Ventilation and air conditioning fan fault early warning method and system based on multi-source data

Abstract

The application provides a kind of based on multi-source data's ventilation air conditioning fan fault early warning method and system, it is related to the online fan fault early warning technical field of ventilation air conditioning system.The method comprises: obtaining the multi-source data of ventilation air conditioning fan;The multi-source data is preprocessed;Dynamic working condition clustering is carried out to the multi-source data after preprocessing using improved DBSCAN clustering model, if the clustering result is that multi-source data belongs to outlier, then carry out fault early warning;If the clustering result is that multi-source data belongs to normal working condition cluster, then carry out local sparse type anomaly detection to multi-source data using isolated forest model;If the detection result is that there is abnormal point in normal working condition cluster, then further judge to abnormal point to obtain final abnormal point by screening;If there is still abnormal point in normal working condition cluster after screening processing, then carry out fault early warning.The application can accurately carry out fault early warning to the online fan of ventilation air conditioning system without manual parameter adjustment or historical fault label.

Description

Technical Field

[0001] This invention relates to the field of online fan fault early warning technology for ventilation and air conditioning systems, and in particular to a method and system for early warning of ventilation and air conditioning fan faults based on multi-source data. Background Technology

[0002] For online fan fault early warning in ventilation and air conditioning systems, existing technologies can be mainly divided into two categories: one is a rule-based method based on thresholds or statistical tests, which first sets fixed upper and lower limits for single or a small number of sensor signals (such as air volume, vibration peak or bearing temperature), and triggers an alarm when the real-time measured value exceeds the warning threshold; the other is to use a combination of dimensionality reduction and clustering or machine learning models, such as first performing principal component analysis or wavelet decomposition on high-dimensional vibration and temperature features, and then using K-means, DBSCAN or support vector machine for operating condition classification and anomaly detection.

[0003] While these methods can detect obvious anomalies to some extent, they still have the following shortcomings:

[0004] 1) Thresholds and parameters are not adaptive: Fixed upper and lower limits cannot take into account the normal fluctuations of air volume, supply air temperature and operating frequency under daily adjustment or system dynamic load, resulting in frequent false alarms; 2) Limitations of single detection model: Simply using density or reconstruction error methods can only capture one type of abnormality in the overall offset or local mutation, making it difficult to take into account both, and it is difficult to control missed alarms and false alarms at the same time.

[0005] Therefore, there is an urgent need for an online unsupervised fault early warning method that is adaptable to multiple operating conditions. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for early warning of ventilation and air conditioning fan failures based on multi-source data.

[0007] According to one aspect of the present invention, a method for early warning of ventilation and air conditioning fan failure based on multi-source data is proposed, the method comprising:

[0008] Acquire multi-source data from ventilation and air conditioning fans;

[0009] The multi-source data is preprocessed;

[0010] An improved DBSCAN clustering model is used to perform dynamic operating condition clustering on preprocessed multi-source data to obtain clustering results; the clustering results include whether the multi-source data belongs to the normal operating condition cluster or to outliers; if the clustering results indicate that the multi-source data belongs to outliers, a fault warning is issued.

[0011] If the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, then the isolated forest model is used to perform local sparse anomaly detection on the multi-source data to obtain the detection result; the detection result includes whether there are anomalies or no anomalies in the normal operating condition cluster.

[0012] If the detection result indicates that there are abnormal points within the normal operating condition cluster, the abnormal points are further judged to filter and obtain the final abnormal points; if there are still abnormal points within the normal operating condition cluster after the filtering process, a fault warning is issued.

[0013] Furthermore, the multi-source data includes fan operating parameters, fan vibration time-domain characteristics, fan vibration spectrum characteristics, and temperature parameters; wherein, fan operating parameters include static pressure, air volume, operating frequency, and motor input current; fan vibration time-domain characteristics include acceleration RMS, peak acceleration, acceleration kurtosis, velocity RMS, peak velocity, velocity kurtosis, and displacement peak-to-peak value; fan vibration spectrum characteristics include acceleration spectrum and velocity spectrum; and temperature parameters include bearing housing temperature, motor winding temperature, and ambient temperature.

[0014] Furthermore, the preprocessing includes: data resampling to transform irregular time interval data into time series data with the same interval; outlier detection and removal; missing value imputation; time window convergence; standardization; and PCA dimensionality reduction.

[0015] Furthermore, the aggregation rules in the time window convergence are as follows: for the fan operating parameters, calculate the average value and standard deviation, and extract the maximum and minimum values; for the fan vibration time domain characteristics, calculate the average value and extract the maximum value; for the fan vibration spectrum characteristics, extract the maximum energy of the bearing fault characteristic frequency point; for the temperature parameters, calculate the average temperature and extract the highest temperature.

[0016] Furthermore, the step of using the improved DBSCAN clustering model to perform dynamic working condition clustering on the preprocessed multi-source data and obtaining the clustering results includes:

[0017] Adaptively determine the two parameters, neighborhood radius and minimum neighbor number, in the DBSCAN clustering model;

[0018] DBSCAN clustering is performed on the preprocessed multi-source data based on the determined neighborhood radius and minimum number of neighbor points; among them, the density connection point set, i.e. the boundary point set, obtained by clustering is physically reassigned to allocate all density connection points to the normal operating condition cluster.

