Coal mill state analysis method based on HMM-GMM model

Abstract

The invention discloses a coal mill state analysis method based on an HMM-GMM model, and belongs to the field of industrial equipment monitoring, and the state analysis method comprises the steps: obtaining the operation state data of a coal mill, and carrying out the preprocessing of the operation state data, so as to obtain the preprocessing data; obtaining a data cluster structure based on the preprocessed data; on the basis of the data cluster structure, obtaining a feature extraction method of operation data of the coal mill; on the basis of the feature extraction method, nonlinear multi-dimensional features capable of effectively representing the running state of the coal mill are obtained; based on the features, a coal mill operation state quantitative evaluation model is obtained; based on the model, obtaining the likelihood probability of performance reduction of the coal mill; and based on the likelihood probability, obtaining a quantitative evaluation index of the running state of the coal mill, and further obtaining the performance reduction degree of the coal mill. The problems that the fault analysis range of the running state of the coal mill is narrow, and pre-analysis is difficult to carry out are solved, so that the precision of state analysis work of the coal mill, the technical application range and the state analysis efficiency are fully improved.

Description

Technical Field

[0001] This application belongs to the field of industrial equipment monitoring, specifically, it relates to a coal mill condition analysis method based on the HMM-GMM model. Background Technology

[0002] As a crucial piece of equipment in thermal power plant systems, the operating status of coal mills significantly impacts the operational quality of other equipment. Therefore, it is essential to analyze and monitor the operating status of coal mills to ensure they operate efficiently and normally. Current methods for coal mill status analysis include manual inspection and status analysis based on operational characteristics. However, these methods consume substantial manual resources, struggle to predict operational faults, and can only identify faults based on pre-defined characteristic states, failing to recognize other potential faults and resulting in a narrow range of analytical parameters.

[0003] Therefore, how to develop a method that can perform pre-analysis of the operating status of coal mills and broaden the scope of fault analysis is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To enable advance analysis of the operating status of coal mills and broaden the scope of fault analysis, this application discloses a coal mill condition analysis method based on the HMM-GMM model, specifically: A coal mill condition analysis method based on the HMM-GMM model, the condition analysis method comprising: The operating status data of the coal mill is acquired, and the operating status data is preprocessed to obtain preprocessed data; Based on the preprocessed data, the data cluster structure is obtained; Based on the data cluster structure, a feature extraction method for coal mill operation data is obtained; Based on feature extraction methods, nonlinear multidimensional features that can effectively characterize the operating status of coal mills are obtained; Based on the aforementioned nonlinear multidimensional characteristics, a quantitative evaluation model for the operating status of the coal mill is obtained; Based on the quantitative evaluation model, the likelihood probability of coal mill performance degradation is obtained; Based on the likelihood probability, a quantitative evaluation index of the coal mill's operating status is obtained to determine the degree of performance degradation of the coal mill.

[0005] Optionally, the step of acquiring the operating status data of the coal mill and preprocessing the operating status data to obtain preprocessed data includes: The operating status data of the coal mill is acquired, including coal feed rate, primary air volume, primary air pressure, primary air temperature, mill outlet temperature, pressure difference, and mill current. Obtain the historical extreme values ​​of the operating status data of the coal mill; Based on the historical operating extreme values, the operating status data of the coal mill is preprocessed, and the preprocessing equation is: ; Where Xnorm represents the preprocessing result of the running status data; X represents the measured data of the running status data; Xmin represents the historical minimum value of the running status data; and Xmax represents the historical maximum value of the running status data.

[0006] Optionally, obtaining the data cluster structure based on the preprocessed data includes: The preprocessed data is processed based on an autoencoder to obtain a data mapping space; Based on the data mapping space, obtain the mapping data; Obtain the concentration level of the mapped data, and when the concentration level is not lower than a preset concentration level, obtain the data cluster structure; Before acquiring the data cluster structure, the method further includes a feature extraction method for acquiring coal mill operating data based on the data cluster structure, including: Based on different feature extraction methods, a three-dimensional spatial distribution scatter plot of the preprocessed data is obtained. Based on the three-dimensional spatial distribution scatter plot, the optimal clustering of the coal mill operating status data is obtained; Based on the optimal clustering, the corresponding feature extraction method is obtained, and the corresponding feature extraction method is an autoencoder.

[0007] Optionally, before the step of obtaining the data cluster structure based on the preprocessed data, the following steps are included: Obtain the running status data type and data processing node corresponding to all the preprocessed data; Based on the running data type, obtain the running status data transformation relationship between the running status data types to obtain the running status data influence relationship; Based on the influence relationship of the aforementioned operational status data, establish the connection relationship between the data processing nodes to obtain the association of all data processing nodes; Based on the association of all data processing nodes, the matching degree of the running status data is obtained, and the running status data with a matching degree not lower than the preset matching degree is obtained, and the corresponding preprocessed data is obtained.

[0008] Optionally, the step of obtaining the matching degree of the running status data based on the association of all data processing nodes, and obtaining the running status data whose matching degree is not higher than a preset matching degree, and obtaining the corresponding preprocessed data, includes: Obtain normal operating status data of the coal mill under normal operating conditions; Based on the connection relationship between the normal operation status data and the data processing nodes, a mathematical model between the data processing nodes is obtained to obtain the association between the data processing nodes; Based on the operating status data on the data processing node, and associated with the data processing node, theoretical operating status data of the data processing node that is associated with the data processing unit is obtained; Obtain the deviation between the theoretical operating state data and the operating state data on the data processing node, where the deviation is the matching degree of the operating state; The matching degree of the running status data is compared with the preset matching degree to obtain the running status data whose matching degree is not higher than the preset matching degree, so as to obtain the non-matching data; Remove the operating status data corresponding to the mismatched data to obtain the normal operating status data of the coal mill.

