment.
Smart Images
Figure CN115392362B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method, system, device, and storage medium for detecting abnormal construction noise data in construction projects, belonging to the field of engineering management technology. Background Technology
[0002] Construction is one of the most important human survival activities. The construction phase of a construction project is the most active phase in terms of resource flow and project management throughout the project's entire life cycle. It is also a crucial phase for forming the final engineering entity and realizing the project's use value. However, there are still many problems in the management of construction sites, such as extensive management methods and reliance on manual inspection equipment.
[0003] To address these issues, governments and relevant enterprises have introduced numerous policies and standards. Among these, strengthening project management through information technology is a key approach. Driven by information technologies such as the Internet of Things and cloud computing, construction projects are exhibiting new characteristics of digitalization and intelligence. Construction sites have become more visible, controllable, and perceptible, giving rise to smart construction sites. The application of smart construction sites in construction projects has improved the on-site management level of construction companies, contributing to the implementation of the scientific development concept and sustainable development. Simultaneously, smart construction sites generate abundant engineering big data, which is a data-driven description of the physical activities of the project. Driven by this engineering big data, the supervision of the construction process has begun... As supervision shifts from offline to remote, researchers are exploring how to analyze sensor big data to replace traditional manual monitoring methods. Anomalies in construction noise time-series data are less frequent than normal, yet they hold greater research value. Detecting anomalies in sensor data or data from multiple related points allows for further investigation into the causes of these anomalies, enabling timely solutions before problems escalate. Taking noise sensors as an example, data-driven methods can detect abnormal noise data subsequences, diagnose monitoring equipment malfunctions, and enable online inspections of noise monitoring equipment at smart construction sites, thereby reducing the supervision costs for regulators in evaluating the quality of various construction projects.
[0004] While academia has developed various solutions to the problem of time series anomaly detection, there is still no anomaly detection method that can be directly applied to the construction engineering field, taking into account the knowledge of engineering management and the characteristics of construction engineering datasets. On the one hand, time series data collected at construction sites not only suffers from large scale, rapid generation, and low quality, but also exhibits new characteristics such as high dimensionality, multiple patterns, and multiple correlations, requiring detection methods with low time complexity to preserve the timeliness of the data. Existing anomaly detection methods are not well adapted to these characteristics. On the other hand, although construction data anomaly detection shares the characteristics of time series anomaly detection, its focus is not on improving the means and methods of anomaly detection, but rather on causality of anomalies. It requires combining domain knowledge and contextual data mining, considering application scenarios and time cost factors, with the aim of improving regulatory methods and reducing regulatory costs. Summary of the Invention
[0005] The purpose of this invention is to provide a method, system, device, and storage medium for detecting abnormal construction noise data in construction projects, thereby solving the problem of a lack of means and methods for abnormal detection in the prior art.
[0006] To achieve the above objectives, the present invention employs the following technical solution:
[0007] In a first aspect, the present invention provides a method for detecting abnormal construction noise data in construction projects, comprising:
[0008] Obtain the dataset to be detected;
[0009] The dataset to be detected is input into a pre-trained anomaly detection model to obtain a list of noise data anomalies, thus completing the anomaly detection of construction noise data in construction projects.
[0010] In conjunction with the first aspect, the anomaly detection model is further described as a Stacking ensemble model, comprising a weak classifier in the first layer and a secondary classifier in the second layer. The weak classifier includes SGD classifier, random forest classifier, extremely random tree classifier, Adaboost classifier, and GBDT classifier.
[0011] In the Stacking ensemble model, the output of the weak classifier is used as the input of the secondary classifier.
[0012] In conjunction with the first aspect, the anomaly detection model is further trained using the following method:
[0013] Obtain a noisy dataset, perform feature engineering on the noisy dataset to obtain a training set;
[0014] Noisy data from the training set is input into weak classifiers. Five weak classifiers are trained using the 5-fold cross-validation method. The output of the weak classifiers is then used to train the secondary classifiers, resulting in a trained anomaly detection model.
[0015] In conjunction with the first aspect, the feature engineering process for the noisy dataset further includes: feature cleaning and preprocessing followed by feature reconstruction;
[0016] Feature cleaning includes filling missing values and deduplicating feature values in noisy data. Missing value filling uses the front-to-back averaging method, and feature value deduplication is based on the one-to-one relationship between timestamps and record values.
