Intermediate frequency furnace combined working condition identification method based on harmonic disturbance data clustering

Abstract

The invention relates to an intermediate frequency furnace combination condition identification method based on harmonic disturbance data clustering, and the method comprises the steps: collecting the power quality monitoring data of all intermediate frequency furnaces from a PCC point, and extracting a feature vector of a fixed dimension through employing a feature extraction method based on a sliding time window; dimensionality reduction is carried out on the extracted high-dimensional feature data through principal component analysis (PCA); clustering all the feature data after dimension reduction by using an HDBSCAN clustering algorithm, and performing timestamp alignment on working condition time sequences of different intermediate frequency furnaces to generate a combined working condition label; constructing a training data set with labels by taking the feature data after dimension reduction as input and the combined working condition labels as output; then, an XGBoost algorithm is adopted to train a classification model, and testing and verification are carried out; newly-collected PCC point data is processed and then input into the trained classification model, and a real-time combined working condition label is output. According to the method, the working condition during parallel operation of the intermediate frequency furnace can be accurately identified only through the electric energy quality data of the common connection point of the wire inlet ends.

Description

Technical Field

[0001] This invention relates to the field of power quality monitoring technology, specifically to a method for identifying the combined operating conditions of a medium-frequency furnace based on harmonic disturbance data clustering. Background Technology

[0002] With the continuous improvement of technology, the widespread application of modern power electronic devices in industrial enterprises has significantly improved production efficiency. However, it has also injected a large amount of harmonics into the power grid, bringing about a series of power quality problems. Induction furnaces, as core smelting equipment in industries such as metallurgy and casting, exhibit significant nonlinear and impulsive characteristics during operation. Their current, voltage, and harmonic characteristics show marked differences at different operating stages (such as preheating, melting, and holding), especially during the melting stage, which is often accompanied by severe harmonic disturbances, significantly impacting the stable operation of the power grid. In actual production, casting workshops often configure multiple induction furnaces to operate in parallel, with alternating operating conditions. This results in the power quality data at the point of common coupling (PCC) at the incoming line being a mixed superposition of the operating conditions of all operating equipment, exhibiting extremely complex characteristics that are difficult to directly analyze. This makes it impossible to accurately distinguish the harmonic pollution characteristics under specific operating conditions, thus hindering the implementation of efficient mitigation measures.

[0003] Existing methods typically use high-dimensional, redundant raw power quality monitoring data directly, lacking effective dimensionality reduction and feature extraction techniques. The mixed signals from multiple medium-frequency furnaces operating in parallel are difficult to analyze. Current research often focuses on single equipment, failing to distinguish complex combined operating conditions (such as "furnace A melting + furnace B holding"). Furthermore, it cannot effectively establish a mapping relationship between power quality monitoring data at the input common connection point (PCC) and the combined operating states of lower-level equipment. Consequently, it is impossible to establish an accurate harmonic spectrum model corresponding to specific operating conditions, resulting in a lack of specificity in the configuration and control strategies of harmonic mitigation devices, low mitigation efficiency, and difficulty in achieving precise mitigation. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying the combined operating conditions of medium-frequency furnaces based on harmonic disturbance data clustering. This method can accurately identify the operating conditions of medium-frequency furnaces operating in parallel using only the power quality data of the common connection point at the incoming line.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: a method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering, comprising the following steps:

[0006] Step S1: Collect power quality monitoring data for all intermediate frequency furnaces from the PCC point;

[0007] Step S2: For the power quality monitoring data of each intermediate frequency furnace, a feature vector of fixed dimension is extracted using a feature extraction method based on a sliding time window;

[0008] Step S3: Dimensionality reduction of the high-dimensional feature data extracted in step S2 is performed using principal component analysis (PCA);

[0009] Step S4: Use the HDBSCAN clustering algorithm to cluster all the dimensionality-reduced feature data, determine the typical operating conditions corresponding to each cluster, and align the time series of operating conditions of different intermediate frequency furnaces with timestamps to generate combined operating condition labels.

[0010] Step S5: Using the feature data reduced by dimensionality in step S3 as input and the combined working condition labels generated in step S4 as output, construct a labeled training dataset; then, based on the labeled training dataset, train a classification model using the XGBoost algorithm, and test and verify it.