[0009] Optionally, the step of comparing the matching degree of the running status data with a preset matching degree to obtain running status data whose matching degree is not higher than the preset matching degree, so as to obtain non-matching data, includes: Based on the data processing node association, adjacent data processing nodes with association relationships are obtained to obtain a data processing node group; Based on the data type of the running status data in the data processing node group, a bidirectional mathematical model for the transmission of two types of running status data is established. Based on the bidirectional transmission mathematical model, the processed data corresponding to two adjacent data processing nodes in the data processing node group are obtained respectively. Obtain the matching degree between the processed data and the running status data corresponding to the data processing node. If the matching degree of two adjacent data processing nodes is not lower than the preset matching degree, the matching degree of the running status data is not lower than the preset matching degree; otherwise, the matching degree of the running status data is not higher than the preset matching degree. When the matching degree of the running status data is not higher than the preset matching degree, the running status data in the data processing node group is mismatched data.

[0010] Optionally, the method based on feature extraction to obtain nonlinear multidimensional features that can effectively characterize the operating state of the coal mill includes: The HMM model is trained to obtain the trained HMM model; Obtain the data cluster structure corresponding to different time points during the operation of the coal mill, so as to obtain the time-data cluster structure; Obtain the structure of all time-data clusters within the time period and input it into the trained HMM model; Based on the trained HMM model, obtain the output state parameters corresponding to the time-data cluster structure; Based on the output state parameters, nonlinear multidimensional features that effectively characterize the operating state of the coal mill are obtained.

[0011] Optionally, the quantitative evaluation model for obtaining the operating status of the coal mill based on the nonlinear multidimensional features includes: Obtain all time points with data cluster structures within a time period to obtain all time-data cluster structures; Based on the aforementioned time-data cluster structure, the probability density equation is obtained, and the probability density equation is: ; Where p(x) represents the probability density of data point x following a Gaussian mixture model; αi represents the weight of the i-th Gaussian component; N(x; μi, σi2) represents the probability density of the Gaussian distribution; μi represents the mean vector of the i-th Gaussian component; σi2 represents the variance of the i-th Gaussian vector; i represents the index of the Gaussian vector; j represents the total index of the Gaussian vector. Based on the probability density equation, a quantitative evaluation index for the operating status of the coal mill is obtained. The equation for the quantitative evaluation index for the operating status of the coal mill is as follows: ; Wherein, CL represents the quantitative evaluation index of the coal mill's operating status; F(x) represents the probability density distribution of the period to be evaluated; and G(x) represents the baseline distribution of the coal mill under normal operating conditions.

[0012] Optionally, obtaining the likelihood probability of performance degradation of the coal mill based on the quantitative evaluation model includes: The nonlinear multidimensional features of the coal mill operation process are obtained as a dataset; Based on the dataset, the nonlinear characteristics of the coal mill under normal operating conditions are obtained; Based on the quantitative evaluation model, the predicted operating status data of the coal mill is obtained; Based on the predicted operating status data and the nonlinear characteristics of the coal mill under normal operating conditions, the numerical deviation is obtained to obtain the likelihood probability of coal mill performance degradation.

[0013] Optionally, the step of obtaining a quantitative evaluation index of the coal mill's operating status based on the likelihood probability to obtain the degree of performance degradation of the coal mill includes: Obtain the likelihood probability of the coal mill's performance degradation over a period of time, and establish a performance degradation curve; Obtain the values ​​of the quantitative evaluation indicators of the coal mill's operating status within the same time period, and establish a quantitative evaluation indicator curve; By overlapping the performance degradation curve and the quantitative evaluation index curve, the overlap interval is obtained; The area of ​​the overlapping interval is obtained, and the ratio of the area of ​​the region formed by the performance decline curve and the quantitative evaluation index curve is obtained. The ratio represents the degree of performance decline of the coal mill.

[0014] The beneficial effects of this application include: 1. The scope of analysis for the operating status of coal mills has been broadened. In the technical solution of this application, an autoencoder is used to perform unsupervised feature learning on the high-dimensional time-series data of the coal mill for the status analysis, and the core data of equipment performance are extracted. This method does not require prior establishment and comparison of fault features, but directly learns based on the obtained data, thereby significantly broadening the scope of analysis for the operating status.

[0015] 2. Pre-analysis of the coal mill's operating status has been achieved. In the technical solution of this application, an operating status analysis model for the coal mill is established based on the HMM-GMM model. This model allows for prediction of the coal mill's operating status over a given period, thereby identifying the main types of coal mill faults and their severity.

[0016] 3. Improved efficiency in analyzing the operating status of coal mills. In this application's technical solution, the acquired operating status data was preprocessed, and data indicating operational problems was filtered out. Model building was then based on the filtered data, thereby reducing the amount of data required for model building and improving analysis efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the embodiments of this application or the prior art will be briefly introduced below. Obviously, the following description is only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The drawings are used to provide a further understanding of this disclosure and constitute a part of the specification. They are used together with the following detailed description to explain this disclosure, but do not constitute a limitation of this disclosure. In the drawings: Figure 1 A flowchart of a coal mill condition analysis method based on an HMM-GMM model is provided for embodiments of this application; Figure 2A three-dimensional spatial distribution scatter plot in a coal mill condition analysis method based on an HMM-GMM model provided in this application embodiment; Figure 3 A schematic diagram of the data processing node connection relationship in a coal mill state analysis method based on an HMM-GMM model provided in this application embodiment; Figure 4 An HMM condition monitoring diagram in a coal mill condition analysis method based on an HMM-GMM model provided in this application embodiment; Figure 5 This diagram illustrates the probability density estimation based on the GMM model in a coal mill condition analysis method based on the HMM-GMM model provided in this application embodiment. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Furthermore, in the embodiments of this application, "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] Coal mills in thermal power plants determine the diameter of the processed coal particles. Under normal operating conditions, they ensure that the coal particle diameter matches the design standard, guaranteeing the normal and stable operation of the power plant. During operation, it is necessary to analyze the coal mill's operating status and predict its operating status for the next period to ensure that the coal mill remains in a normal and stable operating state. Current methods for analyzing the operating status of coal mills mainly employ manual inspection and analysis based on stored feature states. The former consumes a large amount of manual resources and can only perform post-event analysis, making it difficult to conduct pre-event analysis of the coal mill's operating status. The latter can only identify and monitor certain specific fault manifestations, resulting in a narrow scope of coal mill operating status analysis. Therefore, the analysis of the operating status of coal mills needs to address the problems existing in the current analysis methods. To this end, this application discloses a coal mill state analysis method based on the HMM-GMM model.

