AI-based fixed-point and quantitative air leakage monitoring and alarm system and method for ventilation ducts
Through the fixed-point quantitative air leakage monitoring system based on AI, the sensor network and autoencoder model are used to identify air leakage in the ventilation duct in real time, solving the problems of low detection efficiency and poor positioning accuracy in traditional methods, and achieving efficient and automated air leakage monitoring and alarm .
Patent Information
- Application Number
- CN202410666526.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-28
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-28
AI Technical Summary
Traditional ventilation duct leakage monitoring methods have low detection efficiency, poor positioning accuracy and slow response speed, and cannot realize real-time monitoring, resulting in delayed detection and aggravation of air leakage problems.
Using a fixed-point quantitative air leakage monitoring and alarm system based on AI, the pressure, air flow and air flow noise data are collected by building a sensor network, and normal samples and abnormal samples are divided using density-based clustering algorithms, and a full convolutional autoencoder network training is designed. Air leakage detection model, real-time monitoring and triggering air leakage alarm.
Real-time monitoring and rapid identification of air leakage events are achieved, which reduces the dependence on manual inspections, improves the accuracy and automation of air leakage detection, reduces labor costs, and optimizes the monitoring and maintenance efficiency of ventilation systems.
Smart Images

Figure CN118443226B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to an AI-based ventilation duct fixed-point and quantitative air leakage monitoring and alarm system and method. Background Art
[0002] Ventilation duct systems play a vital role in industrial, commercial and residential buildings. Their main uses include conveying air, removing exhaust gas and regulating indoor temperature. These systems ensure indoor air quality and environmental comfort, which have a positive impact on protecting personnel health and production efficiency. However, the sealing and integrity of the ducts are the basis for maintaining system efficiency and safety. Once air leakage occurs, it will not only reduce system efficiency and increase energy consumption, but may also lead to safety hazards, such as leakage of harmful gases.
[0003] Traditionally, air leakage monitoring relies mainly on regular physical inspections and manual testing, such as using tools such as smoke tests or stethoscopes to identify air leakage points. Although these methods are effective to a certain extent, they have obvious defects such as low detection efficiency, poor positioning accuracy, and slow response speed. First, regular inspections cannot achieve real-time monitoring, resulting in air leakage problems that may have existed for some time before they are discovered, thereby exacerbating the severity of the problem. Secondly, manual inspections are difficult to accurately locate air leakage points, especially when the pipeline system is complex or hidden, which brings additional difficulties to maintenance work. Finally, the process from discovering air leakage to taking maintenance measures is time-consuming, which not only delays the resolution of the problem, but also increases the operational risk of the system.
[0004] Given these limitations of traditional methods, there is an urgent need for a new technology that can improve detection efficiency, speed up response, monitor around the clock, and detect air leakage problems in a timely manner. Summary of the invention
[0005] The present application aims to solve at least one of the technical problems in the related art to a certain extent. To this end, one purpose of the present application is to propose an AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm system and method, which realizes real-time monitoring and timely response to air leakage risks in the ventilation duct.
[0006] One aspect of the present application provides an AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method, including:
[0007] Step S100: constructing a sensor network, using sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline;
[0008] The specific method of constructing a sensor network and using sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline is:
[0009] Step S110: deploying a pressure sensor, a flow meter, and an acoustic wave sensor in a potential air leakage area in the ventilation duct;
[0010] Step S120: using a pressure sensor to collect pressure data p in the ventilation duct, and using the continuous time pressure data to obtain a pressure change sequence;
[0011] Step S130: using a flow meter to monitor the fluctuation of air flow, obtaining air flow data k, and using the continuous time air flow data to obtain an air flow change sequence;
[0012] Step S140: using an acoustic wave sensor to monitor the acoustic waves in the ventilation duct to obtain an airflow noise number q, and obtaining an airflow noise sequence according to the airflow noise data of continuous time;
[0013] Step S200: dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples by using a density-based clustering algorithm;
[0014] The specific method of dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples by using a density-based clustering algorithm is as follows:
[0015] Step S210: normalizing the pressure change sequence, the air flow change sequence and the airflow noise sequence, and scaling them to the interval [0, 1];
[0016] Step S220: dividing the pressure change sequence, the air flow change sequence and the airflow noise sequence into N samples by using a sliding window method;
[0017] Step S230: extracting statistics in each window according to the sliding window method, wherein the statistics constitute a feature vector of the window, including: mean, variance, and peak value;
[0018] The specific method of extracting the statistics in each window according to the sliding window method, and the statistics constituting the feature vector of the window is:
[0019] Step S231: Preset the window size α, prediction time step c and sliding step h, and obtain the i-th sample {(P i , K i , Q i ), y i}, where y is a data label, and the data label is divided into normal ventilation and abnormal air leakage;
[0020] Step S232: For each window, calculate the mean, variance, and peak value within the window;
[0021] The mean value includes the pressure mean value, the air flow mean value and the airflow noise mean value. The calculation formula of the pressure mean value is: Among them, p β represents the βth pressure data in the window, and the calculation formula of the mean air flow rate is: Among them, k β represents the βth air flow data in the window, and the calculation formula of the airflow noise mean is: Among them, q β represents the βth airflow noise data in the window;
[0022] The variance includes pressure variance, air flow variance and airflow noise variance. The calculation formula of the pressure variance is: The calculation formula of the air flow variance is: The calculation formula of the airflow noise variance is:
[0023]
[0024] The peak values include pressure peak value, air flow peak value and airflow noise peak value. The expression of pressure peak value is: p max =max(p1,p2,...,p α ), the expression of the peak air flow rate is: k max =max(k1,k2,...,k α ), the expression of the airflow noise peak is: q max =max(q1,q2,...,q α );
[0025] Step S233: Calculate the mean, variance, and peak value in all windows to form the feature vector of the window;
[0026] Step S240: taking the samples corresponding to each window as sample points, clustering the sample points using a density-based clustering algorithm, automatically identifying the types of sample points, dividing the core points and their boundary points into high-density clusters to obtain normal samples, and dividing the noise points into low-density clusters to obtain abnormal samples;