[0011] Step S6: After processing the newly collected PCC point data in steps S2-S3, input the data into the trained classification model and output real-time combined working condition labels.

[0012] Furthermore, step S2 specifically includes:

[0013] The collected power quality time series data is divided into sliding windows of fixed length W, with adjacent windows sliding with a step size s to generate multiple window subsequences;

[0014] Three types of features are extracted from each window: first-order features, frequency domain features, and time-series correlation features, forming a window feature vector;

[0015] Mean pooling aggregates multiple window features into a fixed-dimensional feature vector at the sample level.

[0016] Furthermore, the first-order features include mean, maximum and minimum values, standard deviation, skewness, and kurtosis; the frequency domain features include harmonic content and the energy proportion of each harmonic amplitude; and the time-series correlation features include autocorrelation coefficient and rising / falling slope within the window.

[0017] Furthermore, step S3 specifically includes:

[0018] The harmonic monitoring data in the power quality monitoring data, namely the high-dimensional special data in the feature vector extracted in step S2, is standardized to eliminate the influence of dimensions and obtain standardized harmonic monitoring data.

[0019] For standardized harmonic monitoring data, calculate the covariance matrix;

[0020] For the covariance matrix, solve for its eigenvalues ​​and corresponding eigenvectors;

[0021] Select principal components whose cumulative contribution rate is greater than a set threshold, reduce the original features to a low-dimensional space, and output the reduced feature dataset Y.

[0022] Furthermore, step S4 specifically includes:

[0023] Calculate the core distance and reach distance of each point in the feature dataset Y based on the minimum number of samples k, and construct a minimum spanning tree with the reach distance as the edge weight;

[0024] Based on the cluster stability index, the hierarchical clustering tree is traversed from top to bottom, and the optimal cluster with stability higher than the sum of its subclusters is retained. Unassigned points are marked as noise.

[0025] Based on the operating mechanism of the medium-frequency furnace, the typical operating conditions corresponding to each cluster are determined, including preheating, melting and heat preservation. The time series of operating conditions of multiple medium-frequency furnaces are timestamped to generate combined operating condition labels.

[0026] Furthermore, in step S5, the XGBoost algorithm is used to train the classification model, specifically as follows:

[0027] An objective function is constructed, and regularization terms L1 and L2 are introduced into the objective function. The Softmax multi-class log loss function is adopted, and the hyperparameters including learning rate, tree depth, number of iterations, subsample ratio, and feature sampling ratio are optimized using cross-validation to ensure a balance between classification accuracy and computational efficiency.

[0028] The present invention also provides a computer device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, which implement the above-described method when executed by the processor.

[0029] The present invention also provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the above-described method.

[0030] Compared with existing technologies, this invention has the following advantages: This invention proposes a method for identifying combined operating conditions of medium-frequency furnaces based on harmonic disturbance data clustering. It can accurately identify the current operating status of medium-frequency furnace users using only power quality data from the incoming line's common connection point (PCC). This method first uses principal component analysis (PCA) to reduce the dimensionality of power quality monitoring data, extracting key features and reducing redundancy and complexity. Then, it uses the HDBSCAN clustering algorithm to distinguish typical operating conditions at different operating stages, establishing a set of combined operating conditions, thereby overcoming the problem of mixed and difficult-to-analyze operating conditions when multiple medium-frequency furnaces are operating in parallel. Based on the XGBoost algorithm, a mapping model from mixed power quality signals at the incoming line to specific combined operating conditions is constructed, enabling the identification of the current system operating status using only the electrical quantities at the incoming line. This invention provides crucial support for subsequent precise harmonic mitigation: for each identified combined operating condition, a corresponding harmonic spectrum model at the incoming line can be established, supporting dynamic and directional control of the mitigation device, greatly improving the accuracy and efficiency of harmonic mitigation, helping to improve power grid quality and ensure stable production for enterprises. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the implementation of the method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering, provided in an embodiment of the present invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] like Figure 1 As shown, this embodiment provides a method for identifying the combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering, including the following steps:

[0036] Step S1: Collect power quality monitoring data for all intermediate frequency furnaces from the PCC point.

[0037] Step S2: For the power quality monitoring data of each intermediate frequency furnace, a feature vector of fixed dimension is extracted using a feature extraction method based on a sliding time window.