[0020] like Figure 1 The diagram shown is a flowchart of a coal mill condition analysis method based on an HMM-GMM model provided in this application embodiment. Specifically: S110. Obtain the operating status data of the coal mill and preprocess the operating status data to obtain preprocessed data.

[0021] S120. Based on the preprocessed data, obtain the data cluster structure.

[0022] S130. Based on the data cluster structure, a feature extraction method for coal mill operation data is obtained.

[0023] S140. Based on feature extraction methods, obtain nonlinear multidimensional features that can effectively characterize the operating status of the coal mill.

[0024] S150. Based on the aforementioned nonlinear multidimensional features, obtain a quantitative evaluation model for the operating status of the coal mill.

[0025] S160. Based on the quantitative evaluation model, obtain the likelihood probability of the coal mill performance decline.

[0026] S170. Based on the likelihood probability, obtain a quantitative evaluation index of the coal mill's operating status to obtain the degree of performance degradation of the coal mill.

[0027] The purpose of the above steps is to establish a method that enables continuous and accurate analysis of the operating status of coal mills. The beneficial effects achieved are that it enables pre-analysis in the analysis of the operating status of coal mills, and at the same time, it greatly expands the scope of coal mill operating status analysis to identify possible fault manifestations during coal mill operation.

[0028] The following will explain the specific details of all the above steps: As described in step S110, the purpose of this step is that coal mills have various parameters during operation, and these parameters often have different physical characteristics. If these parameters are used directly, it is difficult to perform unified analysis on this data. To ensure that the acquired operating status data has a unified dimension and can be used directly, preprocessing is required. This preprocessing of the acquired operating status data allows the obtained data to be established within a unified data processing system. Specifically: The operating status data of the coal mill is acquired, including coal feed rate, primary air volume, primary air pressure, primary air temperature, mill outlet temperature, pressure difference, and mill current. Obtain the historical extreme values ​​of the operating status data of the coal mill; Based on the historical operating extreme values, the operating status data of the coal mill is preprocessed, and the preprocessing equation is: ; in, X norm This represents the preprocessing result of the running status data; X Actual measured data representing operational status data; Xmin This represents the historical minimum value of the running status data; X max This represents the historical maximum value of the running status.

[0029] In addition to obtaining the acquired operational status data, it is also necessary to obtain the acquisition time of the data, so as to generate the correspondence between the time and the operational status data.

[0030] Among these steps, the preprocessing results need to be obtained by acquiring the maximum and minimum values ​​of various data types, such as the maximum and minimum values ​​of coal feed rate and the maximum and minimum values ​​of primary air volume.

[0031] In addition to the running status data type mentioned above, other types of data can obviously be included, and this application does not limit them.

[0032] The beneficial effect of step S110 is that it can process the acquired multiple types of data according to a unified processing method and finally obtain preprocessed data results with unified dimensions.

[0033] As described in step S120, the purpose of this step is that the data displayed in the acquired operational status data may be sporadic results or inevitable results from long-term operation. Therefore, in specific processing, it is necessary to analyze the concentration of various types of data. Data with higher concentration is obviously inevitable data, and its use can ensure that the obtained data has higher application value. Therefore, it is necessary to obtain the data cluster structure. Specifically: The preprocessed data is processed based on an autoencoder to obtain a data mapping space; Based on the data mapping space, obtain the mapping data; Obtain the concentration level of the mapped data, and when the concentration level is not lower than a preset concentration level, obtain the data cluster structure; Before acquiring the data cluster structure, the method further includes a feature extraction method for acquiring coal mill operating data based on the data cluster structure, including: Based on different feature extraction methods, a three-dimensional spatial distribution scatter plot of the preprocessed data is obtained. Based on the three-dimensional spatial distribution scatter plot, the optimal clustering of the coal mill operating status data is obtained; Based on the optimal clustering, the corresponding feature extraction method is obtained, and the corresponding feature extraction method is an autoencoder.

[0034] The acquired preprocessed data is directly processed using an autoencoder to obtain a data mapping space.

[0035] The acquired data cluster structure is essentially a range of data with a high degree of concentration.

[0036] Specifically, the positions of all data points within the mapping space are obtained, as well as the distances between each data point and other data points. The average distance is calculated. If the distance between a data point and all other data points exceeds the average distance, then the data point does not belong to the data cluster structure.

[0037] Alternatively, other existing technologies can be used to obtain the intervals with high concentration of data points and the corresponding data points within the mapping space. The resulting region is the data cluster structure. This application does not limit the methods used.

[0038] Among them, feature extraction methods include autoencoders, PCA, and KPCA.

[0039] Among them, such as Figure 2 The image shown is a scatter plot of the three-dimensional spatial distribution in a coal mill state analysis method based on an HMM-GMM model provided in this application embodiment. The structure of all data clusters can be obtained from the scatter plot.