[0017] Preprocessing involves dimensionality reduction and feature selection of the noisy dataset. Dimensionality reduction is achieved through principal component analysis, while feature selection removes irrelevant features and fixed constant feature attributes from the noisy data.
[0018] Feature reconstruction involves extracting statistical feature attributes to describe abnormal patterns in noisy data, and then combining these attributes with those obtained after dimensionality reduction to generate a reconstructed feature set.
[0019] In conjunction with the first aspect, the abnormal noise data patterns are further identified using construction project noise emission standards and noise time series data subsequence feature extraction methods.
[0020] In conjunction with the first aspect, the noise data anomaly patterns further include fixed value anomalies, constant period anomalies, excessively low value anomalies, extreme value anomalies, sudden anomalies, and trend anomalies. Among these, extreme value anomalies and sudden anomalies are noise data point anomaly patterns, while the remaining anomaly patterns are noise subsequence anomaly patterns.
[0021] Secondly, the present invention also provides a construction noise data anomaly detection system, comprising:
[0022] Test Data Acquisition Module: Used to acquire the dataset to be tested;
[0023] Anomaly detection module: Used to input the dataset to be detected into a pre-trained anomaly detection model to obtain a list of noise data anomalies and complete the anomaly detection of construction noise data in construction projects.
[0024] Thirdly, the present invention also provides a construction noise data anomaly detection device, including a processor and a storage medium;
[0025] The storage medium is used to store instructions;
[0026] The processor is configured to operate according to the instructions to perform the steps of the method according to any one of the first aspects.
[0027] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the second aspect, and the present invention also provides the steps of any of the methods described herein.
[0028] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0029] This invention provides a method, system, device, and storage medium for detecting anomalies in construction noise data. By using a pre-trained anomaly detection model to detect noise anomalies in the dataset, it helps improve the quality of sensor big data, providing high-quality data for all stakeholders in the project. Furthermore, it improves on-site supervision methods, promoting the transformation from traditional manual inspection to data-driven remote supervision. This enables the diagnosis of anomalies detected in construction projects and the control of system operation status. After detecting abnormal noise data, it can further diagnose faults in monitoring equipment, enabling online inspection of smart construction site noise monitoring equipment, thereby reducing the cost for regulators to supervise various construction projects. Attached Figure Description
[0030] Figure 1 This is a schematic diagram of a fixed value anomaly provided in an embodiment of the present invention;
[0031] Figure 2 This is a schematic diagram of a constant period anomaly provided in an embodiment of the present invention;
[0032] Figure 3 This is a schematic diagram of extreme value anomalies provided in an embodiment of the present invention;
[0033] Figure 4 This is a schematic diagram of a sudden anomaly provided in an embodiment of the present invention;
[0034] Figure 5 This is a schematic diagram of trend anomalies provided in an embodiment of the present invention;
[0035] Figure 6 This is a flowchart of the feature engineering process provided in an embodiment of the present invention;
[0036] Figure 7 This is a schematic diagram of the anomaly detection model provided in an embodiment of the present invention. Detailed Implementation
[0037] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to illustrate the technical solution of the present invention more clearly, and should not be used to limit the scope of protection of the present invention.
[0038] Example 1
[0039] This invention provides a method for detecting abnormal construction noise data in construction projects, comprising:
[0040] S1. Obtain the dataset to be detected.
[0041] Collect the dataset to be tested from real-world scenarios.
[0042] S2. Input the dataset to be detected into the pre-trained anomaly detection model to obtain a list of noise data anomalies and complete the anomaly detection of construction noise data in the construction project.
[0043] The anomaly detection model in this embodiment of the invention adopts an ensemble model. The basic idea of ensemble learning is to integrate multiple ordinary models with average performance to obtain an ensemble model with excellent performance. The core tasks of the ensemble model are twofold: first, how to obtain multiple ordinary models with average performance; and second, how to effectively integrate these models to obtain an ensemble model with excellent performance.
[0044] Stacking ensemble models are hierarchical model ensemble frameworks that integrate at least two or more classifiers, and can be considered as "model stacking". Stacking ensemble learning methods can effectively fuse multiple classifiers through a tree structure, thereby improving classification accuracy and precision. Typically, the first weak classifier in Stacking will select a model with high fit in order to learn fully from the training data.
