Thermal power plant primary fan health degree evaluation method based on data fusion

Abstract

The invention provides a thermal power plant primary fan health degree assessment method based on data fusion, which comprises the following steps: after data cleaning and steady state screening are carried out on historical data, sampling is carried out on the whole by adopting a probability density equal interval sampling method, and a state vector on each interval is determined by adopting fuzzy clustering; according to the method, typical state vectors are selected to form a memory matrix so as to effectively cover a common working space, a k-nearest neighbor algorithm is adopted to realize dynamic updating of the memory matrix, data with relatively new equipment is used to replace data with relatively poor quality in a historical training set, and the online adaptive capability of the model is improved. A traditional residual threshold method is replaced by sliding window state similarity to serve as an evaluation standard of early warning, a similarity judgment mechanism and a health degree calculation index are given, and the accuracy and timeliness of fault early warning are improved. The system has good adaptability and expandability, and has relatively high application and popularization prospects.

Description

Technical Field

[0001] The invention relates to the technical field of health assessment of primary fans in thermal power plants, and in particular to a health assessment method of primary fans in thermal power plants based on data fusion. Background Art

[0002] The primary fan is an important large auxiliary machine in the air and smoke system of the traditional coal-fired unit boiler. Its main function is to provide primary air for boiler combustion. In a pulverized coal boiler, the primary fan can provide the power source of the primary air, adjust the combustion conditions, and maintain the stable operation of the pulverizing system. The function of the primary air is to carry the pulverized coal into the furnace for combustion. It transports air to the pulverizing system so that the pulverized coal can be blown into the furnace through pipelines and other conveying devices under the carry of the primary air. By adjusting the air volume and air pressure of the primary fan, the amount of primary air entering the furnace can be controlled. The appropriate amount of primary air is crucial for the stable combustion of pulverized coal. In the pulverizing system, the primary air also plays the role of drying and conveying pulverized coal. The working environment of the primary fan is harsh and the probability of failure is high. As an important auxiliary machine, the operating status of the primary fan directly affects the safety and economy of the power plant unit operation. The early health evaluation and fault warning system can help operators detect faults at an early stage and prevent further fault development.

[0003] Traditionally, data-driven health evaluation and fault warning identification methods require a large amount of fault data for training, and even need to divide the fault data into different faults. However, in the actual primary fan fault diagnosis scenario, it is difficult to collect and classify a large amount of fault data. On the one hand, it is often difficult to obtain enough fault samples. In actual industrial scenarios, most primary fans operate in a healthy state, and the operating frequency in a fault state is very low, so there are fewer fault sample data. On the other hand, the required prior knowledge of the fault mechanism and the transfer calculation between equipment states are too complicated. Most of the existing methods in the industry require modeling of multiple fault-related variables of the primary fan separately, which is too complicated for industrial applications. Over time, the primary fan equipment will gradually age, and the initially constructed fault diagnosis model is based on the initial state of the equipment or data within a certain period of time. When the equipment ages, the newly emerging fault features may not be included in the original model. The operating environment of the primary fan is complex and dynamically changing. Changes in these environmental factors will lead to changes in fault types and fault probabilities, so the fault diagnosis model needs to be updated to adapt to the new operating environment. In addition, power plants may adopt new technologies or upgrade equipment. The data characteristics and operating modes brought by these new technologies and new equipment are different from the data and modes based on the original model. Updating the fault diagnosis model can enable the model to make full use of new data information. Traditionally, health evaluation and fault warning models based solely on historical data are difficult to adapt to actual industrial application scenarios. It is necessary to solve the problem of online adaptive updating of the model to further improve the real-time, accuracy and reliability of health evaluation and fault warning methods.

[0004] The comparison with the existing technology is as follows:

[0005] Technical comparison with patent CN112067335A "A method for early warning of power plant blower failure based on multivariate state estimation"

[0006] The evaluation index of patent CN112067335A is mainly concentrated on the deviation of historical data, and the fault warning is determined by the deviation function. The evaluation index of this patent not only includes the health score, but also uses the sliding window state similarity to replace the traditional residual threshold method, taking into account various working conditions and environmental impacts. There is an essential difference between the two in the idea of ​​indicator setting.

[0007] Patent CN112067335A uses a memory matrix built based on historical data, relies on deviation thresholds to achieve early warning, and emphasizes quantitative calculation; while this patent uses fuzzy clustering and k-nearest neighbor algorithms to achieve dynamic updates, combining quantitative and qualitative evaluations to form a comprehensive health assessment standard. The two are essentially different in overall evaluation ideas and implementation methods.

[0008] Patent CN112067335A is mainly applied to power plant blowers, focusing on the fault warning of blowers; while this patent is applied to the health assessment and fault warning of primary blowers of coal-fired units in thermal power plants, focusing on the overall operating status of the equipment. There is an essential difference between the two in the evaluation objects.

[0009] Technical comparison with patent CN118410365A "Fan fault monitoring and diagnosis method based on online learning and multivariate state estimation"

[0010] The evaluation indicators of patent CN118410365A are mainly focused on the preprocessing and monitoring of fault samples, emphasizing the processing of data outliers; while the evaluation indicators of this patent cover a wider range of fields, including equipment health scores, state similarity, etc., forming a more comprehensive evaluation system. There is an essential difference between the two in the scope of indicators.