[0040] The beneficial effect of step S120 is that, in the visualization, more obvious feature separation can be observed based on the cluster structure. The autoencoder data has a semantic structure, with larger inter-cluster spacing and denser internal clusters. Currently, methods for obtaining data cluster structure also include KPCA and PAC. However, KPCA may separate some clusters due to nonlinear mapping, but its cluster structure is clearer than PCA due to the influence of kernel function selection. The middle clusters are partially separated from the right clusters, but the left clusters still have a small amount of overlap. Due to linear constraints, PCA clusters may overlap or be ellipsoidally distributed. The autoencoder successfully captures the intrinsic relationship between data and state that is difficult to represent with PCA / KPCA. It can be seen that the autoencoder can perform good interpretation and learning of nonlinear data and is more suitable for processing the large amount of nonlinear operating data of coal mills. For complex equipment (such as coal mills), its operating state is affected by multiple nonlinear factors. The autoencoder can more accurately capture performance degradation characteristics. In other words, based on the autoencoder, more accurate processing of operating state data can be achieved.

[0041] However, during the implementation of step S120, it was found that directly using all the acquired data, especially when processing it over a long period, resulted in a massive amount of data, placing extremely high demands on the computing resources of the entire system. Therefore, it is necessary to further process the acquired data. To improve data processing efficiency, the obtained runtime data needs to be preprocessed before acquiring the data cluster structure. Specifically: Obtain the running status data type and data processing node corresponding to all the preprocessed data; Based on the running data type, obtain the running status data transformation relationship between the running status data types to obtain the running status data influence relationship; Based on the influence relationship of the aforementioned operational status data, establish the connection relationship between the data processing nodes to obtain the association of all data processing nodes; Based on the association of all data processing nodes, the matching degree of the running status data is obtained, and the running status data with a matching degree not lower than the preset matching degree is obtained, and the corresponding preprocessed data is obtained.

[0042] The data processing node refers to a computing node that carries operational status data. Each data processing node carries operational status data at a certain time point and also contains preprocessed data.

[0043] Each data processing node carries information about the type of operational status data. Among the operational status data collected by the coal mill, some data types can be converted and influence each other, such as primary air volume, primary air pressure, and pressure difference. The data processing nodes corresponding to these types of equipment operational data can be directly connected to establish a data processing node association.

[0044] Among them, for all data processing nodes that can establish associations with data types of running status, direct connections are made to this type of data, and two nodes with a connection relationship are considered adjacent nodes.

[0045] Among them, such as Figure 3The diagram shown illustrates the connection relationships of data processing nodes in a coal mill state analysis method based on an HMM-GMM model provided in this application embodiment. Each data processing node carries operational status data type information. Based on the data type information, the transformation relationships between connected data are obtained. Furthermore, there are seven data processing nodes, A1 to A7. It can be seen that A1 and A4, A1 and A3, A3 and A4, A3 and A6, A4 and A6, A4 and A5, and A5 and A6 are directly related data processing nodes. A1 and A3 to A6 can correspond to primary air volume, primary air pressure, primary air temperature, mill outlet temperature, and pressure difference, respectively. A2 and A7 are data processing nodes that are not directly related to the data in other data processing nodes, and are themselves not directly related either. They can represent coal feed rate and mill current, respectively. Of course, if A2 and A7 are related but not to other nodes, these two data processing nodes can be directly connected. Each node is set with a matching degree (ξ1 to ξ7) based on its operational status data type. For f1 to f8, these represent data transformation functions in the direction of the arrows, representing the operational status data of the coal mill under normal operating conditions carried by two directly related data processing nodes. For example, for A1 and A4, to transform the data processing data corresponding to A1 into the data processing data corresponding to A4, the data transformation function is f7. Furthermore, for the operational status data carried by these two nodes, a transformation from A4 to A1 can also be performed, and the data transformation function is the inverse function of f7. Each pair of directly connected data processing nodes has a data transformation function and its inverse function; the functions appear in pairs.

[0046] Specifically, for any data processing node, the data is transformed according to the transformation relationship of the data on other data processing nodes connected to it, and the processed data corresponding to the connected node is obtained. The original data and the processed data are compared to obtain the matching degree.

[0047] The ultimate goal of the above steps is to filter the preprocessed data. Therefore, this application uses a matching degree comparison method to filter the relevant preprocessed data, setting the necessary filtering criteria. Specifically: Obtain normal operating status data of the coal mill under normal operating conditions; Based on the connection relationship between the normal operation status data and the data processing nodes, a mathematical model between the data processing nodes is obtained to obtain the association between the data processing nodes; Based on the operating status data on the data processing node, and associated with the data processing node, theoretical operating status data of the data processing node that is associated with the data processing unit is obtained; Obtain the deviation between the theoretical operating state data and the operating state data on the data processing node, where the deviation is the matching degree of the operating state; The matching degree of the running status data is compared with the preset matching degree to obtain the running status data whose matching degree is not higher than the preset matching degree, so as to obtain the non-matching data; Remove the operating status data corresponding to the mismatched data to obtain the normal operating status data of the coal mill.

[0048] In the operation of the coal mill, if it is in normal operation, the two data processing nodes with a connection can obviously convert the two types of data they carry. Therefore, it is necessary to set a data conversion method. In this process, the conversion method applied to the coal mill in normal operation can determine whether the two types of data can be converted.

[0049] Among them, the two types of data can be converted between data processing nodes that have a connection relationship. For example, for primary air volume and primary air pressure, during the normal operation of the coal mill, the primary air pressure can be determined based on the primary air volume, or the primary air volume can be determined based on the primary air pressure. Therefore, it is necessary to establish a mathematical model for the mutual conversion between the two.

[0050] Each data processing node carries corresponding operational status data (i.e., operational status data on the data processing node). At the same time, it also receives data processed by a mathematical model, which is the operational data of the device connected to the node and managed by that node (i.e., theoretical operational status data).

[0051] Each data processing node calculates the deviation between the theoretical operating data and the operating status data on the data processing node, and the obtained deviation is the matching degree of the operating status.

[0052] The matching threshold can be set by technical personnel or by the average historical matching value of the node.

[0053] When a mismatch is found, it means that the data between the two current data processing nodes cannot be converted into each other, indicating that at least one of the two types of data can be used to disregard the fact that the coal mill is in a faulty operating state.