[0045] like Figure 7 As shown, the Stacking ensemble model provided in this embodiment of the invention includes a weak classifier in the first layer and a secondary classifier in the second layer. The weak classifier includes SGD classifier, random forest classifier, extremely random tree classifier, Adaboost classifier, and GBDT classifier. In the Stacking ensemble model, the output of the weak classifier is used as the input of the secondary classifier.
[0046] Obtain a noisy dataset, perform feature engineering on the noisy dataset, and obtain a training set.
[0047] like Figure 6 As shown, the above feature engineering process includes feature cleaning and preprocessing. Feature cleaning includes imputing missing values and deduplicating feature values in noisy data. Missing value imputation uses the mean method before and after imputation, and feature value deduplication is based on the one-to-one relationship between timestamps and record values. Preprocessing involves feature dimensionality reduction and feature selection in the noisy dataset. Feature dimensionality reduction is performed using principal component analysis, and feature selection uses principal component analysis and correlation analysis to remove irrelevant features and fixed constant feature attributes from the noisy data. Feature reconstruction involves extracting statistical feature attributes to describe the abnormal patterns of the noisy data and combining them with the feature attributes after feature dimensionality reduction to generate a reconstructed feature form.
[0048] The reconstructed feature table is shown in Table 1:
[0049] Table 1 - Reconstruction Characteristics of Construction Noise Data
[0050]
[0051] The aforementioned noise data anomaly patterns include fixed value anomalies, constant period anomalies, excessively low value anomalies, extreme value anomalies, sudden anomalies, and trend anomalies. Among them, extreme value anomalies and sudden anomalies are noise data point anomaly patterns, while the remaining anomaly patterns are noise subsequence anomaly patterns. Noise data anomaly patterns are identified using construction engineering noise emission standards and noise time series data subsequence feature extraction methods.
[0052] Fixed value anomalies refer to noise data collected by the sensor that does not change within 30 minutes, or fixed value anomalies subsequences that fluctuate very little (less than a preset threshold) within 30 minutes compared to normal noise sequences. Fixed value anomalies are further divided into single fixed value anomalies and multiple fixed value anomalies: a single fixed value anomaly refers to a time series subsequence containing only a single fixed value, while multiple fixed value anomalies refer to a time series subsequence containing multiple fixed values. Figure 1 An example of a fixed value anomaly is given.
[0053] Construction project sensor data is affected by environmental and human factors, resulting in drastic fluctuations and highly random distribution. Considering the distribution of construction project sensor data and the characteristics of dust and noise data at actual construction sites, a constant-period anomaly subsequence is defined as one with small data fluctuations that remain stable within a threshold range over 30 minutes. Compared to fixed-value anomalies, although the data fluctuates, there are clear upper and lower limits to these fluctuations, and the difference between these limits is small. The trend term of the constant-period anomaly data, after seasonal decomposition of the time series, is fitted to a straight line. Figure 2 An example of a constant-period anomaly is given.
[0054] The low value anomaly refers to a pattern in which the recorded values of a subsequence with a time length of 30 minutes are significantly lower than the normal values. In construction projects, due to reasons such as equipment aging, failure, and human operation, there may be situations where the data values collected by sensors are lower than the actual values. The low value anomaly is often related to equipment quality issues. The diagnosis of this anomaly pattern involves a combination of domain knowledge, expert experience judgment, and machine learning.
[0055] Extreme value anomalies are point anomalies consisting of local maxima and minima within a time series subsequence. Unlike mutation anomalies, extreme value anomalies are localized anomalous features within a subsequence; from a global perspective, they may not be classified as anomalies. Their emphasis is on exhibiting anomalous characteristics within a segmented subsequence that differ from other local data. Therefore, a combination of data-driven and knowledge-driven methods is needed for diagnosis. Figure 3 An example of an extreme value anomaly is given.
[0056] A sudden anomaly is a globally abrupt change in a time-series dataset, including sudden increases and sudden decreases. Its statistical characteristics include a large range in the dataset and a significant difference between the maximum or minimum value and the mean of the series. Sudden anomalies are point anomalies, and data-driven unsupervised machine learning methods can achieve high accuracy in their detection. Figure 4 An example of a sudden anomaly is given.
[0057] Trend anomalies specifically include two types: 1) Within a 30-minute time interval, the noise sequence displays any two or more of the top five anomalies; 2) The data shows other characteristics that do not conform to the characteristics of the domain data, such as the average noise level at night being higher than during the day in noise monitoring data of construction projects. Figure 5 An example of an anomaly in the trend is given.
