Coal mill and coal pulverizing system risk early warning method and system

Abstract

The invention provides a coal mill and coal pulverizing system risk early warning method and system, and belongs to the field of process control state estimation and early warning monitoring. The method comprises the following steps: acquiring historical operation data of a coal mill and a coal pulverizing system; for the current operation data of the coal mill and the coal pulverizing system, extracting a neighborhood subset from the historical operation data based on a clustering algorithm; determining a multi-class abnormal risk probability score matrix and a reference risk score of the neighborhood subset; calculating a correlation coefficient between the reference risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix; obtaining a final risk score based on the current operation data and the primitive anomaly detection algorithm corresponding to the row vector with the maximum correlation coefficient; wherein the multi-class abnormal risk probability score matrix is obtained based on multiple primitive anomaly detection algorithms, and the reference risk score is obtained based on a statistical method. The problem of low efficiency of early warning means in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of process control state estimation and early warning monitoring, and in particular to a coal mill and pulverizing system risk early warning method and system. Background Art

[0002] Equipment assets are the core assets of power plants. Equipment reliability and stability are the greatest guarantee for safe and stable production of power plants. Against the background of continuous technological advancement, management methods and concepts, it has become an industry development consensus to use advanced technologies such as big data and artificial intelligence to achieve intelligent equipment management and control and improve equipment operation stability and reliability.

[0003] The coal mill and pulverizing system is an important auxiliary system of the coal-fired unit. The function of the system is to grind coal powder of suitable quality according to the unit output load to ensure the fuel needs of the boiler for power generation. The high efficiency and safety of the pulverizing system operation are the key to achieving the goals of advanced power generation and smart power generation. Many coal-fired units are equipped with a special pulverizing system automatic start-stop control system APS (Automatic Plant Startup and Shutdown System) to ensure the safe start-stop, adjustment and monitoring operations of this subsystem.

[0004] Coal mill and pulverizing system parameter monitoring and early warning detection are the main tasks of the pre-shift operator, which includes dozens of abnormal monitoring such as valve abnormalities, vibration abnormalities, pressure abnormalities, outlet air-powder mixture temperature abnormalities, coal powder pipeline wall temperature abnormalities, etc., involving signal analysis of up to hundreds of sensors.

[0005] The parameter threshold trigger warning method currently used in the industry is relatively old. On the one hand, it requires the on-duty personnel to analyze dozens or even hundreds of sensor parameters one by one, resulting in a heavy management and control burden, and is prone to omissions in equipment data analysis, untimely fault detection, and even equipment damage. On the other hand, the existing technical means ignore the temporal and spatial dependencies between sensor signals. The isolated analysis of a single signal is not conducive to the early detection of equipment failures and sensitive signals. There is still much room for improvement in the relationship between the time and space characteristics of the APS system of the coal mill and pulverizing system.

[0006] Therefore, in order to reduce the workload of operation and maintenance personnel and improve the warning sensitivity of coal mill and pulverizing system, it is very necessary to study the comprehensive risk warning method of coal mill and pulverizing system based on multi-source sensor information. Summary of the invention

[0007] The purpose of the embodiments of the present invention is to provide a coal mill and pulverizing system risk early warning method and system, so as to at least solve the problem of inefficiency of the early warning means in the prior art.

[0008] In order to achieve the above-mentioned object, the first aspect of the present invention provides a coal mill and pulverizing system risk early warning method, comprising:

[0009] Obtain historical operation data of coal mill and pulverizing system;

[0010] For the current operation data of coal mill and pulverizing system, a neighborhood subset is extracted from the historical operation data based on clustering algorithm;

[0011] Determine a multi-class anomaly risk probability score matrix and a baseline risk score for a neighborhood subset;

[0012] Calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix;

[0013] The final risk score is obtained based on the primitive anomaly detection algorithm corresponding to the current running data and the row vector with the largest correlation coefficient;

[0014] Among them, the multi-category abnormal risk probability score matrix is ​​obtained based on multiple primitive anomaly detection algorithms, and the benchmark risk score is obtained based on statistical methods.