[0054] When data is found to be in an abnormal operating state, the preprocessed data corresponding to that data processing node can be obtained.

[0055] Regarding the above steps, it can determine whether the obtained preprocessed data indicates an operational fault in the coal mill. However, this method is only a relatively simple judgment approach and is difficult to handle actual situations. In reality, in the analysis of the coal mill's operational status, some parameters have more stringent requirements, such as primary air pressure and differential pressure, while others have relatively lower requirements, such as primary air volume. In such cases, different weights or analysis priorities are set for the severity of the impact of different parameters. Therefore, it is also necessary to consider how to reflect this technical implementation requirement in the technical solution used. Specifically: Based on the data processing node association, adjacent data processing nodes with association relationships are obtained to obtain a data processing node group; Based on the data type of the running status data in the data processing node group, a bidirectional mathematical model for the transmission of two types of running status data is established. Based on the bidirectional transmission mathematical model, the processed data corresponding to two adjacent data processing nodes in the data processing node group are obtained respectively. Obtain the matching degree between the processed data and the running status data corresponding to the data processing node. If the matching degree of two adjacent data processing nodes is not lower than the preset matching degree, the matching degree of the running status data is not lower than the preset matching degree; otherwise, the matching degree of the running status data is not higher than the preset matching degree. When the matching degree of the running status data is not higher than the preset matching degree, the running status data in the data processing node group is mismatched data.

[0056] For data processing nodes with interconnected relationships, their corresponding data types can be converted to each other. Therefore, two data models can be established between connected data processing nodes; these two mathematical models are bidirectional transfer mathematical models. For example, between primary air volume and primary air pressure, an equation for primary air pressure derived from primary air volume and an equation for primary air volume derived from primary air pressure can be established.

[0057] After obtaining the bidirectional transmission mathematical model, the processed data and the data carried by each data processing node are obtained based on the model, and the data matching degree between the two connected data processing nodes is calculated.

[0058] In the matching degree judgment, cases with a matching degree not higher than the preset matching degree are filtered out, because in this case, it means that the processed data and the data it carries have too large a deviation, and the data obviously has defects.

[0059] In some cases, one node may meet the matching requirements while the other does not. For example, for primary air volume and primary air pressure, since the allowable deviation range for primary air volume is larger and the allowable deviation range for primary air pressure is smaller, the matching degree of primary air volume may meet the requirements while the matching degree of primary air pressure does not meet the requirements. In this case, both data are considered to be mismatched.

[0060] Specifically, all data processing nodes that are connected are grouped together to form data processing node groups, such as... Figure 3 The diagram shown is a schematic diagram of the data processing node connection relationship in a coal mill state analysis method based on the HMM-GMM model provided in an embodiment of this application. The two data processing nodes A1 and A4 are schematic diagrams of one of the data processing node groups. In fact, all data processing nodes with direct connection relationships constitute a data processing node group. For A2 and A7, they are themselves data processing node groups.

[0061] For the data processing node groups A2 and A7, which each contain only a single data processing node, the method for judging mismatched data is as follows: directly obtain the matching degree between the running status data under normal operating conditions and the measured running status data. If a mismatch is found, the obtained running status data is considered abnormal.

[0062] The beneficial effect of the pre-processing step S130 is that it pre-processes all the obtained coal mill operating status data, thereby filtering out the data under normal operating conditions and obtaining the data under abnormal operating conditions of the coal mill. Based on the final result obtained, the status analysis is carried out. This process significantly reduces the amount of data and improves the analysis efficiency.

[0063] As described in step S140, during data processing, based on the HMM model, the overall operating status of the coal mill can be determined during the current operation, and nonlinear multidimensional features can be obtained. This data can then be used for subsequent operation and loading judgments. Specifically: The HMM model is trained to obtain the trained HMM model; Obtain the data cluster structure corresponding to different time points during the operation of the coal mill, so as to obtain the time-data cluster structure; Obtain the structure of all time-data clusters within the time period and input it into the trained HMM model; Based on the trained HMM model, obtain the output state parameters corresponding to the time-data cluster structure; Based on the output state parameters, nonlinear multidimensional features that effectively characterize the operating state of the coal mill are obtained.

[0064] In the HMM model construction and training process, it is assumed that the coal mill has two hidden states: state 0 represents normal performance and state 1 represents performance degradation. The parameters of the HMM model are set as follows: n_components = 2 (hidden states) and n_iter = maximum number of iterations. Discretized data is used for model training. Likelihood probability calculation: After model training, the model.predict method is used to predict the hidden state of the coal mill at each time point. The likelihood probability of performance degradation is obtained by counting the number of times state 1 (performance degradation) appears in the prediction results and dividing by the total number of time points.

[0065] Training the HMM model requires a third-party library. Open the PyCharm terminal and install and use hmmlearn (part of the Scikit-learn ecosystem). Train the HMM model using the following Python code. The code is as follows: import pandas as pd import hmm from hmmlearn import os os.environ['LOKY_MAX_CPU_COUNT'] = '7' # You can modify the number of cores here according to your needs. # Load data data = pd.read_csv('a1test1_kpca_deep_normalized.csv') # Training the HMM model model = hmm.GaussianHMM(n_components=2, covariance_type="diag", n_iter=100) model.fit(data.values) # Save Model import pickle with open('normal_hmm_kpca_model.pkl', 'wb') as file: pickle.dump(model, file) Since the given coal mill operating status data is used as the observation sequence O, the forward algorithm is used to calculate the likelihood probability. The equation reflecting the degree of matching between the sequence and the model is: ; in, Represents the likelihood probability. N Represents the number of states. This represents the m-th computed term in the forward correlation. m This represents an intermediate variable used in the summation iteration. n This indicates the maximum range that can be traversed for summation. λ These represent the initial probability, state transition probability, and observation probability, respectively.