[0058] After obtaining the training set, the noisy data in the training set is input into the weak classifiers. The five weak classifiers are trained using the 5-fold cross-validation method. The output of the weak classifiers is used to train the secondary classifiers to obtain the trained anomaly detection model.
[0059] The dataset to be detected collected in step S1 is input into the pre-trained anomaly detection model to obtain a list of noise data anomalies, thus completing the anomaly detection of construction noise data in the construction project.
[0060] Example 2
[0061] The present invention provides a method for detecting abnormal construction noise data in construction projects, which is implemented through the following process:
[0062] This invention takes high-dimensional time-series data of construction noise from construction projects as the research object, and uses construction noise emission standards and noise time-series data subsequence feature extraction methods to identify abnormal patterns.
[0063] Each anomaly pattern is summarized into a rule, forming a list of anomaly pattern rule combinations. Considering the portability of anomaly detection methods and their applicability to construction engineering data, the following six anomaly patterns are retained: fixed value anomaly, constant period anomaly, excessively low value anomaly, extreme value anomaly, sudden anomaly, and trend anomaly. Among them, extreme value anomaly and sudden anomaly are anomaly patterns of noisy data points, while the remaining anomaly patterns are anomaly patterns of noisy subsequences.
[0064] Fixed value anomalies refer to noise data collected by the sensor that does not change within 30 minutes, or fixed value anomalies subsequences that fluctuate very little (less than a preset threshold) within 30 minutes compared to normal noise sequences. Fixed value anomalies are further divided into single fixed value anomalies and multiple fixed value anomalies: a single fixed value anomaly refers to a time series subsequence containing only a single fixed value, while multiple fixed value anomalies refer to a time series subsequence containing multiple fixed values. Figure 1 An example of a fixed value anomaly is given.
[0065] Construction project sensor data is affected by environmental and human factors, resulting in drastic fluctuations and highly random distribution. Considering the distribution of construction project sensor data and the characteristics of dust and noise data at actual construction sites, a constant-period anomaly subsequence is defined as one with small data fluctuations that remain stable within a threshold range over 30 minutes. Compared to fixed-value anomalies, although the data fluctuates, there are clear upper and lower limits to these fluctuations, and the difference between these limits is small. The trend term of the constant-period anomaly data, after seasonal decomposition of the time series, is fitted to a straight line. Figure 2 An example of a constant-period anomaly is given.
[0066] The low value anomaly refers to a pattern in which the recorded values of a subsequence with a time length of 30 minutes are significantly lower than the normal values. In construction projects, due to reasons such as equipment aging, failure, and human operation, there may be situations where the data values collected by sensors are lower than the actual values. The low value anomaly is often related to equipment quality issues. The diagnosis of this anomaly pattern involves a combination of domain knowledge, expert experience judgment, and machine learning.
[0067] Extreme value anomalies are point anomalies consisting of local maxima and minima within a time series subsequence. Unlike mutation anomalies, extreme value anomalies are localized anomalous features within a subsequence; from a global perspective, they may not be classified as anomalies. Their emphasis is on exhibiting anomalous characteristics within a segmented subsequence that differ from other local data. Therefore, a combination of data-driven and knowledge-driven methods is needed for diagnosis. Figure 3 An example of an extreme value anomaly is given.
[0068] A sudden anomaly is a globally abrupt change in a time-series dataset, including sudden increases and sudden decreases. Its statistical characteristics include a large range in the dataset and a significant difference between the maximum or minimum value and the mean of the series. Sudden anomalies are point anomalies, and data-driven unsupervised machine learning methods can achieve high accuracy in their detection. Figure 4 An example of a sudden anomaly is given.
[0069] Trend anomalies specifically include two types: 1) Within a 30-minute time interval, the noise sequence displays any two or more of the top five anomalies; 2) The data shows other characteristics that do not conform to the characteristics of the domain data, such as the average noise level at night being higher than during the day in noise monitoring data of construction projects. Figure 5 An example of an anomaly in the trend is given.
[0070] Obtain a noisy dataset and perform feature engineering on it:
[0071] like Figure 6 As shown, the noise dataset of the construction project is first obtained, and then feature engineering is performed on it. Combining the characteristics of the noise data and knowledge of the construction engineering field, the processing of the noise data mainly includes two parts: feature cleaning and preprocessing.