[0019] Furthermore, the adaptive determination of the two parameters, neighborhood radius and minimum neighbor number, in the DBSCAN clustering model includes: using preprocessed multi-source data from multiple sets of ventilation and air conditioning fans as the training set, and representing the low-dimensional feature matrix in the training set after PCA dimensionality reduction as follows: Where m is the number of windows and k is the feature dimension, calculate the minimum number of neighbor points. ;right Each sample in Calculate the Euclidean distance from the minPts nearest neighbor, denoted as . ; All Ascending order as sequence Construct gradient sequence , ; Calculate the total gradient Find the minimum index satisfy: ; set the minimum index Corresponding Euclidean distance Set as neighborhood radius .

[0020] Furthermore, the physical spatial reallocation of the density connection point set, i.e., the boundary point set, obtained from clustering, to allocate all density connection points to the normal operating condition cluster includes:

[0021] Filter fuzzy connection points in the density connection point set; the fuzzy connection point is a point in the density connection point set that belongs to the intersection of the neighborhood of the core point of multiple normal operating condition clusters; define all normal operating condition clusters to which a certain fuzzy boundary point belongs as the candidate clusters to which the fuzzy boundary point belongs;

[0022] For each fuzzy connection point, find the row vector corresponding to the fuzzy connection point in the windowed feature matrix obtained after time window convergence, and use the row vector as the original time window statistical vector corresponding to the fuzzy connection point.

[0023] For each fuzzy connection point, the median vector of the original time window statistical vector of each candidate cluster is determined in the windowed feature matrix obtained after time window convergence.

[0024] For each fuzzy connection point, calculate the Euclidean distance between its corresponding original time window statistical vector and the median vector; select the candidate cluster with the smallest Euclidean distance as the cluster to which the fuzzy connection point belongs;

[0025] Remove each fuzzy connection point from the density connection point set and add it to the corresponding family in the normal operating condition cluster;

[0026] For other density ties that are not fuzzy ties, each has only one affiliation cluster. These ties are removed from the density ties set and added to the corresponding affiliation cluster in the normal operating condition cluster.

[0027] Furthermore, if the detection result indicates the presence of anomalies within the normal operating condition cluster, further judgment of the anomalies includes:

[0028] For outliers within a normal operating condition cluster, compare their corresponding outlier score calculated using the isolated forest model with the cluster threshold: if the outlier score is greater than the cluster threshold, then retain the outlier; wherein, the cluster threshold is calculated as follows:

[0029] For each normal operating condition cluster, the intra-cluster threshold is calculated using the following formula. :

[0030] ;

[0031] In the formula, Indicates adaptive confidence level. , Indicates the sample size within a cluster; The empirical quantiles, representing the empirical cumulative distribution function, are determined as follows: Find the quantiles that satisfy the empirical cumulative distribution function. If the abnormal score value s is found, then the search results will be... ,in,

[0032] ;

[0033] In the formula, This represents the j-th normal operating condition cluster. This represents the anomaly score of the k-th sample calculated using the isolated forest model; Let be the indicator function, which is defined as:

[0034] .

[0035] Furthermore, if the detection result indicates the presence of anomalies within the normal operating condition cluster, further judgment of the anomalies includes: verifying the physical deviation of the boundary points of the anomalies to filter out falsely reported anomalies, specifically including:

[0036] For anomalies within the normal operating condition cluster, determine whether they belong to the density connection point set; if they do, extract the row vector corresponding to the anomaly in the windowed feature matrix obtained after time window convergence as the original time window statistical vector corresponding to the anomaly.

[0037] Based on the original time window statistical vector corresponding to the outlier and the median vector of the original time window statistical vector of each cluster corresponding to the normal operating condition cluster, the deviation is calculated. :

[0038] ;

[0039] In the formula, , Indicates outliers The corresponding original time window statistical vector; This represents the median vector of the original time window statistics vectors corresponding to each cluster (i) in the normal operating condition cluster; To avoid small quantities where the denominator is zero; Represents the L2 norm;

[0040] If deviation If the value is less than the preset threshold, the anomaly is determined to be a false alarm and deleted; otherwise, it is retained.

[0041] According to another aspect of the present invention, a ventilation and air conditioning fan fault early warning system based on multi-source data is proposed. The system is used to implement the aforementioned ventilation and air conditioning fan fault early warning method based on multi-source data. The system includes:

[0042] The data acquisition module is configured to acquire multi-source data from ventilation and air conditioning fans;

[0043] A data preprocessing module configured to preprocess the multi-source data;

[0044] The dynamic clustering module is configured to use an improved DBSCAN clustering model to perform dynamic operating condition clustering on preprocessed multi-source data and obtain clustering results; the clustering results include whether the multi-source data belongs to the normal operating condition cluster or to outliers.

[0045] An anomaly detection module is configured to perform local sparse anomaly detection on the multi-source data using an isolated forest model if the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, and obtain the detection result; the detection result includes whether there are anomalies or not within the normal operating condition cluster.

[0046] An anomaly filtering module is configured to further judge the anomaly if the detection result indicates that there is an anomaly in the normal operating condition cluster, so as to filter and obtain the final anomaly.