[0015] Optionally, obtain historical operating data of the coal mill and pulverizing system, including:

[0016] Continuously collect and store the operation data of the coal mill and the pulverizing system. Each set of operation data contains a variety of signal data collected by the coal mill and the pulverizing system at that time point. There is a preset time interval between any two adjacent time points.

[0017] Align the acquisition intervals of different sensor signals in the historical operation data to obtain a structured sensor data matrix;

[0018] The sensor data matrix is ​​normalized to obtain structured historical operation data.

[0019] Optionally, the method further comprises:

[0020] Before determining the multi-class abnormal risk probability score matrix and the benchmark risk score of the neighborhood subset, the multi-class abnormal risk probability score matrix and the benchmark risk score of the historical operation data are determined.

[0021] Optionally, the calculation formula for the serious deviation abnormal risk score of the historical operation data of the coal mill and the pulverizing system is as follows:

[0022] Xy i =TS curr -TS i ,i=1,2...n;

[0023]

[0024] Among them, TS curr Represents the current processing data point, Xy i Represents the current processing data point TS curr and historical data points TS i The space vector formed by the difference, θ ij Represents vector Xy i and vector Xy j The cosine similarity between them, i and j represent any two different data points in the unit historical data set S, μ θ Represents the current processing data point TS curr The mean of all possible cosine similarities constructed with the historical dataset S, rs represents the global severe deviation anomaly risk score of the currently processed data point.

[0025] Optionally, the calculation formula for the local deviation abnormal risk score of the historical operation data of the coal mill and the pulverizing system is as follows:

[0026] density(p)=|{q|q∈C p anddist(p,q)≤ε}|

[0027]

[0028] Among them, C p represents the cluster subcluster to which the target data point p belongs, dist(p,q) represents the distance between data point p and data point q, ε is the preset search radius, {q} represents the set of neighboring points of data point p, |{q}| represents the number of data points in the neighboring point set; kNN(p) represents the set of k nearest neighboring points of data point p in the historical data set, k is the preset number of nearest neighboring points to be considered, density(p) and density(q) represent the local density of data point p and data point q, and LOF(p) represents the local deviation anomaly risk score of data point p.

[0029] Optionally, the calculation formula for the semantic deviation anomaly risk score of the historical operation data of the coal mill and the pulverizing system is as follows:

[0030]

[0031] Where T represents the number of isolated trees in the isolation forest, h t (x) represents the path length of data point x in the tth isolation tree, represents the average path length of data point x in the isolation forest, leaves(t) represents the total number of leaf nodes of the tth isolation tree, H is the negative logarithm of the total number of leaf nodes of all isolation trees in the forest, and Score(x) represents the semantic deviation anomaly risk score of data point x.

[0032] Optionally, the multi-class abnormal risk probability score matrix RM(S norm ) is calculated as follows:

[0033]

[0034] Where R represents the number of primitive anomaly detection algorithms applied, n is the number of historical data collection groups; D r represents the specific rth primitive anomaly detection algorithm, Dx(TS i ) represents the quantitative risk score of the unit data corresponding to time point i under the primitive anomaly detection algorithm in the rth order, Represents real space.

[0035] Optionally, the benchmark risk score Tg(S norm ) is calculated as follows:

[0036] Tg(TS i )=max(D 1 (TS i ), D 2 (TS i ), ...D R (TS i ))

[0037]

[0038] Among them, Tg(TS i ) represents the baseline risk score of the unit operation data point corresponding to time point i, Tg(S norm ) represents the baseline risk score corresponding to all historical data of the unit.

[0039] Optionally, the selection formula of the primitive anomaly detection algorithm is as follows:

[0040]

[0041] r match =max r∈R Pearson(TgVec,RM(S nearby )[r,:])

[0042]

[0043] Among them, Pearson(X,Y) is the calculation formula of Pearson correlation coefficient, E(.) represents expectation, r match Indicates the row vector number with the greatest correlation with the benchmark risk score vector TgVec, TS currRepresents the currently processed data point, Indicates based on r match Select the final primitive anomaly detector that better matches the risk anomaly pattern of the unit data point at the current moment, synScore (TS curr ) represents the final risk anomaly score of the current data point.