[0027] The specific method of taking the samples corresponding to each window as sample points, clustering the sample points using a density-based clustering algorithm, automatically identifying the types of sample points, dividing the core points and their boundary points into high-density clusters to obtain normal samples, and dividing the noise points into low-density clusters to obtain abnormal samples is as follows:
[0028] Step S241: Consider the sample corresponding to the statistic of each window as a sample point a, and calculate all sample point pairs (a i ,a j ) between i ,a j );
[0029] Step S242: Preset the neighborhood radius eps and the minimum number of samples Sam min ;
[0030] The preset neighborhood radius eps and the minimum number of samples Sam min The specific method is:
[0031] Define the sample point set A = {a1, a2, ..., a N}, N represents the total number of sample points in the sample point set, and the c clusters obtained by clustering the sample points using the density-based clustering algorithm are defined as C = {C1, C2, ..., C c};
[0032] Define the objective function
[0033] Among them, C ci represents the ci-th cluster obtained by clustering, |C ci | represents the number of sample points in the ci-th cluster, CP(C ci ) represents the compactness of the ci-th cluster, Noise(C) represents the noise ratio of noise points that are not assigned to any cluster to the total number of sample points, and λ is the regularization coefficient that balances the compactness of the cluster and the noise ratio;
[0034] The compactness of the cluster CP(C ci ) is:
[0035]
[0036] The expression of noise ratio Noise(C) is:
[0037] For the neighborhood radius eps and the minimum number of samples Sam min Perform a grid search, using the current neighborhood radius eps and the minimum number of samples Sam at each step of the search min Run the density-based clustering algorithm to obtain c clusters and calculate the value of the objective function Ω;
[0038] Choose the neighborhood radius eps and the minimum number of samples Sam that maximize the objective function Ω min As the optimal solution, as the final clustering parameters using density-based clustering algorithm;
[0039] Step S243: For each sample point a i , statistical sample point a i The number of samples Sam within the neighborhood radius eps;
[0040] Step S244: Compare the sample number Sam with the minimum sample number Sam min If Sam is greater than or equal to the minimum number of samples Sam min , then the current sample point a i Marked as core point a i ′;
[0041] Step S245: For core point a i ′, add all sample points within its neighborhood radius eps to the same cluster. If the sample points within its neighborhood radius contain other core points, the sample points within the neighborhood radius of the core point are also classified into the same cluster;
[0042] Step S246: marking non-core points within the neighborhood radius of the core point as boundary points, and marking sample points that do not belong to any cluster as noise points;
[0043] Step S247: until all sample points are divided, the sample points divided into the same cluster are marked as normal samples, and the cluster is a high-density cluster, and the noise points that do not belong to the high-density cluster are marked as abnormal samples and divided into a low-density cluster;
[0044] Step S300: designing a fully convolutional autoencoder network, and using normal samples and abnormal samples to train the fully convolutional autoencoder network to obtain an air leakage detection model;
[0045] The air leakage detection model is a model for detecting ventilation abnormality scores in ventilation ducts;
[0046] The specific method of designing a fully convolutional autoencoder network and training the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model is:
[0047] Step S310: designing a fully convolutional autoencoder network, wherein the fully convolutional autoencoder network includes: an encoder, a decoder, and an output of the encoder and an output of the decoder;
[0048] Step S320: using normal samples and abnormal samples as training samples for training an air leakage detection model, wherein the label corresponding to the normal sample is 0, and the label corresponding to the abnormal sample is 1;
[0049] Step S330: designing a loss function Loss, wherein the loss function includes a reconstruction loss and a classification loss, wherein the reconstruction loss selects a smooth L1 loss, and the classification loss selects a binary cross entropy loss;
[0050] The functional expression of the reconstruction loss is:
[0051] Among them, x is the output of the encoder, and y represents the true label of the training sample;
[0052] The functional expression of the classification loss is:
[0053] Among them, y i represents the true label of the i-th training sample, It indicates the probability of predicting that the i-th training sample is an abnormal sample. express The logarithm of express The logarithm of
[0054] The functional expression of the loss function Loss is:
[0055] Among them, ε is the trade-off coefficient;
[0056] Step S340: Divide all training samples into n batches, and input the training samples in the mth batch as input data into the encoder to obtain the output of the encoder, and pass the output of the encoder through the decoder to obtain a reconstructed sequence;
[0057] Step S350: Calculate the reconstruction loss and classification loss to obtain the value of the loss function Loss, and calculate the gradient value ▽Loss(θ m) of the loss function of the batch to the model parameter θ m m ), use the optimizer to update the model parameters to get θm +1 , the model parameter θm +1 The calculation formula is: m+1 =θ m -η×▽Loss(θ m ), where θ m is the model parameter of the current batch, θ m+1 is the updated model parameter, η is the learning rate;
[0058] Step S360: Preset a maximum number of training rounds L. When the number of training rounds reaches the maximum number of training rounds, stop training to obtain a trained air leakage detection model;
[0059] Step S400: input the normal sample into the trained air leakage detection model, output the reconstruction loss of the normal sample, and calculate the mean and standard deviation of the reconstruction loss of the normal sample;
[0060] The normal sample is input into the trained air leakage detection model, and the reconstruction loss of the normal sample is output. The specific method of calculating the mean and standard deviation of the reconstruction loss of the normal sample is:
[0061] Step S410: Input the normal samples in the training samples into the trained air leakage detection model, input the encoder to obtain the output of the encoder, pass the output of the encoder through the decoder to obtain the reconstruction sequence, and calculate the reconstruction loss of the normal samples according to the reconstruction sequence. Among them, X i represents the normal sample of the i-th input, Represents the reconstructed sequence of the output corresponding to the normal sample of the i-th input;
[0062] Step S420: Calculate the mean and standard deviation of the reconstruction loss of normal samples;
[0063] The calculation formula of the mean value of the reconstruction loss of the normal sample is:
[0064] Among them, N zc represents the number of normal samples;
[0065] The calculation formula of the standard deviation of the reconstruction loss of the normal sample is:
[0066]
[0067] Step S500: collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, output the label type of the real-time sample, and if the label type is abnormal, calculate the ventilation abnormality score using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample;
[0068] The real-time pressure, real-time air flow and real-time airflow noise are collected and processed to obtain real-time samples as input data of the air leakage detection model, and the label type of the real-time sample is output. If the label type is abnormal, the specific method of calculating the ventilation abnormality score by using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample is:
[0069] Step S510: using the deployed sensor network to collect real-time pressure, real-time air flow and real-time airflow noise, and sorting the real-time pressure, real-time air flow and real-time airflow noise according to time to obtain a real-time pressure sequence, a real-time air flow sequence and a real-time airflow noise sequence;