[0047] The fault warning module is configured to issue a fault warning when two situations exist: if the clustering result indicates that the multi-source data belongs to an outlier, then a fault warning is issued; if there are still outliers in the normal operating condition cluster after screening, then a fault warning is issued.

[0048] The embodiments of the present invention have the following technical effects:

[0049] This invention proposes a method and system for early warning of ventilation and air conditioning fan failures based on multi-source data. Compared with the commonly used "preset cluster number + single anomaly detection" scheme in the prior art, the "condition-adaptive dynamic clustering + dual anomaly detection" technology adopted in this invention brings significant performance improvement. First, traditional methods often lead to distorted condition classification due to fixed cluster numbers or improper manual parameter adjustment, affecting anomaly identification. This invention adaptively determines two parameters of the DBSCAN clustering algorithm—neighborhood radius and minimum neighbor number—to closely follow the real data distribution and automatically classify diverse operating conditions into high-confidence normal clusters, thus enabling subsequent anomaly detection to be based on accurately aligned operating conditions. Second, this invention introduces a physical space redistribution mechanism for density connection points in the clustering stage. By backtracking the distance between the original time window statistics and the cluster median vector, boundary points that should belong to a certain normal operating condition but were not identified due to density threshold boundaries are accurately assigned to the correct clusters, which not only significantly improves the purity of normal clusters. This provides a more consistent and physical input for subsequent anomaly detection. Furthermore, the globally unified threshold commonly used in existing technologies struggles to account for fluctuations under different operating conditions, resulting in a high false positive rate. This invention performs local sparse detection in an isolated forest trained solely on high-confidence normal cluster samples, combined with global density outlier capture, ensuring that neither sudden faults with overall shifts are missed, nor early signals of local parameter mutations are unearthed. Compared to the false negatives and missed positives easily generated by a single model, this complementary mechanism fundamentally improves the recall and accuracy of diagnosis. Specifically, through operating condition adaptive thresholds based on empirical distribution and physical deviation verification, this invention can actively filter false anomalies in normal transition states such as startup and shutdown, further suppressing false positives. Attached Figure Description

[0050] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a flowchart of a ventilation and air conditioning fan fault early warning method based on multi-source data, as described in an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of a ventilation and air conditioning fan fault early warning system based on multi-source data, as described in an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are part of this invention.

[0054] This invention proposes a method and system for early warning of ventilation and air conditioning fan faults based on multi-source data. The core of this method lies in its adaptive dynamic clustering approach based on multi-source time-window aggregation features. This method automatically identifies normal operating clusters and outliers of the fan using unsupervised density analysis, without requiring pre-defined cluster numbers or reliance on historical fault labels. By adaptively determining the neighborhood radius and minimum number of points for DBSCAN, clustering parameters can be flexibly adjusted to achieve accurate segmentation under multiple operating conditions, providing a solid data foundation for subsequent anomaly detection. This significantly reduces the cost of manual parameter tuning and labeling, improving the universality and reliability of fault warnings. Furthermore, this invention introduces a dual anomaly detection mechanism: first, adaptive DBSCAN captures density anomalies caused by overall shift; then, local mutation anomalies are identified in an isolated forest trained only by high-confidence normal cluster samples. Combined with operating condition adaptive thresholds and physical deviation verification based on empirical distribution, this approach balances overall shift and local mutations, effectively improving fault recall and false alarm suppression capabilities. It overcomes the limitations of single algorithms for different anomaly types, ensuring that unsupervised diagnosis is both comprehensive and accurate.

[0055] This invention proposes a method for early warning of ventilation and air conditioning fan failures based on multi-source data, such as... Figure 1 As shown, the method includes:

[0056] S1. Acquire multi-source data of ventilation and air conditioning fans;

[0057] S2. Preprocess the multi-source data;

[0058] S3. Use the improved DBSCAN clustering model to perform dynamic working condition clustering on the preprocessed multi-source data and obtain the clustering results; the clustering results include whether the multi-source data belongs to the normal working condition cluster or to the outlier; if the clustering result is that the multi-source data belongs to the outlier, then a fault warning is issued;

[0059] S4. If the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, then the isolated forest model is used to perform local sparse anomaly detection on the multi-source data to obtain the detection result; the detection result includes whether there are anomalies or no anomalies in the normal operating condition cluster.

[0060] S5. If the detection result indicates that there are abnormal points within the normal operating condition cluster, the abnormal points are further judged to filter and obtain the final abnormal points; if there are still abnormal points within the normal operating condition cluster after the filtering process, a fault warning is issued.

[0061] The method begins with S1, acquiring multi-source data on ventilation and air conditioning fans.

[0062] According to an embodiment of the present invention, the multi-source data of the ventilation and air conditioning fan includes: fan operating parameters, vibration time-domain characteristics, vibration spectrum characteristics, temperature parameters, etc. Among them, the fan operating parameters include static pressure. Air volume Operating frequency Motor input current Vibration time-domain characteristics include RMS acceleration, peak acceleration, kurtosis acceleration, RMS velocity, peak velocity, kurtosis velocity, and peak-to-peak displacement. Vibration spectral characteristics include acceleration spectrum (total energy of all frequency bands, and energy at characteristic frequency points of bearing faults) and velocity spectrum (total energy of all frequency bands, and energy at characteristic frequency points of rotational speed). Temperature parameters include bearing housing temperature. Motor winding temperature Ambient temperature .