[0044] A second aspect of the present invention provides a coal mill and pulverizing system risk early warning system, comprising:

[0045] Data acquisition module, used to obtain historical operation data of coal mill and pulverizing system;

[0046] The data extraction module is used to extract neighborhood subsets from historical operation data based on clustering algorithms for the current operation data of the coal mill and pulverizing system;

[0047] A first calculation module is used to determine a multi-class abnormal risk probability score matrix and a baseline risk score for a neighborhood subset;

[0048] The second calculation module is used to calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix;

[0049] The third calculation module is used to obtain the final risk score based on the current running data and the primitive anomaly detection algorithm corresponding to the row vector with the largest correlation coefficient;

[0050] Among them, the multi-category abnormal risk probability score matrix is ​​obtained based on multiple primitive anomaly detection algorithms, and the benchmark risk score is obtained based on statistical methods.

[0051] Through the above technical scheme, a risk warning method for a coal mill and a pulverizing system is provided. The method comprehensively captures and learns various abnormal situations that may exist in the historical operation data of the coal mill and the pulverizing system by applying multiple primitive anomaly detection algorithms, and considers the abnormal scores of the unit data at different time points by type; on this basis, for the unit data at the current moment, a highly similar neighborhood subset is extracted from the historical operation data based on a clustering algorithm, and the abnormal discrimination operation based on the neighborhood subset will more efficiently highlight the specific abnormal mode of the unit at the current time point; further, by calculating the correlation coefficient of the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix, the primitive anomaly detector corresponding to the row vector with the largest correlation coefficient is selected as the best abnormal risk assessment method adapted to the unit data at the current moment, thereby realizing comprehensive and effective quantification of the unit abnormal risk score at the current time point, effectively realizing multi-source information joint detection, comprehensively considering the spatiotemporal dependency between different sensor data, and providing a normalized comprehensive risk score with good readability, which can effectively solve the problems of inefficient early warning means and insufficient fault early warning capabilities in the existing technology.

[0052] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0054] Figure 1 It is a flowchart of the steps of a coal mill and pulverizing system risk early warning method provided by an embodiment of the present invention;

[0055] Figure 2 It is a system structure block diagram of a coal mill and pulverizing system risk warning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0057] Figure 1 The following is a flow chart of a coal mill and pulverizing system risk warning method provided by one embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a risk early warning method for a coal mill and a pulverizing system, the method comprising:

[0058] Step S100: Obtain historical operation data of the coal mill and the pulverizing system.

[0059] There can be one or more groups of historical operation data.

[0060] In some implementations of this embodiment, the historical operating data of the coal mill and the pulverizing system include seven types: coal feed rate feedback signal of the coal feeder, coal mill current, coal mill primary air flow rate, coal mill primary air temperature, coal mill primary air pressure, coal mill inlet and outlet pressure difference, and coal mill outlet powder mixture temperature.

[0061] In some implementations of this embodiment, historical operating data of a coal mill and a pulverizing system are obtained, including: obtaining historical operating data of the coal mill and the pulverizing system within a preset time period in the past; aligning acquisition intervals of different sensor signals in the historical operating data to obtain a structured sensor data matrix; and normalizing the sensor data matrix to obtain structured historical operating data.

[0062] In the above implementation, the sensor data matrix is ​​normalized to eliminate the dimensional differences between different sensor signal features, thereby obtaining historical operation data for use by a subsequent anomaly detection algorithm.

[0063] In some implementations of this embodiment, obtaining historical operation data of the coal mill and the pulverizing system in the past preset time period includes: continuously collecting and storing the operation data of the coal mill and the pulverizing system, each set of operation data includes a variety of signal data collected by the coal mill and the pulverizing system at that time point, and there is a preset time interval between any two adjacent time points. The preset time interval is the time interval for collecting data, for example, 10 seconds.

[0064] In some implementations of this embodiment, multiple groups of historical operation data of the coal mill and the pulverizing system in the past preset time period are obtained, which can be expressed by the following formula:

[0065]

[0066] In the formula, S represents the historical operation data matrix of the unit, n is the number of historical operation data collection groups, and m is the number of signal sensors; TS i represents the sensor signal vector collected by the monitoring system of the coal mill and pulverizing system at time point i; S j represents the historical operating data of the jth sensor, for example, the historical operating data of the powder mixture temperature sensor at the coal mill outlet; s ij It represents the measurement value of the jth sensor at time point i.