[0070] Step S520: dividing the real-time pressure sequence, the real-time air flow sequence and the real-time airflow noise sequence by a sliding window method to obtain N′ real-time samples;
[0071] Step S530: using N′ real-time samples as input data of the air leakage detection model, inputting them into an encoder to obtain the output of the encoder, using the output of the encoder through a decoder to obtain a reconstructed sequence of real-time samples, and obtaining the reconstruction loss of the real-time samples through the reconstructed sequence of the real-time samples;
[0072] Step S540: obtaining the label type of the real-time sample according to the output data of the air leakage detection model; if the label type of the real-time sample is abnormal, calculating the ventilation abnormality score of the real-time sample according to the reconstruction loss of the real-time sample and the mean and standard deviation of the reconstruction loss of the normal sample;
[0073] The calculation formula of the ventilation abnormality score is:
[0074] in, represents the reconstruction loss of real-time samples, x now represents a real-time sample of the input, A reconstructed sequence of real-time samples representing the output of the air leakage detection model;
[0075] Step S600: Establishing an abnormal response mechanism, grading the ventilation abnormality scores, and triggering air leakage alarms of different abnormality levels;
[0076] The specific method of establishing an abnormal response mechanism, grading the ventilation abnormality scores, and triggering air leakage alarms of different abnormality degrees is as follows:
[0077] Step S610: setting a slight abnormality threshold U1, a moderate abnormality threshold U2 and a severe abnormality threshold U3;
[0078] Step S620: classifying the ventilation abnormality score according to the slight abnormality threshold U1, the moderate abnormality threshold U2 and the severe abnormality threshold U3;
[0079] When W<U1, the ventilation state of the ventilation duct is normal;
[0080] When U1≤W<U2, the ventilation state of the ventilation duct is slightly abnormal;
[0081] When U2≤W<U3, the ventilation state of the ventilation duct is moderately abnormal;
[0082] When W ≥ U3, the ventilation status of the ventilation duct is seriously abnormal;
[0083] Step S630: triggering different air leakage alarms according to the degree of abnormality of the ventilation state of the ventilation duct.
[0084] One aspect of the present application provides an AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm system, including:
[0085] The sensor deployment data collection module is used to build a sensor network and use sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline;
[0086] The sample data clustering and division module is used to divide the pressure data, air flow data and airflow noise data into normal samples and abnormal samples through a density-based clustering algorithm;
[0087] The network design model training module is used to design a fully convolutional autoencoder network and train the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model.
[0088] A normal sample standard calculation module is used to input normal samples into the trained air leakage detection model, output the reconstruction loss of the normal samples, and calculate the mean and standard deviation of the reconstruction loss of the normal samples;
[0089] The ventilation anomaly score calculation module is used to collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, and output the label type of the real-time sample. If the label type is abnormal, the ventilation anomaly score is calculated using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample;
[0090] The air leakage degree grading alarm module is used to establish an abnormal response mechanism, grade the ventilation abnormality scores, and trigger air leakage alarms of different abnormal degrees.
[0091] One aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the AI-based fixed-point and quantitative air leakage monitoring and alarm method for ventilation ducts are implemented.
[0092] One aspect of the present application provides a readable storage medium storing a computer program suitable for loading by a processor to execute the steps in an AI-based fixed-point and quantitative air leakage monitoring and alarm method for ventilation ducts.
[0093] Compared with the prior art, the ventilation duct fixed-point quantitative air leakage monitoring and alarm system and method proposed in this application have the following advantages:
[0094] This application can monitor and quickly identify potential air leakage events in real time through automatic cluster analysis, reducing the time and resource consumption of manual inspection. It automatically distinguishes normal samples from abnormal samples through density-based clustering algorithms, reducing the dependence on professional knowledge and improving the automation of air leakage identification. It automatically finds the most optimized clustering parameters through grid search with preset neighborhood radius and minimum number of samples, so that the performance of the algorithm can be maintained at the best state under different working conditions.
[0095] This application designs and trains a fully convolutional autoencoder network as an air leakage detection model to learn the deep features of the input data. By combining reconstruction errors and classification labels, the model can more accurately distinguish normal samples from abnormal samples, thereby improving the accuracy of air leakage detection. The automated air leakage detection model reduces the reliance on regular physical inspections and manual testing, reduces the labor cost of air leakage detection and positioning, and also improves the continuity and coverage of detection.
[0096] This application combines the reconstruction loss of real-time samples with the reconstruction loss statistics of normal samples to dynamically calculate the ventilation anomaly score, which can more accurately evaluate the degree of abnormality of real-time samples;
[0097] This application establishes a degree grading and response mechanism based on abnormality scores, which can not only effectively improve the monitoring and maintenance efficiency of the ventilation system, but also take targeted measures according to the severity of the abnormality, optimize resource allocation, and enhance system safety, demonstrating its innovation and practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0098] Figure 1 A method flow chart of the AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method provided in this application;
[0099] Figure 2 Functional module diagram of the AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm system provided in this application;
[0100] Figure 3 A schematic diagram of the structure of an electronic device provided in this application;
[0101] Figure 4 It is a schematic diagram of the structure of a readable storage medium provided by this application. DETAILED DESCRIPTION
[0102] In order to better understand the present application, various aspects of the present application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of exemplary embodiments of the present application and do not limit the scope of the present application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.
[0103] In the accompanying drawings, the size, dimensions and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are for illustration only and are not strictly drawn to scale. As used herein, the terms "substantially", "approximately" and similar terms are used as terms of approximation, not as terms of degree, and are intended to illustrate the inherent deviations in measurements or calculations that will be recognized by those of ordinary skill in the art. In addition, in this application, the order in which the steps are described does not necessarily represent the order in which these processes occur in actual operation, unless otherwise specified or can be derived from the context.