[0063] In S2, the multi-source data is preprocessed.

[0064] According to embodiments of the present invention, the original multi-source, high-dimensional, and noisy operational data is transformed into low-dimensional, robust, and comparable feature samples through a process of "time window convergence + standardization + PCA dimensionality reduction". This process aligns the time-series information of different channels, suppresses occasional shocks, unifies the dimensional scale, and removes redundant correlations.

[0065] The original data may be obtained from different sources, and the time intervals during data collection from different sources may be inconsistent, resulting in different time intervals for the data obtained from each source. In irregularly sampled data, it is necessary to transform data with different time intervals into data with the same time interval (e.g., 15 minutes). Therefore, firstly, the original multi-source data is resampled to transform the irregular time interval data into time series data with the same interval. The following methods can be used: 1) Forward imputation: filling missing values ​​with the values ​​of the previous time point; 2) Linear interpolation: interpolating missing values ​​according to the changing trend between time points to obtain a smooth time series; 3) Mean interpolation: filling missing values ​​according to the average value of adjacent data points.

[0066] Then, after data resampling, outlier detection and removal are performed on the resampling data. For example, the quartile rule, the Raida criterion, etc., can be used.

[0067] Then, missing values ​​resulting from outlier removal or those present in the original data are imputed. As an example, the moving average method can be used for imputation. The moving average method imputes missing values ​​by calculating the average of the data over a rolling time window before and after the missing value. Its effectiveness depends on the choice of window size. Smaller windows are sensitive to data changes but are easily affected by noise; larger windows help smooth the data but may delay information capture. The window size is determined based on the rate of data change. Cross-validation with simulated missing values ​​is used to evaluate the imputation effect under different window sizes. Specifically, a high-quality continuous variable time series segment is selected as a "reference complete dataset." Missing values ​​are manually and randomly introduced into the reference dataset, and the imputation effect (RMSE, CV) of the moving average method with different window sizes is evaluated to determine the optimal window size for that variable. This optimal window size is then applied to the imputation of the entire dataset.

[0068] Then, time window aggregation is performed. Time window aggregation aggregates instantaneous records at the same time interval according to a sliding window of fixed length, using various statistical measures to characterize the overall operating status within that time period. This step is used to smooth noise and single-point fluctuations and extract persistent trends and extreme features, making each sample more comprehensively reflect the working conditions during that period. Specifically, it includes: 1) Window settings: 11) Window length settings: The more samples in the window (i.e., the longer the window), the more stable the statistics (including mean, standard deviation, etc.) and the less susceptible they are to fluctuations at individual points; the longer the window, the more the algorithm must wait for the complete window data to produce results, and the slower the fault warning will be. The optimal window can be determined based on the time scale of the fault characteristics. For example, for fast-changing faults (such as valve blockage or impeller impact caused by foreign object inhalation), there are often obvious abnormalities within minutes or ten minutes, and a short window (30min–1h) is more sensitive. For slow-changing faults (such as bearing lubricant deterioration or winding insulation degradation causing a slow temperature rise), it takes hours or even longer to show the deviation, and a long window (2h–4h) is more helpful in smoothing random noise and highlighting the gradual signal. 12) Sliding step size: 15 minutes (a new window is generated for every record); 2) Aggregation rules: Based on the physical meaning of fault warning, feature information entropy and algorithm efficiency, the following aggregation rules are formed: a) Static pressure / air volume / operating frequency / current: calculate the average value to reflect continuous operating conditions; calculate the standard deviation to evaluate operating stability; extract the maximum and minimum values ​​to capture extreme operating conditions; b) Vibration time domain characteristics: Acceleration RMS: calculate the average value (continuous intensity) + extract the maximum value (transient anomaly); Acceleration peak value: extract the maximum value (maximum impact intensity); Acceleration kurtosis: extract the maximum value (most severe impact event); Velocity RMS / peak value / kurtosis: same as acceleration processing rules; c) Vibration spectrum characteristics: Acceleration spectrum: extract the maximum energy of bearing fault characteristic frequency points; d) Temperature parameters: calculate the average temperature (overall temperature rise trend); extract the highest temperature (local overheating risk).

[0069] Output a windowed feature matrix through time-window convergence operation: Where m is the number of windows, This represents the total number of features extracted from each variable according to the aggregation rules above.

[0070] Then, Z-score standardization is performed to eliminate differences in the dimensions and magnitudes of different physical quantities, ensuring that subsequent distance- or variance-based algorithms are not dominated by certain large numerical features and can fairly utilize information from each dimension. Specifically, for the windowed feature matrix... Calculate the mean for each column j. and standard deviation and apply Obtain the standardized matrix .

[0071] Then, PCA dimensionality reduction is performed to remove redundant correlations and construct a low-dimensional, denoised feature space. Specifically, for the standardized matrix... Calculate the covariance matrix: For the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues. With feature vectors Determine the minimum principal component number k that satisfies Projected to the front Principal components: ,get Output the reduced-dimensional feature matrix. This serves as the input for subsequent clustering and anomaly detection.