[0066] This application discloses a practical application effect, using historical operating status data of a coal mill, with a sampling time of 6 seconds, spanning from 6:50:00 to 19:26:00 on November 1, 2017. Based on this data, the likelihood probability of coal mill performance degradation is analyzed using a Hidden Markov Model (HMM), providing a theoretical basis for predictive maintenance of the equipment.

[0067] Likelihood probability assessment: Nonlinear features of the coal mill under normal operating conditions are extracted. A trained Hidden Markov Model (HMM) is used for likelihood probability calculation as a benchmark assessment. Features extracted from test data are also used for likelihood probability calculation. The resulting coal mill operating condition monitoring curve is shown below. Figure 4 The diagram shown is an HMM condition monitoring diagram in a coal mill condition analysis method based on an HMM-GMM model provided in an embodiment of this application.

[0068] The beneficial effect of step S130 is that the higher the health performance score of the condition monitoring curve, the higher the corresponding likelihood probability, and the closer the equipment state is to the normal state, i.e., the "healthier" the equipment performance. As shown in the figure, the HMM condition monitoring method using the autoencoder feature extraction method detects a health performance score of 0.1~0.3 for the coal mill in the test set data, enabling it to detect the earliest trend of declining equipment performance. This is because the autoencoder can extract the inherent deep-level mapping relationship between the coal mill's operating data and its state, capturing subtle performance changes in the equipment state at an early stage of performance decline. Furthermore, the existing HMM condition monitoring methods using KPCA and PCA feature extraction methods detect a health performance score of approximately 0.2~0.6 for the coal mill in the test set data, indicating a higher probability of detecting a decline in coal mill performance compared to the autoencoder method, and a slower response to performance decline. Simultaneously, this likelihood probability can be compared with a set threshold. When the probability exceeds the threshold, an early warning mechanism is triggered, reminding staff to pay attention to the operating status of the coal mill and to arrange predictive maintenance work in advance to avoid production interruptions caused by equipment failures, thereby affecting power production. Furthermore, by combining typical patterns of various observed indicators when the coal mill's performance deteriorates, it is possible to analyze whether the current operating status of the coal mill is closer to a normal state or a performance degradation state, providing stronger support for equipment maintenance decisions.

[0069] As described in step S150, the purpose of this step is to determine the cause of the problem after determining the overall operating status of the coal mill. Therefore, this is done by determining quantitative evaluation indicators for operation and a quantitative evaluation model for the operating status. Specifically... Obtain all time points with data cluster structures within a time period to obtain all time-data cluster structures; Based on the aforementioned time-data cluster structure, the probability density equation is obtained, and the probability density equation is: ; in, p ( x ) represents a data point x The probability density follows a Gaussian mixture model; α i Indicates the first i The weights of each Gaussian component; N ( x ; μ i , σ i2 () represents the probability density of a Gaussian distribution; μ i Indicates the first i The mean vector of Gaussian components; σi2 Indicates the first i The variance of a Gaussian vector; i Indicates the index of the Gaussian vector; j Indicates the total number of Gaussian vectors; Based on the probability density equation, a quantitative evaluation index for the operating status of the coal mill is obtained. The equation for the quantitative evaluation index for the operating status of the coal mill is as follows: ; in, CL Indicators representing quantitative assessments of the operating status of coal mills; F ( x () represents the probability density distribution for the period to be evaluated; G ( x () represents the baseline distribution under normal operating conditions of the coal mill.

[0070] Specifically, the baseline probability obtained from the coal mill operating data extracted by the autoencoder and the likelihood probability obtained from the test set data are used as the monitoring indicator of state. The state assessment effect is then compared with the images obtained from the data obtained by PCA and KPCA feature extraction methods.

[0071] In PyCharm, you can use code to build and train a GMM model by installing scikit-learn. The code is as follows: best_n_components1 = 1 best_aic1 = np.inf for n in range(1, 11): gmm = GaussianMixture(n_components=n, covariance_type='full') gmm.fit(data1) aic = gmm.aic(data1) if aic < best_aic1: best_aic1 = aic best_n_components1 = n best_n_components2 = 1 best_aic2 = np.inf for n in range(1, 11): gmm = GaussianMixture(n_components=n, covariance_type='full') gmm.fit(data2) aic = gmm.aic(data2) if aic < best_aic2: best_aic2 = aic best_n_components2 = n # Training GMM Model gmm1=GaussianMixture(n_components=best_n_components1, covariance_type='full') gmm1.fit(data1) gmm2=GaussianMixture(n_components=best_n_components2, covariance_type='full') gmm2.fit(data2) The calculated probability density of the test set is used to plot a comparison curve, which shows the degree of deviation between the distribution during the evaluation period and the baseline distribution. Distribution overlap calculation and evaluation result overlap calculation: The area of ​​the overlapping region between the two distributions is calculated by numerical integration. For data extracted using the autoencoder method, CL=0.0168. (Example...) Figure 5 The diagram shown is a schematic representation of the probability density estimation based on the GMM model in a coal mill condition analysis method based on the HMM-GMM model provided in an embodiment of this application.

[0072] The beneficial effect of step S150 is that, based on the determination of the quantitative evaluation model, the operating status of the coal mill in the next period of time can be predicted based on the model, and the obtained normal operation data and predicted data can be compared to determine the probability and extent of the performance decline of the coal mill.

[0073] As described in step S160, the purpose of this step is to process the data generated during the operation of the coal mill. Specifically, the processing method involves acquiring data from the normal operating state and comparing it with the obtained predicted operating data to obtain the deviation. Clearly, the larger the deviation, the greater the decrease in likelihood probability, and the more abnormal the coal mill's operating state. Conversely, the smaller the deviation, the more normal the coal mill's operating state. Specifically: The nonlinear multidimensional features of the coal mill operation process are obtained as a dataset; Based on the dataset, the nonlinear characteristics of the coal mill under normal operating conditions are obtained; Based on the quantitative evaluation model, the predicted operating status data of the coal mill is obtained; Based on the predicted operating status data and the nonlinear characteristics of the coal mill under normal operating conditions, the numerical deviation is obtained to obtain the likelihood probability of coal mill performance degradation.