[0104] It should also be understood that expressions such as "including", "comprising", "having", "containing" and / or "comprising" are open rather than closed expressions in this specification, which indicate the presence of the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features rather than just the individual elements in the list. In addition, when describing embodiments of the present application, "may" is used to mean "one or more embodiments of the present application". And, the term "exemplary" is intended to refer to an example or illustration.
[0105] Unless otherwise specified, all words (including engineering terms and scientific and technological terms) used in this article have the same meaning as those commonly understood by ordinary technicians in the field to which this application belongs. It should also be understood that, unless clearly stated in this application, words defined in common dictionaries should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.
[0106] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0107] Example 1
[0108] like Figure 1 As shown, the AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method provided in this application includes:
[0109] Step S100: constructing a sensor network, using sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline;
[0110] The specific method of constructing a sensor network and using sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline is:
[0111] Step S110: deploying a pressure sensor, a flow meter and an acoustic wave sensor in a potential air leakage area in the ventilation duct;
[0112] Step S120: using a pressure sensor to collect pressure data p in the ventilation duct, and using the continuous time pressure data to obtain a pressure change sequence;
[0113] Step S130: using a flow meter to monitor the fluctuation of air flow, obtaining air flow data k, and using the continuous time air flow data to obtain an air flow change sequence;
[0114] Step S140: using an acoustic wave sensor to monitor the acoustic waves in the ventilation duct to obtain an airflow noise number q, and obtaining an airflow noise sequence according to the airflow noise data of continuous time;
[0115] This step can accurately collect data by using professional sensors such as pressure sensors, flow meters and acoustic sensors, which improves the accuracy and reliability of monitoring data and provides a basis for subsequent data analysis. By analyzing these data, abnormal patterns can be identified to find the existence of air leakage;
[0116] Step S200: dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples by using a density-based clustering algorithm;
[0117] The normal sample refers to the pressure data, air flow data and airflow noise data monitored during normal ventilation without air leakage in the ventilation duct;
[0118] The abnormal samples refer to the pressure data, air flow data and airflow noise data monitored when air leakage occurs in the ventilation duct;
[0119] The specific method of dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples by a density-based clustering algorithm is:
[0120] Step S210: normalizing the pressure change sequence, the air flow change sequence and the airflow noise sequence, and scaling them to the interval [0, 1];
[0121] Step S220: dividing the pressure change sequence, the air flow change sequence and the airflow noise sequence into N samples by using a sliding window method;
[0122] Step S230: extracting statistics in each window according to the sliding window method, wherein the statistics constitute a feature vector of the window, including: mean, variance, and peak value;
[0123] The specific method of extracting the statistics in each window according to the sliding window method, and the statistics constituting the feature vector of the window is:
[0124] Step S231: Preset the window size α, prediction time step c and sliding step h, and obtain the i-th sample {(P i , K i , Q i ), y i}, where y is a data label, and the data label is divided into normal ventilation and abnormal air leakage;
[0125] Step S232: For each window, calculate the mean, variance, and peak value within the window;
[0126] The mean value includes the pressure mean value, the air flow mean value and the airflow noise mean value. The calculation formula of the pressure mean value is: Among them, p β represents the βth pressure data in the window, and the calculation formula of the mean air flow rate is: Among them, k β represents the βth air flow data in the window, and the calculation formula of the airflow noise mean is: Among them, q β represents the βth airflow noise data in the window;
[0127] The variance includes pressure variance, air flow variance and airflow noise variance. The calculation formula of the pressure variance is: The calculation formula of the air flow variance is: The calculation formula of the airflow noise variance is:
[0128] The peak values include pressure peak value, air flow peak value and airflow noise peak value. The expression of pressure peak value is: p max =max(p1,p2,...,p α ), the expression of the peak air flow rate is: k max =max(k1,k2,...,k α ), the expression of the airflow noise peak is: q max=max(q1,q2,...,q α );
[0129] Step S233: Calculate the mean, variance, and peak value in all windows to form the feature vector of the window;
[0130] Step S240: taking the samples corresponding to each window as sample points, clustering the sample points using a density-based clustering algorithm, automatically identifying the types of sample points, dividing the core points and their boundary points into high-density clusters to obtain normal samples, and dividing the noise points into low-density clusters to obtain abnormal samples;
[0131] The specific method of taking the samples corresponding to each window as sample points, clustering the sample points using a density-based clustering algorithm, automatically identifying the types of sample points, dividing the core points and their boundary points into high-density clusters to obtain normal samples, and dividing the noise points into low-density clusters to obtain abnormal samples is as follows:
[0132] Step S241: Consider the sample corresponding to the statistic of each window as a sample point a, and calculate all sample point pairs (a i ,a j ) between i ,a j );
[0133] Step S242: Preset the neighborhood radius eps and the minimum number of samples Sam min ;
[0134] The preset neighborhood radius eps and the minimum number of samples Sam min The specific method is:
[0135] Define the sample point set A = {a1, a2, ..., a N}, N represents the total number of sample points in the sample point set, and the c clusters obtained by clustering the sample points using the density-based clustering algorithm are defined as C = {C1, C2, ..., C c};
[0136] Define the objective function
[0137] Among them, C ci represents the ci-th cluster obtained by clustering, |C ci | represents the number of sample points in the ci-th cluster, CP(C ci ) represents the compactness of the ci-th cluster, Noise(C) represents the noise ratio of noise points that are not assigned to any cluster to the total number of sample points, and λ is the regularization coefficient that balances the compactness of the cluster and the noise ratio;
[0138] The compactness of the cluster CP(C ci) is:
[0139]
[0140] The expression of noise ratio Noise(C) is:
[0141] For the neighborhood radius eps and the minimum number of samples Sam min Perform a grid search, using the current neighborhood radius eps and the minimum number of samples Sam at each step of the search min Run the density-based clustering algorithm to obtain c clusters and calculate the value of the objective function Ω;
[0142] Choose the neighborhood radius eps and the minimum number of samples Sam that maximize the objective function Ω min As the optimal solution, as the final clustering parameters using density-based clustering algorithm;
[0143] Step S243: For each sample point a i , statistical sample point a i The number of samples Sam within the neighborhood radius eps;
[0144] Step S244: Compare the sample number Sam with the minimum sample number Sam min If Sam is greater than or equal to the minimum number of samples Sam min , then the current sample point a i Marked as core point a i ′;
[0145] Step S245: For core point a i ′, add all sample points within its neighborhood radius eps to the same cluster. If the sample points within its neighborhood radius contain other core points, the sample points within the neighborhood radius of the core point are also classified into the same cluster;
[0146] Step S246: marking non-core points within the neighborhood radius of the core point as boundary points, and marking sample points that do not belong to any cluster as noise points;
[0147] Step S247: until all sample points are divided, the sample points divided into the same cluster are marked as normal samples, and the cluster is a high-density cluster, and the noise points that do not belong to the high-density cluster are marked as abnormal samples and divided into a low-density cluster;
[0148] The basic idea of the density-based clustering algorithm is: for a given sample point, if there is at least Sam minIf the number of samples is greater than , then the sample point is regarded as a core point, and all points that can be densely connected by the core point form a cluster. The remaining sample points are marked as boundary points or noise points according to whether they are within the neighborhood radius eps of the core point.