[0072] In the above preprocessing process, a sample index mapping is established, that is, each row of data in any intermediate matrix (resampling → windowing → standardization → PCA dimensionality reduction) has the same global sample ID (which can be an integer index or a timestamp + window position).

[0073] As an example, one could aggregate the original 15-minute records into a 2-hour sliding window, calculate a set of statistical characteristics (including mean, standard deviation, maximum value, etc.), and form a matrix. Each row represents a window (feature sample vector). After PCA dimensionality reduction, these Projecting the 3D window features onto k-dimensional space yields... Each row is a vector of the window in the PCA space, denoted as . In the sample index mapping, it can be based on row index in (or global sample ID) directly locate The OK.

[0074] In S3, the improved DBSCAN clustering model is used to perform dynamic working condition clustering on the preprocessed multi-source data to obtain clustering results; the clustering results include whether the multi-source data belongs to the normal working condition cluster or to outliers; if the clustering results indicate that the multi-source data belongs to outliers, a fault warning is issued.

[0075] According to an embodiment of the present invention, after obtaining the low-dimensional feature matrix Then, this step uses adaptive density clustering to divide the feature samples into normal operating condition clusters and outliers.

[0076] The two parameters of the DBSCAN clustering algorithm—neighborhood radius and minimum neighbor count—need to be adaptively determined without manual parameter tuning. This step considers both the feature dimension k and the sample size m, ensuring stable operation in both small clusters and large sample sizes. The improved DBSCAN clustering algorithm adaptively determines the neighborhood radius and minimum neighbor count using the following process:

[0077] For the low-dimensional feature matrix after PCA dimensionality reduction Where m is the number of windows and k is the feature dimension, calculate the minimum number of neighbor points. ;right Each sample in Calculate its up to the th The Euclidean distance between the nearest neighbors is denoted as . ; All Ascending order as sequence Construct gradient sequence , ; Calculate the total gradient Find the minimum index satisfy: ; set the minimum index Corresponding Euclidean distance Set as neighborhood radius .

[0078] An improved DBSCAN clustering algorithm with the adaptively determined neighborhood radius and minimum number of neighbor points is used to cluster the low-dimensional feature matrix after PCA dimensionality reduction. The clustering result includes: 1) a normal operating condition cluster containing several core points and their neighboring points; 2) an outlier set O (i.e., noise points); and 3) a density-connected point set B (i.e., boundary point set). A core point is defined as: given a neighborhood radius... Given the minimum number of points minPts, if a sample point is in Within the space A point is considered a core point if its neighborhood (including itself) contains at least minPts samples; density connection point set That is, the boundary point set is defined as: if a certain sample point It does not satisfy the core point condition, i.e., the number of its neighborhood points is less than the minimum number of points minPts, but it has at least one core point. Make Then it is called These are density connection points, also known as boundary points. Considered to be located within the neighborhood of at least one core point of a cluster; outlier set Outliers are defined as: a sample point that is neither a core point nor satisfies the density connectivity condition (i.e., it is not at any core point). If a sample point is not within its neighborhood, it is called an outlier, or a noise point.

[0079] During DBSCAN clustering, the following two situations can lead to information loss or ambiguous judgments: a) Multi-cluster crossover: A density connection point falls into multiple clusters simultaneously. Within the neighborhood, it's unclear which cluster it's closer to; b) Unassigned: If all density connection points are indiscriminately assigned to a certain cluster, their true physical spatial affiliation may be ignored. Therefore, this embodiment performs physical spatial reallocation of the density connection point set, assigning all density connection points to the normal operating condition cluster. The density connection point set is processed according to the following procedure. Perform fine-grained allocation.

[0080] S31. Identify the fuzzy connections in the density connection point set and their candidate clusters;

[0081] For each density connection point First, determine its location. The neighborhood of the core point in the space belongs to which clusters: For each cluster in the normal operating condition cluster, find the core point. Make If density connection point If it belongs to only one cluster, it will be directly assigned to that cluster without further processing.

[0082] If it belongs to multiple clusters, its affiliation needs to be further determined, i.e., proceed to the next step; when a certain density connection point belongs to multiple clusters, the density connection point is defined as a fuzzy connection point, and the multiple clusters to which it belongs are defined as candidate clusters; that is, the fuzzy connection point is the point in the density connection point set that belongs to the intersection of the neighborhood of the core point of multiple normal operating condition clusters; all normal operating condition clusters in which a certain fuzzy boundary point is located are defined as candidate clusters for the affiliation of the fuzzy boundary point.

[0083] S32. For each fuzzy connection point, the windowed feature matrix obtained after time-window convergence. Find the row vector corresponding to the fuzzy connection point in the data, and use the features of this row vector as the original time window statistical vector corresponding to the fuzzy connection point. ;

[0084] S33. For each fuzzy connection point, determine the median vector of the original time window statistical vector of each candidate cluster in the windowed feature matrix obtained after time window convergence.