[0072] Feature cleaning includes imputing missing values and deduplicating feature values in noisy data. Missing value imputation uses the mean method before and after imputation, and feature value deduplication is based on the one-to-one relationship between timestamps and record values. Preprocessing involves feature dimensionality reduction and feature selection in the noisy dataset. Feature dimensionality reduction is performed using principal component analysis, and feature selection uses principal component analysis and correlation analysis to remove irrelevant features and fixed constant feature attributes from the noisy data. Feature reconstruction involves extracting statistical feature attributes that describe the abnormal patterns of the noisy data and combining them with the feature attributes after feature dimensionality reduction to generate a reconstructed feature form.
[0073] The reconstructed feature table is shown in Table 1:
[0074] Table 1 - Reconstruction Characteristics of Construction Noise Data
[0075]
[0076] The basic idea of ensemble learning is to combine multiple ordinary models with average performance to obtain a high-performance ensemble model. The core tasks of ensemble model are twofold: first, how to obtain multiple ordinary models with average performance, and second, how to effectively combine these models to obtain a high-performance ensemble model.
[0077] Stacking ensemble models are hierarchical model ensemble frameworks that integrate at least two or more classifiers, and can be considered as "model stacking". Stacking ensemble learning methods can effectively fuse multiple classifiers through a tree structure, thereby improving classification accuracy and precision. Typically, the first weak classifier in Stacking will select a model with high fit in order to learn fully from the training data.
[0078] like Figure 7 As shown, the Stacking ensemble model provided in this embodiment of the invention includes a weak classifier in the first layer and a secondary classifier in the second layer. The weak classifier includes SGD classifier, random forest classifier, extremely random tree classifier, Adaboost classifier, and GBDT classifier. In the Stacking ensemble model, the output of the weak classifier is used as the input of the secondary classifier.
[0079] like Figure 7 As shown, based on feature engineering, n datasets are constructed according to actual training needs, and then m datasets are randomly selected from the datasets as training sets; combined with anomaly patterns, SGD classifier, random forest classifier, extremely random tree classifier, Adaboost classifier, and GBDT classifier are selected as weak classifiers for training.
[0080] The training of the weak classifiers is as follows: Based on the cross-validation method, five weak classifiers are trained so that the prediction results of the weak classifiers on the test set can be used as the input of the logistic regression secondary classifier of the Stacking ensemble model during testing; in order to prevent data overfitting, a five-fold cross-validation method is used to improve the robustness of the model.
[0081] The training set is randomly divided into n equal parts. Each time, n-1 parts are randomly selected with replacement as the training set, and the remaining dataset is used as the test set. This process is repeated n times to obtain n test results. The final prediction result is obtained by averaging the test results.
[0082] Table 2 - Noise Data Anomaly Detection Process Based on Stacking Ensemble Model
[0083]
[0084]
[0085] In Table 2, processes 1, 2, and 3 are the processes of training the weak classifier (i.e., the primary classifier), processes 5 to 9 are the prediction results of the trained weak classifier, and the prediction results are used as the training set of the secondary classifier. Process 11 is the training of the secondary classifier using the prediction results of the weak classifier, resulting in the final trained Stacking ensemble model.
[0086] After the Stacking ensemble model is trained, the noise dataset to be detected is input into it to obtain a list of noise data anomalies, thus completing the anomaly detection.
[0087] This invention provides a method for detecting anomalies in construction noise data. First, by identifying common anomaly patterns in the noise data, the original data undergoes feature reconstruction based on domain knowledge and statistical methods. Then, the anomaly detection problem is transformed into a multi-classification problem in machine learning. Five weak classifiers are selected, and the reconstructed feature dataset and anomaly pattern list are input for training. Five-fold cross-validation is used to improve the robustness of the learning process. Finally, based on a Stacking ensemble model, the output of the weak classifiers is used as the input of the secondary classifiers for iterative training. The trained Stacking ensemble model is stored, and anomaly detection is performed using the trained model to output a list of noise data anomalies.
[0088] Example 3
[0089] This invention provides a construction noise data anomaly detection system, comprising:
[0090] Test Data Acquisition Module: Used to acquire the dataset to be tested;
[0091] Anomaly detection module: Used to input the dataset to be detected into a pre-trained anomaly detection model to obtain a list of noise data anomalies and complete the anomaly detection of construction noise data in construction projects.