[0038] In this embodiment, step S2 specifically includes:

[0039] 1) Divide the collected power quality time series data into sliding windows of fixed length W, with adjacent windows sliding with a step size s, to generate multiple window subsequences;

[0040] 2) Extract three types of features for each window: first-order features, frequency domain features, and time-series correlation features to form a window feature vector.

[0041] The first-order features include mean, maximum and minimum values, standard deviation, skewness and kurtosis; the frequency domain features include harmonic content and the energy proportion of each harmonic amplitude; and the time-series correlation features include autocorrelation coefficient and rising / falling slope within the window.

[0042] 3) Multiple window features are aggregated into a fixed-dimensional feature vector at the sample level by mean pooling.

[0043] Step S3: Dimensionality reduction of the high-dimensional feature data extracted in step S2 is performed using principal component analysis (PCA).

[0044] In this embodiment, step S3 specifically includes:

[0045] 1) Standardize the harmonic monitoring data in the power quality monitoring data, i.e. the high-dimensional special data in the feature vector extracted in step S2, to eliminate the influence of dimensions and obtain standardized harmonic monitoring data.

[0046] 2) Calculate the covariance matrix for standardized harmonic monitoring data;

[0047] 3) For the covariance matrix, solve for its eigenvalues ​​and corresponding eigenvectors;

[0048] 4) Select principal components whose cumulative contribution rate is greater than a set threshold, reduce the dimensionality of the original features to a low-dimensional space, and output the dimensionality-reduced feature dataset Y. In this embodiment, the threshold is set to 90%.

[0049] Step S4: Use the HDBSCAN clustering algorithm to cluster all the dimensionality-reduced feature data, determine the typical operating conditions corresponding to each cluster, and align the time series of operating conditions of different intermediate frequency furnaces with timestamps to generate combined operating condition labels.

[0050] In this embodiment, step S4 specifically includes:

[0051] 1) Calculate the core distance and reach distance of each point in the feature dataset Y based on the minimum number of samples k, and construct the minimum spanning tree with the reach distance as the edge weight;

[0052] 2) Based on the cluster stability index, traverse the hierarchical clustering tree from top to bottom, retain the optimal cluster whose stability is higher than the sum of its subclusters, and mark unassigned points as noise;

[0053] 3) Based on the operating mechanism of the medium-frequency furnace, determine the typical operating conditions corresponding to each cluster, including preheating, melting and heat preservation. Time stamp alignment is performed on the operating condition time series of multiple medium-frequency furnaces to generate combined operating condition labels.

[0054] Step S5: Using the feature data after dimensionality reduction in step S3 as input and the combined working condition labels generated in step S4 as output, construct a labeled training dataset; then, based on the labeled training dataset, train a classification model using the XGBoost algorithm, and test and verify it.

[0055] In step S5, the XGBoost algorithm is used to train the classification model, and its specific implementation is as follows:

[0056] An objective function is constructed, and regularization terms L1 and L2 are introduced into the objective function. The Softmax multi-class log loss function is adopted, and the hyperparameters including learning rate, tree depth, number of iterations, subsample ratio, and feature sampling ratio are optimized using cross-validation to ensure a balance between classification accuracy and computational efficiency.

[0057] Step S6: After processing the newly collected PCC point data in steps S2-S3, input the data into the trained classification model and output real-time combined working condition labels.

[0058] The relevant content of this invention will be further explained below.

[0059] 1. Principal Component Analysis (PCA)

[0060] Because the harmonic monitoring data from intermediate frequency furnaces has a high dimensionality, directly using it for cluster analysis would significantly reduce the differences in Euclidean distance between data points, leading to the "curse of dimensionality" and reducing the accuracy of cluster analysis. Furthermore, noise and redundancy issues become increasingly complex and difficult to handle with increasing dimensionality, potentially reducing the usability of clustering algorithms and resulting in excessive computational demands, impacting subsequent model training. To address the problems caused by the high dimensionality of harmonic monitoring data, this invention employs Principal Component Analysis (PCA) to reduce the dimensionality of the harmonic data. The specific steps are as follows:

[0061] Because the harmonic monitoring data of the three-phase symmetrical intermediate frequency furnace is

[0062] (1)

[0063] Among them, represents the fundamental active power in the \(i\)-th monitoring period; \(I_1\) (i) represents the fundamental current in the \(i\)-th monitoring period; \(I\) h *(i) represents the content rate of the \(h\)-th harmonic current in the \(i\)-th monitoring period; \(U\) h (i) represents the \(h\)-th harmonic voltage in the \(i\)-th monitoring period; \(t\) is the number of samples of harmonic monitoring data.