[0085] The median vector is calculated using the following formula:

[0086] ;

[0087] in, Let represent the median vector of the j-th candidate cluster, and d represent the dimension index of the feature. , which also represents the d-th eigencomponent of the median vector; median The dot (.) represents the median operator, which calculates the median of a given set of data. Representing feature samples In the windowed feature matrix The corresponding original time window statistical vector; Representing vectors The specific value of the dth characteristic component.

[0088] S34. For each fuzzy connection point, calculate the Euclidean distance between its corresponding original time window statistical vector and the median vector of the original time window statistical vector of each corresponding candidate cluster; select the candidate cluster with the smallest Euclidean distance as the cluster to which the fuzzy connection point belongs.

[0089] S35. Remove each fuzzy connection point from the density connection point set and add it to the corresponding family in the normal operating condition family.

[0090] After the above steps, all density connection points are clearly assigned, preventing information loss due to multiple clusters covering or not being assigned. The final output is the normal operating condition cluster set. The set includes both core points and redistributed density-connected points; outlier set. Only retain noise points.

[0091] Furthermore, a sample index is established during the clustering process. and normal operating condition cluster index Mapping clusters between Cluster(i) is a label mapping function that converts the sample number into the cluster number to which the sample belongs. After clustering, each sample i is assigned to a cluster, such as cluster 1, cluster 2, ..., cluster M; or, if it is an outlier, it does not belong to any cluster. Cluster(i) is defined as: cluster(i) = j; that is, sample i... Classified into operating condition cluster C j In the middle. For example, if the data corresponding to sample number i=57 is classified as a member of "cluster 3" after clustering, then cluster(57)=3; if If a cluster belongs to the outlier set O, then "cluster(i)=0" can be defined or no cluster number can be returned to distinguish it from the normal cluster.

[0092] In S4, if the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, then the isolated forest model is used to perform local sparse anomaly detection on the multi-source data to obtain the detection result; the detection result includes whether there are anomalies or no anomalies in the normal operating condition cluster.

[0093] According to an embodiment of the present invention, if the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, then the isolated forest model is used to perform local sparse anomaly detection on the multi-source data to obtain the detection result.

[0094] First, global density-based anomaly detection is performed. The DBSCAN algorithm is used to quickly capture anomalies that deviate from the overall density, i.e., all outliers that have not been assigned to any cluster. It can be used directly as an anomaly point for fault warning.

[0095] Secondly, the isolated forest algorithm is used to perform local sparse anomaly detection on the normal operating condition cluster, identifying multiple anomaly points within the cluster. The construction of the isolated forest model includes: determining the number of trees. For sample sizes m in the thousands, t=100; if the sample size is tens of thousands, t can be appropriately increased to 150-200; constructing a system containing... Isolated forest model of trees; using the isolated forest model to analyze the low-dimensional feature matrix corresponding to the normal operating condition cluster. Samples in Calculate anomaly scores .

[0096] In S5, if the detection results output by the isolated forest model indicate that there are outliers in the normal operating condition cluster, the outliers are further judged to filter and obtain the final outliers; if there are still outliers in the normal operating condition cluster after filtering, a fault warning is issued.

[0097] Specifically, this includes: S51, for each normal operating condition cluster, the cluster threshold is calculated according to the following formula. :

[0098] ;

[0099] In the formula, Indicates adaptive confidence level. , This represents the sample size within a cluster. When the number of samples in a cluster is small, the confidence level is lowered to avoid the quantiles being too aggressive due to the small sample size. 0.05 is an empirical coefficient that can be used as an adjustable parameter; 0.98 is the global upper limit to avoid the threshold being too high. The empirical quantiles representing the empirical cumulative distribution function (ECDF) are determined as follows: Find the quantiles that satisfy the empirical cumulative distribution function. The minimum anomaly score s is then found. ,in Indicates in cluster The percentage of samples with a score no greater than s (empirical distribution):

[0100] ;

[0101] In the formula, This represents the j-th normal operating condition cluster. This represents the result calculated using the Isolation Forest algorithm. Abnormal scores for each sample; Let be the indicator function, which is defined as:

[0102] .

[0103] For example, if the intra-cluster score is [0.1, 0.2, 0.5, 0.9], then (Because there are 3 values ​​≤ 0.5); Conversely .

[0104] S52. For multiple outliers within a normal operating condition cluster, compare the outlier score calculated using the isolated forest algorithm with the cluster threshold. The value; anomaly scores greater than the intra-cluster threshold. Selected feature samples are retained; otherwise, they are deleted.

[0105] Furthermore, for multiple anomalies within the normal operating condition cluster, boundary point physical deviation verification is performed to filter out false alarm anomalies; false alarms caused by normal transition states such as start-up / shutdown are deleted to ensure that the retained local anomalies have sufficient physical deviation. Specifically, this includes:

[0106] S53. For multiple outliers within the normal operating condition cluster, determine whether they belong to the density connection point set; if so, extract the row vector corresponding to the outlier in the windowed feature matrix obtained after time window convergence as the original time window statistical vector corresponding to the outlier. ;

[0107] S54. Based on the original time window statistical vector corresponding to the anomaly point and the median vector of the original time window statistical vector of each cluster corresponding to the normal operating condition cluster, calculate the deviation. :

[0108] ;

[0109] In the formula, , This represents the original time window statistical vector corresponding to the outlier; This represents the median vector of the original time window statistical vectors corresponding to each cluster (i) in the normal operating condition cluster; ε is a small quantity used to avoid the denominator being zero, for example, it can take the value 1e-6, or be set according to the data volume.