[0074] Among them, the obtained nonlinear multidimensional data can be directly used as a dataset.

[0075] The obtained quantitative evaluation model will be directly used to acquire operational status data over a subsequent period of time.

[0076] The two types of data are directly processed to calculate the deviation, which is the likelihood probability of a decrease in the performance of the coal mill.

[0077] As described in step S170, the purpose of this step is that, in the process of analyzing the condition of a coal mill, it is often necessary to predict the operating condition of the coal mill over a period of time based on the obtained results. Therefore, appropriate methods are needed to obtain the results. Specifically: Obtain the likelihood probability of the coal mill's performance degradation over a period of time, and establish a performance degradation curve; Obtain the values ​​of the quantitative evaluation indicators of the coal mill's operating status within the same time period, and establish a quantitative evaluation indicator curve; By overlapping the performance degradation curve and the quantitative evaluation index curve, the overlap interval is obtained; The area of ​​the overlapping interval is obtained, and the ratio of the area of ​​the region formed by the performance decline curve and the quantitative evaluation index curve is obtained. The ratio represents the degree of performance decline of the coal mill.

[0078] For both the obtained likelihood probability values ​​and the values ​​of the quantitative evaluation indicators, corresponding curves can be established.

[0079] Based on the obtained curve, an equation can be fitted to the curve, and then the curve can be plotted for the next period of time based on the fitted equation.

[0080] This allows us to obtain the overlapping intervals of the two curves over a period of time, the area of ​​the overlapping intervals, and the total area of ​​the region formed by the two curves.

[0081] Among them, the overlap of the generated curves of normal state and test state is obtained. The higher the overlap, the higher the similarity between the two data systems. Since the normal state curve is obtained based on the normal operation data, the higher the overlap between the test data and the normal operation data, the closer the generated test data represents the current operation of the coal mill is to the normal state, thereby realizing the quantitative prediction of the operating state of the coal mill in the future.

[0082] The beneficial effect of step S150 is that it can quantitatively predict the operating status of the coal mill over a period of time and ensure the accuracy of the prediction.

[0083] The beneficial effects of this application include: 1. The scope of analysis for the operating status of coal mills has been broadened. In the technical solution of this application, an autoencoder is used to perform unsupervised feature learning on the high-dimensional time-series data of the coal mill for the status analysis, and the core data of equipment performance are extracted. This method does not require prior establishment and comparison of fault features, but directly learns based on the obtained data, thereby significantly broadening the scope of analysis for the operating status.

[0084] 2. Pre-analysis of the coal mill's operating status has been achieved. In the technical solution of this application, an operating status analysis model for the coal mill is established based on the HMM-GMM model. This model allows for prediction of the coal mill's operating status over a given period, thereby identifying the main types of coal mill faults and their severity.

[0085] 3. Improved efficiency in analyzing the operating status of coal mills. In this application's technical solution, the acquired operating status data was preprocessed, and data indicating operational problems was filtered out. Model building was then based on the filtered data, thereby reducing the amount of data required for model building and improving analysis efficiency.

[0086] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to computer program instructions. The aforementioned computer program can be stored in a non-volatile storage medium, and when executed, it performs the steps of the above method embodiments. Alternatively, if the integrated unit of the present invention is implemented as a software functional module and sold or used as an independent product, it can also be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention.

[0087] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for analyzing the condition of a coal mill based on an HMM-GMM model, characterized in that, The state analysis method includes: The operating status data of the coal mill is acquired, and the operating status data is preprocessed to obtain preprocessed data; Based on the preprocessed data, the data cluster structure is obtained; Based on the data cluster structure, a feature extraction method for coal mill operation data is obtained; Based on feature extraction methods, nonlinear multidimensional features that can effectively characterize the operating status of coal mills are obtained; Based on the aforementioned nonlinear multidimensional characteristics, a quantitative evaluation model for the operating status of the coal mill is obtained; Based on the quantitative evaluation model, the likelihood probability of coal mill performance degradation is obtained; Based on the likelihood probability, a quantitative evaluation index of the coal mill's operating status is obtained to determine the degree of performance degradation of the coal mill.

2. The method for coal mill condition analysis based on the HMM-GMM model according to claim 1, characterized in that, The process of acquiring the operating status data of the coal mill and preprocessing the operating status data to obtain preprocessed data includes: The operating status data of the coal mill is obtained, including coal feed rate, primary air volume, primary air pressure, primary air temperature, mill outlet temperature, coal mill pressure difference, and mill current. Obtain the historical extreme values ​​of the operating status data of the coal mill; Based on the historical operating extreme values, the operating status data of the coal mill is preprocessed, and the preprocessing equation is: ； in, X norm This represents the preprocessing result of the running status data; X Actual measured data representing operational status data; X min This represents the historical minimum value of the running status data; X max This represents the historical maximum value of the running status.

3. The method for coal mill condition analysis based on the HMM-GMM model according to claim 1, characterized in that, The step of obtaining the data cluster structure based on the preprocessed data includes: The preprocessed data is processed based on an autoencoder to obtain a data mapping space; Based on the data mapping space, obtain the mapping data; Obtain the concentration level of the mapped data, and when the concentration level is not lower than a preset concentration level, obtain the data cluster structure; Before acquiring the data cluster structure, the method further includes a feature extraction method for acquiring coal mill operating data based on the data cluster structure, including: Based on different feature extraction methods, a three-dimensional spatial distribution scatter plot of the preprocessed data is obtained. Based on the three-dimensional spatial distribution scatter plot, the optimal clustering of the coal mill operating status data is obtained; Based on the optimal clustering, the corresponding feature extraction method is obtained, and the corresponding feature extraction method is an autoencoder.