[0067] After obtaining multiple sets of historical operating data of the coal mill and pulverizing system in the past preset time period, the Min-Max normalization method is used to normalize the sensor signals in the historical operating data to eliminate the influence of the numerical dimension differences of different sensor signals on the subsequent model, thereby improving the performance of the risk warning model.

[0068] The Min-Max normalization operation is defined by the following formula:

[0069]

[0070] Among them, i represents the time measurement point, j represents the sensor number, and s ij represents the original measurement value of the jth sensor at time point i, s ij norm represents the normalized result of eliminating the dimension characteristics of the sensor, max(S j ) and min(S j ) represents the maximum and minimum values ​​of the historical operating data of sensor No. j.

[0071] Step S200: For the current operation data of the coal mill and the pulverizing system, a neighborhood subset is extracted from the historical operation data based on a clustering algorithm.

[0072] For example, based on a clustering algorithm, a neighborhood subset that is highly similar to the current operating data is extracted from the historical operating data, such as a neighborhood subset that is most similar to the current operating data.

[0073] In some implementations of this embodiment, the KNN (K Nearest Neighbor) method is used to find m neighboring points of a sample from historical operation data, and the distance between samples is calculated using the Euclidean distance, where the Euclidean distance is calculated as follows:

[0074] dis(i,j)=||X i -X j || 2 ;

[0075] Among them, X i and X j represents any two data points.

[0076] According to the minimum distance principle, the neighborhood subset S of the current running data is extracted nearby .

[0077] In this embodiment, a neighborhood subset of the operation data of the coal mill and the pulverizing system at the current moment is extracted in a targeted manner, so that the abnormal risks of the coal mill and the pulverizing system can be evaluated more accurately.

[0078] Step S300. Determine a multi-class anomaly risk probability score matrix and a benchmark risk score for a neighborhood subset, wherein the multi-class anomaly risk probability score matrix is ​​obtained based on a plurality of primitive anomaly detection algorithms, and the benchmark risk score is obtained based on a statistical method.

[0079] In some implementations of this embodiment, first, by applying multiple primitive anomaly detection algorithms, various abnormal situations that may exist in the operating data of the coal mill and the pulverizing system are comprehensively captured and learned. From the perspective of fault diagnosis theory, the abnormal situations can be divided into three categories, namely, severe deviation, local deviation and semantic deviation.

[0080] The risk probabilities of the above three types of abnormal situations are quantitatively characterized by applying a variety of targeted primitive anomaly detection algorithms. In this embodiment, the angle-based outlier detection algorithm (ABOD: Angle-Based Outlier Detection) is used for severe deviations, the CBLOF (Clustering-Based Local Outlier Factor) detection algorithm is used for local deviations, and the isolation forest (iForest: Isolation Forest) detection algorithm is used for semantic deviations.

[0081] Define the primitive anomaly detector combination Detectors:

[0082] Detectors=[D 1 ,D 2 ,...D R ];

[0083] Where R represents the number of primitive anomaly detection algorithms applied; D 1 ,D 2 ,...D R They respectively represent a primitive anomaly detection algorithm, such as the CBLOF anomaly detection algorithm.

[0084] The statistical method for calculating the benchmark risk score for the neighborhood subset is the same as the statistical method for calculating the benchmark risk score for the historical operating data.

[0085] In some implementations of this embodiment, the primitive anomaly detector combination Detectors is applied to the neighborhood subset S nearby , and obtain the multi-class abnormal risk probability score matrix RM(S nearby ) and the baseline risk score vector TgVec, defined as follows:

[0086]

[0087] Where R represents the number of primitive anomaly detection algorithms applied, m is the preset number of nearest neighbor data points to be considered, and D r represents the specific rth primitive anomaly detection algorithm, and p represents any one of the neighborhood subsets S nearby data point, Tg(.) represents the baseline risk score calculation method defined in step S200.

[0088] Step S400: Calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix.

[0089] Step S500: Obtain the final risk score based on the primitive anomaly detection algorithm corresponding to the current running data and the row vector with the largest correlation coefficient.