[0149] This step uses a density-based clustering algorithm to analyze the monitoring data in the ventilation duct, which can not only achieve efficient detection and accurate positioning of air leakage, but also provide a data-driven solution and provide scientific decision-making support for the maintenance and management of the ventilation system;
[0150] Step S300: designing a fully convolutional autoencoder network, and using normal samples and abnormal samples to train the fully convolutional autoencoder network to obtain an air leakage detection model;
[0151] The air leakage detection model is a model for detecting ventilation abnormality scores in ventilation ducts;
[0152] The specific method of designing a fully convolutional autoencoder network and training the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model is:
[0153] Step S310: designing a fully convolutional autoencoder network, wherein the fully convolutional autoencoder network includes: an encoder, a decoder, and an output of the encoder and an output of the decoder;
[0154] Step S320: using normal samples and abnormal samples as training samples for training an air leakage detection model, wherein the label corresponding to the normal sample is 0, and the label corresponding to the abnormal sample is 1;
[0155] Step S330: designing a loss function Loss, wherein the loss function includes a reconstruction loss and a classification loss, wherein the reconstruction loss selects a smooth L1 loss, and the classification loss selects a binary cross entropy loss;
[0156] The functional expression of the reconstruction loss is:
[0157] Among them, x is the output of the encoder, and y represents the true label of the training sample;
[0158] The functional expression of the classification loss is:
[0159] Among them, y i represents the true label of the i-th training sample, It indicates the probability of predicting that the i-th training sample is an abnormal sample. express The logarithm of express The logarithm of
[0160] The functional expression of the loss function Loss is:
[0161] Among them, ε is the trade-off coefficient;
[0162] Step S340: Divide all training samples into n batches, and input the training samples in the mth batch as input data into the encoder to obtain the output of the encoder, and pass the output of the encoder through the decoder to obtain a reconstructed sequence;
[0163] Step S350: Calculate the reconstruction loss and classification loss to obtain the value of the loss function Loss, and calculate the gradient value ▽Loss(θ m) of the loss function of the batch to the model parameter θ m m ), use the optimizer to update the model parameters to get θ m+1 , the model parameters θ m+1 The calculation formula is: m+1 =θ m -η×▽Loss(θ m ), where θ m is the model parameter of the current batch, θ m+1 is the updated model parameter, η is the learning rate;
[0164] Step S360: Preset a maximum number of training rounds L. When the number of training rounds reaches the maximum number of training rounds, stop training to obtain a trained air leakage detection model;
[0165] The maximum number of training rounds L refers to the maximum number of training rounds for training the air leakage detection model, which is preset according to the actual model training requirements;
[0166] In this step, a fully convolutional autoencoder network is designed and trained as an air leakage detection model. By combining reconstruction loss and classification loss for optimization, the fully convolutional autoencoder can learn the deep features of the input data. By combining reconstruction error and classification labels, the model can more accurately distinguish normal samples from abnormal samples, thereby improving the accuracy of air leakage detection.
[0167] Step S400: input the normal sample into the trained air leakage detection model, output the reconstruction loss of the normal sample, and calculate the mean and standard deviation of the reconstruction loss of the normal sample;
[0168] The normal sample is input into the trained air leakage detection model, and the reconstruction loss of the normal sample is output. The specific method of calculating the mean and standard deviation of the reconstruction loss of the normal sample is:
[0169] Step S410: Input the normal samples in the training samples into the trained air leakage detection model, input the encoder to obtain the output of the encoder, pass the output of the encoder through the decoder to obtain the reconstruction sequence, and calculate the reconstruction loss of the normal samples according to the reconstruction sequence. Among them, X i represents the normal sample of the i-th input, Represents the reconstructed sequence of the output corresponding to the normal sample of the i-th input;
[0170] Step S420: Calculate the mean and standard deviation of the reconstruction loss of normal samples;
[0171] The calculation formula of the mean value of the reconstruction loss of the normal sample is:
[0172] Among them, N zc represents the number of normal samples;
[0173] The calculation formula of the standard deviation of the reconstruction loss of the normal sample is:
[0174]
[0175] In this step, by inputting normal samples into the trained air leakage detection model and calculating the mean and standard deviation of the reconstruction loss, a comparison benchmark of normal samples is provided for the subsequent calculation of the ventilation anomaly score of real-time samples.