[0110] S55, if deviation If the value is less than the preset threshold, the anomaly is determined to be a false alarm and deleted; otherwise, it is retained. The preset threshold can be 0.1.

[0111] In the above steps, training is only performed on high-confidence normal cluster samples to avoid contamination by anomalous samples; the quantile method based on ECDF with an upper limit can take into account the fluctuation characteristics of small and large clusters; deviation filtering can eliminate false alarms in normal transition states; combining global density and local sparsity can improve detection recall and accuracy. Steps S4 and S5 use a two-layer mechanism of "density anomaly capture + isolated forest detection" to complementarily cover different types of anomalies, and finally fuse the results to ensure high recall and low false alarms.

[0112] Once multi-source data is confirmed to be abnormal, an early warning message with engineering significance is generated. The early warning message may include, for example, the start and end times of the abnormal time window.

[0113] Another embodiment of the present invention proposes a ventilation and air conditioning fan fault early warning system based on multi-source data. The system is used to implement the aforementioned ventilation and air conditioning fan fault early warning method based on multi-source data, such as... Figure 2 As shown, the system includes:

[0114] Data acquisition module 210 is configured to acquire multi-source data from ventilation and air conditioning fans;

[0115] Data preprocessing module 220 is configured to preprocess the multi-source data;

[0116] The dynamic clustering module 230 is configured to use an improved DBSCAN clustering model to perform dynamic operating condition clustering on the preprocessed multi-source data and obtain clustering results; the clustering results include whether the multi-source data belongs to the normal operating condition cluster or to outliers.

[0117] The anomaly detection module 240 is configured to perform local sparse anomaly detection on the multi-source data using an isolated forest model if the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, and obtain the detection result; the detection result includes whether there are anomalies or no anomalies in the normal operating condition cluster.

[0118] Anomaly filtering module 250 is configured to further judge the anomaly if the detection result indicates that there is an anomaly in the normal operating condition cluster, so as to filter and obtain the final anomaly.

[0119] The fault warning module 260 is configured to issue a fault warning when two situations exist: if the clustering result indicates that the multi-source data belongs to outliers, then a fault warning is issued; if there are still outliers in the normal operating condition cluster after screening, then a fault warning is issued.

[0120] The function of the ventilation and air conditioning fan fault early warning system based on multi-source data described in this embodiment of the invention can be explained by the aforementioned ventilation and air conditioning fan fault early warning method based on multi-source data. Therefore, for the parts not described in detail in the system embodiment, please refer to the above method embodiment, and they will not be repeated here.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of ventilation and air conditioning fan failures based on multi-source data, characterized in that, include: Acquire multi-source data from ventilation and air conditioning fans; The multi-source data is preprocessed; An improved DBSCAN clustering model was used to perform dynamic working condition clustering on preprocessed multi-source data to obtain clustering results. The clustering results include whether the multi-source data belongs to the normal operating condition cluster or to outliers; If the clustering result indicates that the multi-source data belongs to outliers, then a fault warning is issued; The method of using an improved DBSCAN clustering model to perform dynamic working condition clustering on preprocessed multi-source data and obtaining clustering results includes: Adaptively determining the neighborhood radius and minimum neighbor number in the DBSCAN clustering model includes: using preprocessed multi-source data from multiple sets of ventilation and air conditioning fans as the training set; the low-dimensional feature matrix in the training set after PCA dimensionality reduction is represented as: Where m is the number of windows and k is the feature dimension, calculate the minimum number of neighbor points. ;right Each sample in Calculate the Euclidean distance from the minPts nearest neighbor, denoted as . ; All Ascending order as sequence Construct gradient sequence , ; Calculate the total gradient Find the minimum index satisfy: ; set the minimum index Corresponding Euclidean distance Set as neighborhood radius ; DBSCAN clustering is performed on the preprocessed multi-source data based on a determined neighborhood radius and minimum neighbor number. Specifically, the density connectivity point set (i.e., the boundary point set) obtained from clustering is physically reassigned to allocate all density connectivity points to normal operating condition clusters, including: Filter fuzzy connection points in the density connection point set; the fuzzy connection point is a point in the density connection point set that belongs to the intersection of the neighborhood of the core point of multiple normal operating condition clusters; define all normal operating condition clusters to which a certain fuzzy boundary point belongs as the candidate clusters to which the fuzzy boundary point belongs; For each fuzzy connection point, find the row vector corresponding to the fuzzy connection point in the windowed feature matrix obtained after time window convergence, and use the row vector as the original time window statistical vector corresponding to the fuzzy connection point. For each fuzzy connection point, the median vector of the original time window statistical vector of each candidate cluster is determined in the windowed feature matrix obtained after time window convergence. For each fuzzy connection point, calculate the Euclidean distance between its corresponding original time window statistical vector and the median vector; select the candidate cluster with the smallest Euclidean distance as the cluster to which the fuzzy connection point belongs; Remove each fuzzy connection point from the density connection point set and add it to the corresponding family in the normal operating condition cluster; For other density ties that are not fuzzy ties, they have only one belonging cluster. They are removed from the density ties set and added to the corresponding belonging cluster in the normal operating condition cluster. If the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, then the isolated forest model is used to perform local sparse anomaly detection on the multi-source data to obtain the detection result; the detection result includes whether there are anomalies or no anomalies in the normal operating condition cluster. If the detection result indicates that there are abnormal points within the normal operating condition cluster, the abnormal points are further judged to filter and obtain the final abnormal points; if there are still abnormal points within the normal operating condition cluster after the filtering process, a fault warning is issued.