[0090] Specifically, the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class anomaly risk probability score matrix is ​​calculated, the primitive anomaly detection algorithm corresponding to the row vector with the largest correlation coefficient is applied to the operating data at the current moment, and the risk score of the primitive anomaly detection algorithm is output as the final result.

[0091] In some implementations of this embodiment, the baseline risk score vector TgVec and the multi-class abnormal risk probability score matrix RM(S) of the neighborhood subset are obtained for the current moment operation data. nearby ), calculate the baseline risk score vector TgVec and the multi-class abnormal risk probability score matrix RM(S nearby ) The correlation coefficient of each row vector is selected, and the primitive anomaly detector corresponding to the row vector with the largest correlation coefficient is selected as the best anomaly risk assessment method for the current running data activation adaptation, which is defined as follows:

[0092]

[0093] r match =max r∈R Pearson(TgVec,RM(S nearby )[r,:])

[0094]

[0095] Among them, Pearson(X,Y) is the calculation formula of Pearson correlation coefficient, E(.) represents expectation, r match Indicates the row vector number with the greatest correlation with the benchmark risk score vector TgVec, TS curr Represents the current processing data point, D rmatch Indicates based on r match Select the final primitive anomaly detector that better matches the risk anomaly pattern of the unit data point at the current moment, synScore (TS curr ) represents the final risk anomaly score of the current data point.

[0096] In some implementations of this embodiment, the method further includes: before determining the multi-class abnormal risk probability score matrix and the baseline risk score of the neighborhood subset, determining the multi-class abnormal risk probability score matrix and the baseline risk score of the historical operation data.

[0097] By applying multiple primitive anomaly detection algorithms, a multi-category anomaly risk probability score matrix of historical operation data is obtained, and the baseline risk score of all historical operation data is obtained based on statistical methods, realizing model training based on joint detection of multi-source information.

[0098] In these implementations, by calculating the multi-class abnormal risk probability score matrix and the benchmark risk score of the historical operation data, the final risk score of the current operation data and the score of the historical operation data can be displayed on the operation and maintenance team GUI interface in the form of a continuous curve, so as to better present the "whole process" of the gradual decline of the unit. After calculating the multi-class abnormal risk probability score matrix and the benchmark risk score of the historical operation data, the multi-class abnormal risk probability score matrix and the benchmark risk score of the neighborhood subset can be obtained through index query without repeated calculation.

[0099] In some implementations of this embodiment, first, by applying multiple primitive anomaly detection algorithms, various abnormal situations that may exist in the operating data of the coal mill and the pulverizing system are comprehensively captured and learned. From the perspective of fault diagnosis theory, the abnormal situations can be divided into three categories, namely, severe deviation, local deviation and semantic deviation.

[0100] The risk probabilities of the above three types of abnormal situations are quantitatively characterized by applying a variety of targeted primitive anomaly detection algorithms. In this implementation, the angle-based outlier detection algorithm (ABOD: Angle-Based Outlier Detection) is used for severe deviations, the CBLOF (Clustering-Based Local Outlier Factor) detection algorithm is used for local deviations, and the iForest (Isolation Forest) detection algorithm is used for semantic deviations. The historical operation data TS at different time points is quantitatively evaluated by type. i The anomaly score.

[0101] Define the primitive anomaly detector combination Detectors:

[0102] Detectors=[D 1 ,D 2 ,...D R ]

[0103] Where R represents the number of primitive anomaly detection algorithms applied; D 1 ,D 2 ,...D R They respectively represent a primitive anomaly detection algorithm, such as the CBLOF anomaly detection algorithm.

[0104] Specifically, in this step, the ABOD algorithm is used to quantitatively evaluate the serious deviation of the current operating data.