[0176] Step S500: collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, output the label type of the real-time sample, and if the label type is abnormal, calculate the ventilation abnormality score using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample;
[0177] The label type of the real-time sample refers to the classification of the ventilation status in the ventilation duct, including abnormal and normal;
[0178] The real-time pressure, real-time air flow and real-time airflow noise are collected and processed to obtain real-time samples as input data of the air leakage detection model, and the label type of the real-time sample is output. If the label type is abnormal, the specific method of calculating the ventilation abnormality score by using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample is:
[0179] Step S510: using the deployed sensor network to collect real-time pressure, real-time air flow and real-time airflow noise, and sorting the real-time pressure, real-time air flow and real-time airflow noise according to time to obtain a real-time pressure sequence, a real-time air flow sequence and a real-time airflow noise sequence;
[0180] Step S520: dividing the real-time pressure sequence, the real-time air flow sequence and the real-time airflow noise sequence by a sliding window method to obtain N′ real-time samples;
[0181] Step S530: using N′ real-time samples as input data of the air leakage detection model, inputting them into an encoder to obtain the output of the encoder, using the output of the encoder through a decoder to obtain a reconstructed sequence of real-time samples, and obtaining the reconstruction loss of the real-time samples through the reconstructed sequence of the real-time samples;
[0182] Step S540: obtaining the label type of the real-time sample according to the output data of the air leakage detection model; if the label type of the real-time sample is abnormal, calculating the ventilation abnormality score of the real-time sample according to the reconstruction loss of the real-time sample and the mean and standard deviation of the reconstruction loss of the normal sample;
[0183] The calculation formula of the ventilation abnormality score is: in, represents the reconstruction loss of real-time samples, x now represents a real-time sample of the input, A reconstructed sequence of real-time samples representing the output of the air leakage detection model;
[0184] This step can help optimize the maintenance plan and operation strategy of the ventilation duct through real-time monitoring and anomaly detection systems. By timely discovering and dealing with small-scale air leakage points, large-scale inspection and maintenance work can be avoided, thus improving the overall operation efficiency of the system.
[0185] Step S600: Establishing an abnormal response mechanism, grading the ventilation abnormality scores, and triggering air leakage alarms of different abnormality levels;
[0186] The specific method of establishing an abnormal response mechanism, grading the ventilation abnormality scores, and triggering air leakage alarms of different abnormality degrees is as follows:
[0187] Step S610: setting a slight abnormality threshold U1, a moderate abnormality threshold U2 and a severe abnormality threshold U3;
[0188] The slight abnormality threshold U1, the moderate abnormality threshold U2 and the severe abnormality threshold U3 are set according to the actual situation and the experience of engineers;
[0189] Step S620: classifying the ventilation abnormality score according to the slight abnormality threshold U1, the moderate abnormality threshold U2 and the severe abnormality threshold U3;
[0190] When W<U1, the ventilation state of the ventilation duct is normal;
[0191] When U1≤W<U2, the ventilation state of the ventilation duct is slightly abnormal;
[0192] When U2≤W<U3, the ventilation state of the ventilation duct is moderately abnormal;
[0193] When W ≥ U3, the ventilation status of the ventilation duct is seriously abnormal;
[0194] Step S630: triggering different air leakage alarms according to the abnormal degree of the ventilation state of the ventilation duct;
[0195] This step not only effectively improves the monitoring and maintenance efficiency of the ventilation system by establishing a degree classification and response mechanism based on the abnormality score, but also takes targeted measures according to the severity of the abnormality, optimizes resource allocation, and enhances system security, demonstrating its innovation and practicality.
[0196] The present application deploys multiple sensor monitoring points at different locations of the ventilation duct. Each monitoring point collects pressure, flow and noise data. When an abnormal ventilation state is detected, the ventilation abnormality scores of each monitoring point are compared. The monitoring point with a higher ventilation abnormality score is likely to be closer to the air leak, thereby achieving fixed-point, quantitative real-time monitoring of the ventilation abnormality location.
[0197] Example 2
[0198] like Figure 2 As shown, the AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm system provided in this application includes:
[0199] The sensor deployment data collection module is used to build a sensor network and use sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline;
[0200] The sample data clustering and division module is used to divide the pressure data, air flow data and airflow noise data into normal samples and abnormal samples through a density-based clustering algorithm;
[0201] The network design model training module is used to design a fully convolutional autoencoder network and train the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model.
[0202] A normal sample standard calculation module is used to input normal samples into the trained air leakage detection model, output the reconstruction loss of the normal samples, and calculate the mean and standard deviation of the reconstruction loss of the normal samples;
[0203] The ventilation anomaly score calculation module is used to collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, and output the label type of the real-time sample. If the label type is abnormal, the ventilation anomaly score is calculated using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample;
[0204] The air leakage degree grading alarm module is used to establish an abnormal response mechanism, grade the ventilation abnormality scores, and trigger air leakage alarms of different abnormal degrees.
[0205] Example 3
[0206] Figure 3 1 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 3 As shown, according to another aspect of the present application, an electronic device is also provided. The electronic device may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are executed by one or more processors, the above-mentioned ventilation duct fixed-point quantitative air leakage monitoring and alarm method based on AI can be executed.
[0207] The method or system according to the embodiment of the present application can also be used by Figure 3 The electronic device architecture shown in FIG. Figure 3As shown, the electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in the electronic device, such as a ROM or a hard disk, may store the AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method provided in the present application. The AI-based fixed-point quantitative air leakage monitoring and alarm method for ventilation ducts may, for example, include: constructing a sensor network, using sensors to monitor and collect pressure data, air flow data and airflow noise data in the duct, dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples through a density-based clustering algorithm, designing a full convolutional autoencoder network, using normal samples and abnormal samples to train the full convolutional autoencoder network to obtain an air leakage detection model, inputting normal samples into the trained air leakage detection model, outputting the reconstruction loss of normal samples, calculating the mean and standard deviation of the reconstruction loss of normal samples, collecting real-time pressure, real-time air flow and real-time airflow noise and processing them to obtain real-time samples as input data for the air leakage detection model, outputting the label type of the real-time sample, and if the label type is abnormal, using the mean and standard deviation of the reconstruction loss of normal samples and the reconstruction loss of the real-time sample to calculate the ventilation abnormality score, establishing an abnormal response mechanism, and grading the degree of ventilation abnormality scores to trigger air leakage alarms of different abnormal degrees. Furthermore, the electronic device may also include a user interface. Of course, Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.
[0208] Example 4
[0209] Figure 4 Schematic diagram of a readable storage medium structure provided by an embodiment of the present application. Figure 4 As shown, it is a readable storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are executed by the processor, the ventilation duct fixed-point quantitative air leakage monitoring and alarm method based on AI according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.