[0092] Example 4
[0093] An embodiment of the present invention provides a construction noise data anomaly detection device, comprising a processor and a storage medium;
[0094] The storage medium is used to store instructions;
[0095] The processor is configured to operate according to the instructions to perform the steps of the following method:
[0096] Obtain the dataset to be detected;
[0097] The dataset to be detected is input into a pre-trained anomaly detection model to obtain a list of noise data anomalies, thus completing the anomaly detection of construction noise data in construction projects.
[0098] Example 5
[0099] The computer-readable storage medium provided in this embodiment of the invention stores a computer program thereon, which, when executed by a processor, implements the steps of the following method:
[0100] Obtain the dataset to be detected;
[0101] The dataset to be detected is input into a pre-trained anomaly detection model to obtain a list of noise data anomalies, thus completing the anomaly detection of construction noise data in construction projects.
[0102] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0103] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0106] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for detecting abnormal construction noise data in construction projects, characterized in that, include: Obtain the dataset to be detected; The dataset to be detected is input into a pre-trained anomaly detection model to obtain a list of noise data anomalies, thus completing the anomaly detection of construction noise data in construction projects. The anomaly detection model is a Stacking ensemble model, which includes a weak classifier in the first layer and a secondary classifier in the second layer. The weak classifiers include SGD classifier, random forest classifier, extremely random tree classifier, Adaboost classifier, and GBDT classifier. In the Stacking ensemble model, the output of the weak classifier is used as the input of the secondary classifier; The anomaly detection model is trained using the following method: Obtain a noisy dataset, perform feature engineering on the noisy dataset to obtain a training set; Noisy data from the training set is input into weak classifiers. Five weak classifiers are trained using the 5-fold cross-validation method. The output of the weak classifiers is used to train the secondary classifiers to obtain a trained anomaly detection model. The feature engineering process for the noisy dataset includes: feature cleaning and preprocessing followed by feature reconstruction; Feature cleaning includes filling missing values and deduplicating feature values in noisy data. Missing value filling uses the front-to-back averaging method, and feature value deduplication is based on the one-to-one relationship between timestamps and record values. Preprocessing involves dimensionality reduction and feature selection of the noisy dataset. Dimensionality reduction is achieved through principal component analysis, while feature selection removes irrelevant features and fixed constant feature attributes from the noisy data. Feature reconstruction involves extracting statistical feature attributes to describe the abnormal patterns in noisy data, and combining these attributes with those obtained after dimensionality reduction to generate a reconstructed feature table. The reconstructed feature table includes the mean, maximum, minimum, range, standard deviation, 25th percentile, 50th percentile, 75th percentile, kurtosis, and skewness. The abnormal noise data patterns are identified using construction project noise emission standards and noise time series data subsequence feature extraction methods. The noise data anomaly patterns include fixed value anomaly, constant period anomaly, excessively low value anomaly, extreme value anomaly, sudden anomaly, and trend anomaly. Among them, extreme value anomaly and sudden anomaly are noise data point anomaly patterns, while the other anomaly patterns are noise subsequence anomaly patterns. Fixed value anomalies refer to noise data that does not change or fluctuates within a preset threshold within 30 minutes. They include single fixed value anomalies and multiple fixed value anomalies. A single fixed value anomaly means that there is only a single fixed value in a time series subsequence, while multiple fixed value anomalies mean that there are multiple fixed values in a time series subsequence. A constant periodic anomaly refers to noisy data that remains stable within a threshold range for 30 minutes, and whose trend term after seasonal decomposition of the time series can be fitted as a straight line. An abnormally low value refers to a subsequence record value that is lower than the normal value within a time period of 30 minutes; Extreme value anomalies refer to point anomalies of local maxima and minima in a time series subsequence; Sudden anomalies refer to a sudden change in the global range of a time series data point, including sudden increases and sudden decreases; Trend anomalies refer to the occurrence of two or more anomalies within 30 minutes, or the display of other characteristics that do not conform to the characteristics of domain data.
2. A construction noise data anomaly detection system based on the method of claim 1, characterized in that, include: Test Data Acquisition Module: Used to acquire the dataset to be tested; Anomaly detection module: Used to input the dataset to be detected into a pre-trained anomaly detection model to obtain a list of noise data anomalies and complete the anomaly detection of construction noise data in construction projects.
3. A device for detecting abnormal construction noise data in construction projects, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to claim 1.
4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method of claim 1.