[0105] In high-dimensional space, traditional distance-based anomaly discrimination methods may fail because the distances between all data points tend to be consistent, losing the ability to measure differences. ABOD is a method that uses the angular relationship between data points to identify outliers. It can provide a measurement method that is not affected by scale. The formula is defined as follows:

[0106] Xy i =TS curr -TS i ,i=1,2...n;

[0107]

[0108] Among them, TS curr Represents the current processing data point, Xy i Represents the current processing data point TS curr and historical data points TS i The space vector formed by the difference, θ ij Represents vector Xy i and vector Xy j The cosine similarity between them, i and j represent any two different data points in the historical running data set S, μ θ Represents the current processing data point TS curr The mean of all possible cosine similarities constructed with the historical running dataset S, rs represents the global severe deviation anomaly risk score of the currently processed data point.

[0109] Specifically, in this step, the CBLOF algorithm is used to quantitatively evaluate the local deviation of the current operating data.

[0110] The CBLOF algorithm combines the concepts of clustering and local anomaly factor. It first determines the intrinsic structure of the data set through clustering and divides the data set into multiple clusters. This makes it more efficient to detect local deviation anomalies within clusters of similar data points. Then, within each cluster, the local deviation of the target data point is quantitatively characterized by measuring the local density deviation between the target data point and its neighboring points.

[0111] Preferably, the KMeans clustering algorithm is used to divide the data set into subclusters. The KMeans clustering algorithm is divided into an assignment step and an update step, wherein the assignment step is to calculate the distance between each data point and all the centroids and assign the point to the cluster represented by the nearest centroid; the update step is to calculate the mean of all the points in each cluster and assign it to the centroid of the current cluster. Repeat the assignment step and the update step until the clustering result meets the stop condition. The KMeans algorithm execution process is defined as follows:

[0112]

[0113] Among them, μ j represents the centroid of the set of data points in cluster(j), |cluster(j)| is the number of data points in cluster(j), and x is the number of data points in cluster(j); i is the coordinate of the data point x in the i-th dimension, μ ji is the center of mass μ j The coordinate in the i-th dimension; K is a preset value representing the number of clusters to be clustered, and J is the total internal cluster distance, which is also the objective function to be minimized by the clustering algorithm iteration.

[0114] Preferably, the LOF local anomaly factor is used to calculate the local anomaly risk score of the target data point, which is defined as follows:

[0115] density(p)=|{q|q∈C p anddist(p,q)≤ε}|

[0116]

[0117] Among them, C p represents the cluster subcluster to which the target data point p belongs, dist(p,q) represents the distance between data point p and data point q, ε is the preset search radius, {q} represents the set of neighboring points of data point p, |{q}| represents the number of data points in the neighboring point set; kNN(p) represents the set of k nearest neighboring points of data point p in the historical running data set, k is the preset number of nearest neighboring points to be considered, density(p) and density(q) represent the local density of data point p and data point q, and LOF(p) represents the local deviation abnormality risk score of data point p.

[0118] Specifically, in this step, an isolation forest (iForest: Isolation Forest) detection algorithm is used to quantitatively evaluate the semantic deviation of the current running data.

[0119] The core idea of ​​iForest is to use random partitioning to isolate data points. For the isolation tree of the isolation forest, the "path length" of a data point is defined as the number of layers from the root of the isolation tree to the leaf node where the data point cannot be further isolated. In normal data distribution, the path length of a data point is usually longer than that of an abnormal data point. This is because abnormal data points are relatively isolated in the feature space and are more easily isolated by the random partitioning of the iForest algorithm. Therefore, it is reasonable to use the iForest algorithm to calculate the semantic anomaly risk score of the data point, which is defined as follows:

[0120]

[0121] Where T represents the number of isolated trees in the isolation forest, h t (x) represents the path length of data point x in the tth isolation tree, represents the average path length of data point x in the isolation forest, leaves(t) represents the total number of leaf nodes of the tth isolation tree, H is the negative logarithm of the total number of leaf nodes of all isolation trees in the forest, and Score(x) represents the semantic deviation anomaly risk score of data point x.

[0122] Apply the primitive anomaly detector combination Detectors to the historical operation data S processed in step S100 norm , the multi-class abnormal risk probability score matrix RM(S norm ) is defined as follows:

[0123]

[0124] Where R represents the number of primitive anomaly detection algorithms applied, n is the number of historical operation data collection groups; D r represents the specific rth primitive anomaly detection algorithm, D r (TS i ) represents the quantitative risk score of the unit data corresponding to time point i under the primitive anomaly detection algorithm in the rth order.