4. The method for analyzing the condition of a coal mill based on an HMM-GMM model according to claim 1, characterized in that, Before the step of obtaining the data cluster structure based on the preprocessed data, the following is included: Obtain the running status data type and data processing node corresponding to all the preprocessed data; Based on the running data type, obtain the running status data transformation relationship between the running status data types to obtain the running status data influence relationship; Based on the influence relationship of the aforementioned operational status data, establish the connection relationship between the data processing nodes to obtain the association of all data processing nodes; Based on the association of all data processing nodes, the matching degree of the running status data is obtained, and the running status data with a matching degree not lower than the preset matching degree is obtained, and the corresponding preprocessed data is obtained.

5. The method for analyzing the condition of a coal mill based on an HMM-GMM model according to claim 4, characterized in that, The process involves associating all data processing nodes to obtain the matching degree of the running status data, acquiring running status data whose matching degree is not higher than a preset matching degree, and obtaining the corresponding preprocessed data, including: Obtain normal operating status data of the coal mill under normal operating conditions; Based on the connection relationship between the normal operation status data and the data processing nodes, a mathematical model between the data processing nodes is obtained to obtain the association between the data processing nodes; Based on the operating status data on the data processing node, and associated with the data processing node, theoretical operating status data of the data processing node that is associated with the data processing unit is obtained; Obtain the deviation between the theoretical operating state data and the operating state data on the data processing node, where the deviation is the matching degree of the operating state; The matching degree of the running status data is compared with the preset matching degree to obtain the running status data whose matching degree is not higher than the preset matching degree, so as to obtain the non-matching data; Remove the operating status data corresponding to the mismatched data to obtain the normal operating status data of the coal mill.

6. A method for analyzing the condition of a coal mill based on an HMM-GMM model according to claim 4 or 5, characterized in that, The step of comparing the matching degree of the running status data with a preset matching degree to obtain running status data whose matching degree is not higher than the preset matching degree, so as to obtain non-matching data, includes: Based on the data processing node association, adjacent data processing nodes with association relationships are obtained to obtain a data processing node group; Based on the data type of the running status data in the data processing node group, a bidirectional mathematical model for the transmission of two types of running status data is established. Based on the bidirectional transmission mathematical model, the processed data corresponding to two adjacent data processing nodes in the data processing node group are obtained respectively. Obtain the matching degree between the processed data and the running status data corresponding to the data processing node. If the matching degree of two adjacent data processing nodes is not lower than the preset matching degree, the matching degree of the running status data is not lower than the preset matching degree; otherwise, the matching degree of the running status data is not higher than the preset matching degree. When the matching degree of the running status data is not higher than the preset matching degree, the running status data in the data processing node group is mismatched data.

7. The method for coal mill condition analysis based on the HMM-GMM model according to claim 1, characterized in that, The feature extraction method obtains nonlinear multidimensional features that can effectively characterize the operating state of the coal mill, including: The HMM model is trained to obtain the trained HMM model; Obtain the data cluster structure corresponding to different time points during the operation of the coal mill, so as to obtain the time-data cluster structure; Obtain the structure of all time-data clusters within the time period and input it into the trained HMM model; Based on the trained HMM model, obtain the output state parameters corresponding to the time-data cluster structure; Based on the output state parameters, nonlinear multidimensional features that effectively characterize the operating state of the coal mill are obtained.

8. The method for coal mill condition analysis based on the HMM-GMM model according to claim 1, characterized in that, The quantitative evaluation model for obtaining the operating status of the coal mill based on the aforementioned nonlinear multidimensional features includes: Obtain all time points with data cluster structures within a time period to obtain all time-data cluster structures; Based on the aforementioned time-data cluster structure, the probability density equation is obtained, and the probability density equation is: ； in, p ( x ) represents a data point x The probability density follows a Gaussian mixture model; α i Indicates the first i The weights of each Gaussian component; N ( x ; μ i , σ i2 () represents the probability density of a Gaussian distribution; μ i Indicates the first i The mean vector of Gaussian components; σ i2 Indicates the first i The variance of a Gaussian vector; i Indicates the index of the Gaussian vector; j Indicates the total number of Gaussian vectors; Based on the probability density equation, a quantitative evaluation index for the operating status of the coal mill is obtained. The equation for the quantitative evaluation index for the operating status of the coal mill is as follows: ； in, CL Indicators representing quantitative assessments of the operating status of coal mills; F ( x () represents the probability density distribution for the period to be evaluated; G ( x () represents the baseline distribution under normal operating conditions of the coal mill.

9. The method for analyzing the condition of a coal mill based on an HMM-GMM model according to claim 1, characterized in that, The step of obtaining the likelihood probability of performance degradation of the coal mill based on the quantitative evaluation model includes: The nonlinear multidimensional features of the coal mill operation process are obtained as a dataset; Based on the dataset, the nonlinear characteristics of the coal mill under normal operating conditions are obtained; Based on the quantitative evaluation model, the predicted operating status data of the coal mill is obtained; Based on the predicted operating status data and the nonlinear characteristics of the coal mill under normal operating conditions, a numerical deviation is obtained to determine the likelihood probability of a performance degradation in the coal mill.

10. A method for analyzing the condition of a coal mill based on an HMM-GMM model according to claim 1, characterized in that, The process of obtaining quantitative evaluation indicators of the coal mill's operating status based on the likelihood probability, in order to determine the degree of performance degradation of the coal mill, includes: Obtain the likelihood probability of the coal mill's performance degradation over a period of time, and establish a performance degradation curve; Obtain the values ​​of the quantitative evaluation indicators of the coal mill's operating status within the same time period, and establish a quantitative evaluation indicator curve; By overlapping the performance degradation curve and the quantitative evaluation index curve, the overlap interval is obtained; The area of ​​the overlapping interval is obtained, and the ratio of the area of ​​the region formed by the performance decline curve and the quantitative evaluation index curve is obtained. The ratio represents the degree of performance decline of the coal mill.

Images