[0064] 1) For the harmonic monitoring data \(X\), considering the different dimensions between variables, it is necessary to eliminate the differences in different dimensions. Therefore, standardization processing is adopted to obtain the standardized harmonic monitoring data \(Z\) as shown in the following formula.

[0065] (2)

[0066] Where:

[0067] (3)

[0068] Among them, \(n\) represents the number of dimensions of the power quality monitoring data adopted; is the element in the \(i\)-th row and \(j\)-th column of \(X\), is the mean value of the \(j\)-th column of \(X\).

[0069] 2) For the standardized harmonic monitoring data \(Z\), calculate its covariance matrix \(\sum\) as shown in the following formula.

[0070] (4)

[0071] Where:

[0072] (5)

[0073] Among them, \(Z\) x (i) represents the element in the \(i\)-th row and \(x\)-th column of \(Z\); represents the mean value of the \(x\)-th column of \(Z\).

[0074] 3) For the covariance matrix \(\sum\), solve its eigenvalues and corresponding eigenvectors. The eigenvalues of the covariance matrix are denoted as \(\lambda_1, \lambda_2, \cdots, \lambda\) n , and the corresponding eigenvectors are \(a_1, a_2, \cdots, a\) n . The eigenvectors are expressed as:

[0075] (6)

[0076] 4) The value of the number \(m\) (\(m < n\)) of the retained principal components is obtained according to the following formula.

[0077] (7)

[0078] Here, β is the selection threshold for principal components. The larger the value of β, the more principal components are selected, and the more original data information is retained. Since principal components with small eigenvalues ​​retain very limited information, and the ultimate goal of principal component analysis is to achieve data dimensionality reduction, a β value of 90% is selected after comprehensive consideration.

[0079] 5) Calculate the m principal components e1, e2, ..., e3 according to the following formula. m .

[0080] (8)

[0081] 6) Using m principal components e1, e2, ..., e m The dataset is constructed as shown in the following formula, and the dimensionality-reduced power quality monitoring data Y is obtained.

[0082] (9)

[0083] Using the above method, the dimension of harmonic monitoring data in power quality monitoring data is reduced from n-dimensional to m-dimensional, which helps to improve the accuracy of subsequent mode division.

[0084] 2. Typical Working Condition Clustering Mining Method Based on HDBSCAN

[0085] To automatically distinguish the operating states of different intermediate frequency furnaces from monitoring data, this invention employs the HDBSCAN clustering algorithm. This algorithm constructs a hierarchical tree structure based on reachability and selects the optimal clustering result by combining cluster stability indicators, thus avoiding the problem of traditional DBSCAN requiring preset threshold parameters.

[0086] First, set the dataset to D, for points Distance to its k nearest neighbors Defined as a point core distance ,Right now:

[0087] (10)

[0088] Secondly, define any two points and Interconnection distance for:

[0089] (11)

[0090] in, Point and The Euclidean distance between them This represents the k nearest neighbors of a point.

[0091] The following steps will be taken to implement this:

[0092] 1) Calculate the core distance and the reach distance.

[0093] Calculate the core distance for each point in dataset D. and each pair of points and Interconnection distance , where k is the minimum number of samples.

[0094] 2) Construct the minimum spanning tree and the clustering hierarchy tree

[0095] Using the distance between nodes as edge weights, construct a complete graph and extract its Minimum Spanning Tree (MST). Sort the edges of the MST in descending order of weight. As the threshold gradually increases, different connected components gradually split, thus forming a hierarchical clustering tree.

[0096] 3) Cluster stability calculation

[0097] definition The reciprocal of the density level, i.e.:

[0098] (12)

[0099] Where distance is the distance threshold; A larger value indicates a higher density scale. For a given cluster C, its stability is defined as:

[0100] (13)

[0101] Among them, for point , It is the λ value when it leaves the cluster. It is the λ value when it joins the cluster.