2. The method for early warning of ventilation and air conditioning fan failure based on multi-source data according to claim 1, characterized in that, The multi-source data includes fan operating parameters, fan vibration time-domain characteristics, fan vibration spectrum characteristics, and temperature parameters. Among them, the fan operating parameters include static pressure, air volume, operating frequency, and motor input current; the fan vibration time-domain characteristics include RMS acceleration, peak acceleration, kurtosis acceleration, RMS velocity, peak velocity, kurtosis velocity, and peak-to-peak displacement; the fan vibration spectrum characteristics include acceleration spectrum and velocity spectrum; and the temperature parameters include bearing housing temperature, motor winding temperature, and ambient temperature.

3. The method for early warning of ventilation and air conditioning fan failure based on multi-source data according to claim 2, characterized in that, The preprocessing includes: data resampling to transform irregular time interval data into time series data with the same interval; outlier detection and removal; missing value imputation; time window convergence; standardization; and PCA dimensionality reduction.

4. The method for early warning of ventilation and air conditioning fan failure based on multi-source data according to claim 3, characterized in that, The aggregation rules in the time window convergence are as follows: for the fan operating parameters, calculate the average value and standard deviation, and extract the maximum and minimum values; for the fan vibration time domain characteristics, calculate the average value and extract the maximum value; for the fan vibration spectrum characteristics, extract the maximum energy of the bearing fault characteristic frequency point; for the temperature parameters, calculate the average temperature and extract the highest temperature.

5. A method for early warning of ventilation and air conditioning fan failure based on multi-source data according to any one of claims 1-4, characterized in that, If the detection result indicates the presence of anomalies within the normal operating condition cluster, further judgment of the anomalies includes: For outliers within a normal operating condition cluster, compare their corresponding outlier score calculated using the isolated forest model with the cluster threshold: if the outlier score is greater than the cluster threshold, then retain the outlier; wherein, the cluster threshold is calculated as follows: For each normal operating condition cluster, the intra-cluster threshold is calculated using the following formula. : ； In the formula, Indicates adaptive confidence level. , Indicates the sample size within a cluster; The empirical quantiles, representing the empirical cumulative distribution function, are determined as follows: Find the quantiles that satisfy the empirical cumulative distribution function. If the abnormal score value s is found, then the search results will be... ,in, ； In the formula, This represents the j-th normal operating condition cluster. This represents the anomaly score of the k-th sample calculated using the isolated forest model; Let be the indicator function, which is defined as: 。 6. A method for early warning of ventilation and air conditioning fan failure based on multi-source data according to claim 5, characterized in that, If the detection result indicates the presence of anomalies within the normal operating condition cluster, further judgment of the anomalies includes: verifying the physical deviation of the boundary points of the anomalies to filter out falsely reported anomalies, specifically including: For anomalies within the normal operating condition cluster, determine whether they belong to the density connection point set; if they do, extract the row vector corresponding to the anomaly in the windowed feature matrix obtained after time window convergence as the original time window statistical vector corresponding to the anomaly. Based on the original time window statistical vector corresponding to the outlier and the median vector of the original time window statistical vector of each cluster corresponding to the normal operating condition cluster, the deviation is calculated. : ； In the formula, , Indicates outliers The corresponding original time window statistical vector; This represents the median vector of the original time window statistics vectors corresponding to each cluster (i) in the normal operating condition cluster; To avoid small quantities where the denominator is zero; Represents the L2 norm; If deviation If the value is less than the preset threshold, the anomaly is determined to be a false alarm and deleted; otherwise, it is retained.

7. A ventilation and air conditioning fan fault early warning system based on multi-source data, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: The data acquisition module is configured to acquire multi-source data from ventilation and air conditioning fans; A data preprocessing module configured to preprocess the multi-source data; The dynamic clustering module is configured to use an improved DBSCAN clustering model to perform dynamic operating condition clustering on preprocessed multi-source data and obtain clustering results; the clustering results include whether the multi-source data belongs to the normal operating condition cluster or to outliers. An anomaly detection module is configured to perform local sparse anomaly detection on the multi-source data using an isolated forest model if the clustering result indicates that the multi-source data belongs to the normal operating condition cluster, and obtain the detection result; the detection result includes whether there are anomalies or not within the normal operating condition cluster. An anomaly filtering module is configured to further judge the anomaly if the detection result indicates that there is an anomaly in the normal operating condition cluster, so as to filter and obtain the final anomaly. The fault warning module is configured to issue a fault warning when two situations exist: if the clustering result indicates that the multi-source data belongs to an outlier, then a fault warning is issued; if there are still outliers in the normal operating condition cluster after screening, then a fault warning is issued.

Images