[0125] Then, the TS at each time point is obtained by taking the mean, median or maximum value and other statistical methods. i The benchmark risk score Tg(TS i ).

[0126] Preferably, the benchmark risk score Tg(S) of all historical operation data of the unit is set by selecting the maximum value. norm ), the process can be expressed by the following formula:

[0127] Tg(TS i )=max(D 1 (TS i ), D 2 (TS), ...D R (TS i ))

[0128]

[0129] Among them, Tg(TS i ) represents the baseline risk score of the coal mill and pulverizing system operation data point corresponding to time point i, Tg(S norm ) represents the baseline risk score corresponding to all historical operating data of the unit.

[0130] The solution and method provided in this embodiment provide a comprehensive and effective solution that takes abnormal modes into consideration for parameter monitoring and risk warning of coal grinding units and pulverizing systems. It not only effectively realizes multi-source information joint detection and comprehensively considers the temporal and spatial dependencies between different sensor data, but also provides a normalized comprehensive risk score with good readability, which can effectively solve the problems of inefficiency of existing warning methods and insufficient fault warning capabilities.

[0131] Figure 2 1 is a block diagram of a coal mill and pulverizing system risk warning system provided by an embodiment of the present invention. Figure 2 As shown, an embodiment of the present invention provides a risk early warning system for a coal mill and a pulverizing system, the system comprising:

[0132] Data acquisition module, used to obtain historical operation data of coal mill and pulverizing system;

[0133] The data extraction module is used to extract neighborhood subsets from historical operation data based on clustering algorithms for the current operation data of the coal mill and pulverizing system;

[0134] A first calculation module is used to determine a multi-class abnormal risk probability score matrix and a baseline risk score for a neighborhood subset;

[0135] The second calculation module is used to calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix;

[0136] The third calculation module is used to obtain the final risk score based on the current running data and the primitive anomaly detection algorithm corresponding to the row vector with the largest correlation coefficient;

[0137] Among them, the multi-category abnormal risk probability score matrix is ​​obtained based on multiple primitive anomaly detection algorithms, and the benchmark risk score is obtained based on statistical methods.

[0138] The embodiment of the present invention further provides a computer-readable storage medium, on which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned coal mill and pulverizing system risk warning method.

[0139] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor (processor) perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0140] The optional embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical scheme of the embodiments of the present invention can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0141] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A coal mill and pulverizing system risk early warning method, characterized in that: include: Obtain historical operation data of coal mill and pulverizing system; For the current operation data of coal mill and pulverizing system, a neighborhood subset is extracted from the historical operation data based on clustering algorithm; Determine a multi-class anomaly risk probability score matrix and a baseline risk score for a neighborhood subset; Calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix; The final risk score is obtained based on the primitive anomaly detection algorithm corresponding to the current running data and the row vector with the largest correlation coefficient; Among them, the multi-category abnormal risk probability score matrix is ​​obtained based on multiple primitive anomaly detection algorithms, and the benchmark risk score is obtained based on statistical methods.

2. A coal mill and pulverizing system risk early warning method according to claim 1, characterized in that: Obtain historical operating data of coal mill and pulverizing system, including: Continuously collect and store the operation data of the coal mill and the pulverizing system. Each set of operation data contains a variety of signal data collected by the coal mill and the pulverizing system at that time point. There is a preset time interval between any two adjacent time points. Align the acquisition intervals of different sensor signals in the historical operation data to obtain a structured sensor data matrix; The sensor data matrix is ​​normalized to obtain structured historical operation data.

3. A coal mill and pulverizing system risk early warning method according to claim 1, characterized in that: The method further comprises: Before determining the multi-class abnormal risk probability score matrix and the benchmark risk score of the neighborhood subset, the multi-class abnormal risk probability score matrix and the benchmark risk score of the historical operation data are determined.