[0102] 4) Select the best cluster

[0103] The compressed hierarchical tree is traversed from top to bottom. For each cluster in the tree, its stability is compared to the sum of the stability of its subclusters. If the stability of the current cluster is greater than the sum of the stability of all its subclusters, the current cluster is retained and all its subclusters are discarded; otherwise, the stability of the current cluster is set to the sum of the stability of its subclusters, and the check continues downward. This results in a set of the most stable, non-overlapping clusters; points not assigned to any cluster are marked as noise.

[0104] 5) Extraction of typical working conditions and construction of labels

[0105] Based on the above clustering results and combined with the operating cycle mechanism of intermediate frequency furnaces, the typical operating conditions corresponding to each cluster are determined and assigned operating condition labels. At the same time, the time series of operating conditions of different intermediate frequency furnaces are timestamped to integrate the operating status of multiple devices with a unified time base, ultimately forming a combined operating condition label that can comprehensively characterize the overall power consumption characteristics of the incoming line.

[0106] 3. Construction of a Combined Working Condition Classification Model Based on XGBoost

[0107] To achieve accurate identification of combined operating conditions of multiple intermediate frequency furnaces based on incoming power quality data, this invention employs the XGBoost (eXtreme Gradient Boosting) algorithm to construct a classification model. This model takes the features reduced by PCA as input and the combined operating condition labels obtained from HDBSCAN clustering as output. Through supervised learning, it establishes a mapping relationship from mixed harmonic signals to operating states, thereby enabling real-time identification of operating conditions based solely on incoming measurement data.

[0108] 1) Model Principle

[0109] XGBoost, as an efficient gradient boosting tree model, can effectively handle high-dimensional nonlinear features, possesses strong generalization ability and high classification accuracy, and is suitable for classification problems involving multivariable and nonlinear operating conditions, such as those of medium-frequency furnaces. Its objective function is as follows:

[0110] (14)

[0111] in, The loss function measures the true label. and predicted value The differences between them; This is a regularization term used to limit the complexity of a single tree and avoid overfitting.

[0112] 2) Model training methods

[0113] Loss function: Softmax multi-class log loss function is used.

[0114] (15)

[0115] in, Let i be an indicator variable for whether sample i belongs to category k. To predict probabilities.

[0116] Regularization constraints: introduced into the objective function and Regularization terms, combined with structural parameters such as the maximum depth of the tree, the number of leaf nodes, and the minimum number of split samples, effectively control model complexity and enhance generalization ability.

[0117] Hyperparameter optimization: Cross-validation is used to fine-tune hyperparameters such as learning rate, tree depth, number of iterations, subsample ratio, and feature sampling ratio to ensure a balance between classification accuracy and computational efficiency.

[0118] 3) Classification model output

[0119] After training, the XGBoost model can directly input the newly acquired power quality features of the incoming line and output the corresponding combined operating condition label, realizing a fast and accurate mapping from mixed signals to specific operating conditions.

[0120] 4 Feature Extraction Method Based on Sliding Time Window

[0121] Since harmonic monitoring data collected under different operating conditions often appear as time series of varying lengths, directly inputting them into traditional machine learning models such as XGBoost would result in a mismatch in input dimensions, making model training impossible. Therefore, this invention employs a statistical feature extraction method based on a sliding time window to convert time series data of arbitrary length into feature vectors of uniform dimensions, thereby achieving standardized processing of the input features and making them suitable for the XGBoost classification model.

[0122] 1) Time window division

[0123] Power quality monitoring time-series data is divided into sliding windows of fixed length 𝑊, with adjacent windows sliding at a step size 𝑆 to ensure continuous capture of continuous changes in operating conditions. This method can generate multiple window subsequences of consistent length from original sequences of different lengths.

[0124] 2) Statistical feature extraction

[0125] A series of statistical characteristics were calculated for the harmonic monitoring sequence within each time window, including:

[0126] First-order features: mean, maximum, minimum, standard deviation, skewness, kurtosis;

[0127] Frequency domain characteristics: harmonic content (THD), energy percentage of each harmonic amplitude;

[0128] Time-series correlation: autocorrelation coefficient, rising / falling slope within the window;

[0129] Each time window can ultimately be mapped to a feature vector of fixed length:

[0130] (16)

[0131] Where m is the number of features, which is independent of the time series length.