[0210] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart may be implemented as a computer software program. For example, the present application provides a non-temporary machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: constructing a sensor network, using sensors to monitor and collect pressure data, air flow data and airflow noise data in the pipeline, dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples through a density-based clustering algorithm, designing a full convolutional autoencoder network, using normal samples and abnormal samples to train the full convolutional autoencoder network to obtain an air leakage detection model, inputting the normal sample into the trained air leakage detection model, outputting the reconstruction loss of the normal sample, calculating the mean and standard deviation of the reconstruction loss of the normal sample, collecting real-time pressure, real-time air flow and real-time airflow noise and processing them to obtain real-time samples as input data of the air leakage detection model, outputting the label type of the real-time sample, if the label type is abnormal, using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample to calculate the ventilation abnormality score, establishing an abnormal response mechanism, grading the degree of ventilation abnormality scores, and triggering air leakage alarms of different abnormal degrees. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed.
[0211] The methods, apparatuses, and devices of the present application may be implemented in many ways. For example, the methods, apparatuses, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.
[0212] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.
[0213] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method, characterized in that: include: Build a sensor network and use sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline; The pressure data, air flow data and airflow noise data are divided into normal samples and abnormal samples through a density-based clustering algorithm; Design a fully convolutional autoencoder network, and use normal samples and abnormal samples to train the fully convolutional autoencoder network to obtain an air leakage detection model; Design a fully convolutional autoencoder network, wherein the fully convolutional autoencoder network includes: an encoder, a decoder, and an output of the encoder and an output of the decoder; The normal samples and abnormal samples are used as training samples for training the air leakage detection model. The label corresponding to the normal sample is 0, and the label corresponding to the abnormal sample is 1. Input the normal samples into the trained air leakage detection model, output the reconstruction loss of the normal samples, and calculate the mean and standard deviation of the reconstruction loss of the normal samples; Collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, output the label type of the real-time sample, and if the label type is abnormal, use the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample to calculate the ventilation abnormality score; Establish an abnormal response mechanism to grade ventilation abnormality scores and trigger air leakage alarms of different abnormality levels; The normal sample is input into the trained air leakage detection model, and the reconstruction loss of the normal sample is output. The specific method of calculating the mean and standard deviation of the reconstruction loss of the normal sample is: The normal samples in the training samples are input into the trained air leakage detection model, and then input into the encoder to obtain the output of the encoder. The output of the encoder is passed through the decoder to obtain the reconstructed sequence, and the reconstruction loss of the normal samples is calculated based on the reconstructed sequence. Among them, X i represents the normal sample of the i-th input, Represents the reconstructed sequence of the output corresponding to the normal sample of the i-th input; Calculate the mean and standard deviation of the reconstruction loss of normal samples; The calculation formula of the mean value of the reconstruction loss of the normal sample is: Among them, N zc represents the number of normal samples; The calculation formula of the standard deviation of the reconstruction loss of the normal sample is: The real-time pressure, real-time air flow and real-time airflow noise are collected and processed to obtain real-time samples as input data of the air leakage detection model, and the label type of the real-time sample is output. If the label type is abnormal, the specific method of calculating the ventilation abnormality score by using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample is: The real-time pressure, real-time air flow and real-time airflow noise are collected by using the deployed sensor network, and the real-time pressure, real-time air flow and real-time airflow noise are sorted according to time to obtain the real-time pressure sequence, real-time air flow sequence and real-time airflow noise sequence; The real-time pressure sequence, the real-time air flow sequence and the real-time airflow noise sequence are divided into N′ real-time samples by the sliding window method; N′ real-time samples are used as input data of the air leakage detection model, input into an encoder to obtain the output of the encoder, the output of the encoder is passed through a decoder to obtain a reconstructed sequence of real-time samples, and the reconstruction loss of the real-time samples is obtained through the reconstructed sequence of the real-time samples; The label type of the real-time sample is obtained according to the output data of the air leakage detection model. If the label type of the real-time sample is abnormal, the ventilation abnormality score of the real-time sample is calculated according to the reconstruction loss of the real-time sample and the mean and standard deviation of the reconstruction loss of the normal sample; The calculation formula of the ventilation abnormality score is: in, represents the reconstruction loss of real-time samples, x now represents a real-time sample of the input, Reconstructed sequence representing real-time samples of the air leakage detection model output.
2. The AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method according to claim 1 is characterized in that: The specific method of dividing the pressure data, air flow data and airflow noise data into normal samples and abnormal samples by a density-based clustering algorithm is: Normalize the pressure change sequence, air flow change sequence and airflow noise sequence and scale them to the interval [0, 1]; The pressure change sequence, air flow change sequence and airflow noise sequence are divided into N samples by using a sliding window method; Extracting statistics within each window according to the sliding window method, wherein the statistics constitute a feature vector of the window, including: mean, variance, and peak value; The samples corresponding to each window are taken as sample points, and the density-based clustering algorithm is used to cluster the sample points. The types of sample points are automatically identified, and the core points and their boundary points are divided into high-density clusters to obtain normal samples, and the noise points are divided into low-density clusters to obtain abnormal samples.
3. The AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method according to claim 2 is characterized in that: The specific method of extracting the statistics in each window according to the sliding window method and forming the feature vector of the window by the statistics is: Preset the window size α, prediction time step c and sliding step h, and use the sliding window method to get the i-th sample as {(P i , K i , Q i ), y i }, where y is a data label, and the data label is divided into normal ventilation and abnormal air leakage; For each window, calculate the mean, variance, and peak value within the window; The mean value includes the pressure mean value, the air flow mean value and the airflow noise mean value. The calculation formula of the pressure mean value is: Among them, p β represents the βth pressure data in the window, and the calculation formula of the mean air flow rate is: Among them, k β represents the βth air flow data in the window, and the calculation formula of the airflow noise mean is: Among them, q β represents the βth airflow noise data in the window; The variance includes pressure variance, air flow variance and airflow noise variance. The calculation formula of the pressure variance is: The calculation formula of the air flow variance is: The calculation formula of the airflow noise variance is: The peak values include pressure peak value, air flow peak value and airflow noise peak value. The expression of pressure peak value is: p max =max(p1,p2,...,p α ), the expression of the peak air flow rate is: k max =max(k1,k2,...,k α ), the expression of the airflow noise peak is: q max =max(q1,q2,...,q α ); The mean, variance, and peak value in all windows are calculated to form the feature vector of the window.