4. A coal mill and pulverizing system risk early warning method according to claim 3, characterized in that: The calculation formula for the serious deviation abnormal risk score of the historical operation data of the coal mill and pulverizing system is as follows: Xy i =TS curr -TS i ,i=1,2...n; Among them, TS curr Represents the current processing data point, Xy i Represents the current processing data point TS curr and historical data points TS i The space vector formed by the difference, θ ij Represents vector Xy i and vector Xy j The cosine similarity between them, i and j represent any two different data points in the unit historical data set S, μ θ Represents the current processing data point TS curr The mean of all possible cosine similarities constructed with the historical dataset S, rs represents the global severe deviation anomaly risk score of the currently processed data point.

5. A coal mill and pulverizing system risk early warning method according to claim 3, characterized in that: The calculation formula for the local deviation abnormal risk score of the historical operation data of the coal mill and pulverizing system is as follows: density(p)=|{q|q∈C p anddist(p,q)≤ε}| Among them, C p represents the cluster subcluster to which the target data point p belongs, dist(p,q) represents the distance between data point p and data point q, ε is the preset search radius, {q} represents the set of neighboring points of data point p, |{q}| represents the number of data points in the neighboring point set; kNN(p) represents the set of k nearest neighboring points of data point p in the historical data set, k is the preset number of nearest neighboring points to be considered, density(p) and density(q) represent the local density of data point p and data point q, and LOF(p) represents the local deviation anomaly risk score of data point p.

6. A coal mill and pulverizing system risk early warning method according to claim 3, characterized in that: The calculation formula for the semantic deviation anomaly risk score of the historical operation data of the coal mill and pulverizing system is as follows: Where T represents the number of isolated trees in the isolation forest, h t (x) represents the path length of data point x in the tth isolation tree, represents the average path length of data point x in the isolation forest, leaves(t) represents the total number of leaf nodes of the tth isolation tree, H is the negative logarithm of the total number of leaf nodes of all isolation trees in the forest, and Score(x) represents the semantic deviation anomaly risk score of data point x.

7. A coal mill and pulverizing system risk early warning method according to claim 3, characterized in that: The multi-class abnormal risk probability score matrix RM(S) of the historical operation data of the coal mill and pulverizing system norm ) is calculated as follows: Where R represents the number of primitive anomaly detection algorithms applied, n is the number of historical data collection groups; D r represents the specific rth primitive anomaly detection algorithm, D r (TS i ) represents the quantitative risk score of the unit data corresponding to time point i under the primitive anomaly detection algorithm in the rth order.

8. A coal mill and pulverizing system risk early warning method according to claim 3, characterized in that: The benchmark risk score Tg(S norm ) is calculated as follows: Tg(TS i )=max(D1(TS i ),L2(TS i ),...D R (TS i )) Among them, Tg(TS i ) represents the baseline risk score of the unit operation data point corresponding to time point i, Tg(S norm ) represents the baseline risk score corresponding to all historical data of the unit.

9. A coal mill and pulverizing system risk early warning method according to claim 1, characterized in that: The selection formula of the primitive anomaly detection algorithm is as follows: r match =max r∈R Pearson(TgVec,RM(S nearby )[r,:]) Among them, Pearson(X,Y) is the calculation formula of Pearson correlation coefficient, E(.) represents expectation, r match Indicates the row vector number with the greatest correlation with the benchmark risk score vector TgVec, TS curr Represents the current processing data point, D rmatch Represents based on r match Select the final primitive anomaly detector that better matches the risk anomaly pattern of the unit data point at the current moment, synScore (TS curr ) represents the final risk anomaly score of the current data point.

10. A coal mill and pulverizing system risk early warning system, characterized in that: include: Data acquisition module, used to obtain historical operation data of coal mill and pulverizing system; The data extraction module is used to extract neighborhood subsets from historical operation data based on clustering algorithms for the current operation data of the coal mill and pulverizing system; A first calculation module is used to determine a multi-class abnormal risk probability score matrix and a baseline risk score for a neighborhood subset; The second calculation module is used to calculate the correlation coefficient between the baseline risk score vector of the neighborhood subset and each row vector of the multi-class abnormal risk probability score matrix; The third calculation module is used to obtain the final risk score based on the current running data and the primitive anomaly detection algorithm corresponding to the row vector with the largest correlation coefficient; Among them, the multi-category abnormal risk probability score matrix is ​​obtained based on multiple primitive anomaly detection algorithms, and the benchmark risk score is obtained based on statistical methods.

Images