[0132] 3) Feature sequence integration

[0133] If the original time series corresponding to a sample is divided into K time windows and K m-dimensional feature vectors are extracted, then these window-level features need to be aggregated into a single sample-level fixed-dimensional feature vector. This invention employs the mean pooling method to calculate the average value of each feature dimension j across all K windows:

[0134] (17)

[0135] Finally, an m-dimensional feature vector X is generated as the input representation of the sample. This method can preserve the statistical properties of the overall sequence while smoothing out noise interference, ensuring that input sequences of different lengths can be converted into features of uniform dimension, thus meeting the input requirements of the XGBoost classification model.

[0136] This embodiment also provides a computer device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, which implement the above-described method when executed by the processor.

[0137] This embodiment also provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method.

[0138] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering, characterized in that, Includes the following steps: Step S1: Collect power quality monitoring data for all intermediate frequency furnaces from the PCC point; Step S2: For the power quality monitoring data of each intermediate frequency furnace, a feature vector of fixed dimension is extracted using a feature extraction method based on a sliding time window; Step S3: Dimensionality reduction of the high-dimensional feature data extracted in step S2 is performed using principal component analysis (PCA); Step S4: Use the HDBSCAN clustering algorithm to cluster all the dimensionality-reduced feature data, determine the typical operating conditions corresponding to each cluster, and align the time series of operating conditions of different intermediate frequency furnaces with timestamps to generate combined operating condition labels. Step S5: Using the feature data reduced by dimensionality in step S3 as input and the combined working condition labels generated in step S4 as output, construct a labeled training dataset; then, based on the labeled training dataset, train a classification model using the XGBoost algorithm, and test and verify it. Step S6: After processing the newly collected PCC point data in steps S2-S3, input the data into the trained classification model and output real-time combined working condition labels.

2. The method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering according to claim 1, characterized in that, Step S2 specifically includes: The collected power quality time series data is divided into sliding windows of fixed length W, with adjacent windows sliding with a step size s to generate multiple window subsequences; Three types of features are extracted from each window: first-order features, frequency domain features, and time-series correlation features, forming a window feature vector; Mean pooling aggregates multiple window features into a fixed-dimensional feature vector at the sample level.

3. The method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering according to claim 2, characterized in that, The first-order features include mean, maximum and minimum values, standard deviation, skewness, and kurtosis; the frequency domain features include harmonic content and the energy proportion of each harmonic amplitude; and the time-series correlation features include autocorrelation coefficient and rising / falling slope within the window.

4. The method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering according to claim 1, characterized in that, Step S3 specifically includes: The harmonic monitoring data in the power quality monitoring data, namely the high-dimensional special data in the feature vector extracted in step S2, is standardized to eliminate the influence of dimensions and obtain standardized harmonic monitoring data. For standardized harmonic monitoring data, calculate the covariance matrix; For the covariance matrix, solve for its eigenvalues ​​and corresponding eigenvectors; Select principal components whose cumulative contribution rate is greater than a set threshold, reduce the original features to a low-dimensional space, and output the reduced feature dataset Y.

5. The method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering according to claim 1, characterized in that, Step S4 specifically includes: Calculate the core distance and reach distance of each point in the feature dataset Y based on the minimum number of samples k, and construct a minimum spanning tree with the reach distance as the edge weight; Based on the cluster stability index, the hierarchical clustering tree is traversed from top to bottom, and the optimal cluster with stability higher than the sum of its subclusters is retained. Unassigned points are marked as noise. Based on the operating mechanism of the medium-frequency furnace, the typical operating conditions corresponding to each cluster are determined, including preheating, melting and heat preservation. The time series of operating conditions of multiple medium-frequency furnaces are timestamped to generate combined operating condition labels.

6. The method for identifying combined operating conditions of an intermediate frequency furnace based on harmonic disturbance data clustering according to claim 1, characterized in that, In step S5, the XGBoost algorithm is used to train the classification model, specifically as follows: An objective function is constructed, and regularization terms L1 and L2 are introduced into the objective function. The Softmax multi-class log loss function is adopted, and the hyperparameters including learning rate, tree depth, number of iterations, subsample ratio, and feature sampling ratio are optimized using cross-validation to ensure a balance between classification accuracy and computational efficiency.

7. A computer device, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as claimed in any one of claims 1-6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by a processor, the method described in any one of claims 1-6 is implemented.

Images