4. The AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method according to claim 3 is characterized in that: The specific method of taking the samples corresponding to each window as sample points, clustering the sample points using a density-based clustering algorithm, automatically identifying the types of sample points, dividing the core points and their boundary points into high-density clusters to obtain normal samples, and dividing the noise points into low-density clusters to obtain abnormal samples is as follows: The sample corresponding to the statistic of each window is regarded as a sample point a, and all sample point pairs (a i ,a j ) between the Euclidean distance d(a i ,a j ); Preset neighborhood radius eps and minimum number of samples Sam min ; The preset neighborhood radius eps and the minimum number of samples Sam min The specific method is: Define the sample point set A = {a1, a2, ..., a N }, N represents the total number of sample points in the sample point set, and the c clusters obtained by clustering the sample points using the density-based clustering algorithm are defined as C = {C1, C2, ..., C c }; Define the objective function Among them, C ci represents the ci-th cluster obtained by clustering, |C ci | represents the number of sample points in the ci-th cluster, CP(C ci ) represents the compactness of the ci-th cluster, Noise(C) represents the noise ratio of noise points that are not assigned to any cluster to the total number of sample points, and λ is the regularization coefficient that balances the compactness of the cluster and the noise ratio; The compactness of the cluster CP(C ci ) is: The expression of noise ratio Noise(C) is: For the neighborhood radius eps and the minimum number of samples Sam min Perform a grid search, using the current neighborhood radius eps and the minimum number of samples Sam at each step of the search min Run the density-based clustering algorithm to obtain c clusters and calculate the value of the objective function Ω; Choose the neighborhood radius eps and the minimum number of samples Sam that maximize the objective function Ω min As the optimal solution, as the final clustering parameters using density-based clustering algorithm; For each sample point a i , statistical sample point a i The number of samples Sam within the neighborhood radius eps; Compare the sample number Sam with the minimum sample number Sam min If Sam is greater than or equal to the minimum number of samples Sam min , then the current sample point a i Marked as core point a i ′; For the core point a i ′, add all sample points within its neighborhood radius eps to the same cluster. If the sample points within its neighborhood radius contain other core points, the sample points within the neighborhood radius of the core point are also classified into the same cluster; Mark the non-core points within the neighborhood radius of the core point as boundary points, and mark the sample points that do not belong to any cluster as noise points; Until all sample points are divided, the sample points divided into the same cluster are marked as normal samples, and the cluster is a high-density cluster. The noise points that do not belong to the high-density cluster are marked as abnormal samples and divided into a low-density cluster.
5. The AI-based ventilation duct fixed-point quantitative air leakage monitoring and alarm method according to claim 4 is characterized in that: The specific method of designing a fully convolutional autoencoder network and training the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model is: Design a loss function Loss, wherein the loss function includes reconstruction loss and classification loss, wherein the reconstruction loss selects smooth L1 loss, and the classification loss selects binary cross entropy loss; The functional expression of the reconstruction loss is: Among them, x is the output of the encoder, and y represents the true label of the training sample; The functional expression of the classification loss is: Among them, y i represents the true label of the i-th training sample, It indicates the probability of predicting that the i-th training sample is an abnormal sample. express The logarithm of express The logarithm of The functional expression of the loss function Loss is: Among them, ε is the trade-off coefficient; All training samples are divided into n batches. For the training samples in the mth batch, they are input into the encoder as input data to obtain the output of the encoder. The output of the encoder is passed through the decoder to obtain the reconstructed sequence. Calculate the reconstruction loss and classification loss to get the value of the loss function Loss, and calculate the loss function of the batch for the model parameter θ m The gradient value Use the optimizer to update the model parameters to get θ m+1 , the model parameters θ m+1 The calculation formula is: θ m+1 =θ m -η×▽Loss(θ m ), where θ m is the model parameter of the current batch, θ m+1 is the updated model parameter, η is the learning rate; A maximum number of training rounds L is preset. When the number of training rounds reaches the maximum number of training rounds, the training is stopped to obtain a trained air leakage detection model.
6. A ventilation duct fixed-point quantitative air leakage monitoring and alarm system based on AI, which is implemented based on the ventilation duct fixed-point quantitative air leakage monitoring and alarm method based on AI as described in any one of claims 1 to 5, characterized in that: include: The sensor deployment data collection module is used to build a sensor network and use sensors to monitor and collect pressure data, air flow data, and airflow noise data in the pipeline; The sample data clustering and division module is used to divide the pressure data, air flow data and airflow noise data into normal samples and abnormal samples through a density-based clustering algorithm; The network design model training module is used to design a fully convolutional autoencoder network and train the fully convolutional autoencoder network using normal samples and abnormal samples to obtain an air leakage detection model. A normal sample standard calculation module is used to input normal samples into the trained air leakage detection model, output the reconstruction loss of the normal samples, and calculate the mean and standard deviation of the reconstruction loss of the normal samples; The ventilation anomaly score calculation module is used to collect real-time pressure, real-time air flow and real-time airflow noise and process them to obtain real-time samples as input data of the air leakage detection model, and output the label type of the real-time sample. If the label type is abnormal, the ventilation anomaly score is calculated using the mean and standard deviation of the reconstruction loss of the normal sample and the reconstruction loss of the real-time sample; The air leakage degree grading alarm module is used to establish an abnormal response mechanism, grade the ventilation abnormality scores, and trigger air leakage alarms of different abnormal degrees.
7. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the AI-based ventilation duct fixed-point and quantitative air leakage monitoring and alarm method as described in any one of claims 1 to 5 are implemented.
8. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor to execute the steps in the AI-based ventilation duct fixed-point and quantitative air leakage monitoring and alarm method as described in any one of claims 1 to 5.