[0011] Patent CN118410365A uses online learning and DBSCAN algorithm for data processing, and uses adaptive similarity threshold for fault warning, emphasizing real-time performance and data accuracy; while this patent uses improved memory matrix and k-nearest neighbor algorithm to achieve dynamic update, combined with sliding window method for state similarity assessment, focusing on online adaptive capabilities. There is an essential difference between the two in terms of technical implementation.

[0012] Patent CN118410365A is mainly used for fan fault monitoring, focusing on the diagnosis and monitoring of fans; while this patent is applicable to primary fans of coal-fired units in thermal power plants, focusing on overall health assessment and fault warning. The two are essentially different in application fields and target objects. Summary of the invention

[0013] In order to solve the above technical problems, the present invention proposes a health assessment method for primary fans in thermal power plants based on data fusion, aiming to overcome the shortcomings of existing methods and improve the operating safety and reliability of coal-fired power plants.

[0014] To achieve the above object, the technical solution adopted by the present invention is:

[0015] The multivariate state estimation technology (MSET) does not require an explicit model training process, but only uses historical normal data to build a memory matrix, and then calculates the similarity between the current state and the normal state. It has the advantages of low training cost, simple program, and strong generalization ability. The present invention uses the multivariate state estimation technology to perform a health assessment and fault warning of a wind turbine.

[0016] The estimation accuracy and computational efficiency of the MSET model depend on the memory matrix, which should consist of a sufficient number of historical vectors that are similar to the states of the observation vectors. This goal is usually achieved by increasing the size of the memory matrix, but a large memory matrix will have an adverse effect on the computational speed of matrix multiplication and nonlinear operations. In addition, too many samples with low similarity to the observation vector will lead to increased information redundancy. This in turn reduces the accuracy of the estimate. In theory, the most ideal matrix consists of all normal operation history data, but in practical applications, the amount of normal operation history data is large, and the large size of the matrix will lead to huge computational complexity, high computer performance requirements, and long computational time, which cannot meet the real-time requirements of the industry. However, the process memory matrix constructed in most MSET applications is composed of historical observation vectors, without a specific sampling method or using equal interval sampling method to select, and it is impossible to ensure that the sampling density of typical operating states is sufficient. In addition, the random sampling technique involves randomly selecting some data from historical data to construct the memory matrix, but this method is highly random and cannot guarantee coverage of all states. In addition, due to changes in operating status and frequent maintenance that cause changes in objects, the accuracy of MSET estimation using a fixed memory matrix may be reduced, making it difficult to estimate equipment under various operating conditions.

[0017] Therefore, the memory matrix should cover a typical normal operating state space and minimize the size of the matrix. The probability density can reflect the probability of the variable value in a certain area. Selecting the state vector of the process memory matrix according to the probability density can effectively cover the common working space. The prediction accuracy of MSET essentially depends on historical data similar to the observed data. Therefore, effectively constructing the process memory matrix based on probability characteristics is conducive to improving the prediction accuracy. Considering that the unit load can represent the working state of the equipment, the probability density is sampled at equal intervals based on the load. When selecting the state vectors that form the improved process memory matrix D, the purpose is to remove redundant or similar state vectors; the probability density equal interval sampling method is used to sample the population, and fuzzy clustering is used to determine the state vectors in each interval, and typical state vectors are selected to form the memory matrix.

[0018] The present invention provides a health assessment method for a primary fan in a thermal power plant based on data fusion. After data cleaning and steady-state screening of historical data, a probability density equal-interval sampling method is used to sample the population, and a state vector of a process memory matrix is ​​selected to effectively cover a common working space. Fuzzy clustering is used to determine the state vector on each interval, and a typical state vector is selected to form a memory matrix. Newer data of the equipment is used to replace data of poor quality in a historical training set, thereby improving the online adaptive ability of the model. The method comprises the following steps:

[0019] S1: After dimensionality reduction, denoising and data cleaning of historical data, steady-state data is screened to provide a data set for the model;

[0020] S2: Taking the maximum and minimum values ​​of the unit load as the upper and lower limits, the filtered data is divided into M equal intervals, recorded as [L1, L2, ..., L i , …L M ];

[0021] S3: Use the probability density equal interval sampling method to sample the population, integrate the probability density on each equal interval, and obtain the load value of the state vector in the load interval L i The probability P i , multiplied by the total number of state vectors K in the process memory matrix, to obtain the number of state vectors K in each load interval i , K i =P i ×K;

[0022] S4: At each load interval L i Fuzzy clustering is used to determine the state vector X on each interval i , set the load interval L i The optimal number of clusters is set as K i ;

[0023] S5: Each load interval L i The selected state vector X i Add to the process memory matrix D until the number of Xi reaches Ki;

[0024] S6: Collect the primary wind turbine data in real time, and calculate the estimated vector X based on the observed vector and the acquired memory matrix est , the sliding window method is used to calculate the overall similarity of the vector. The overall similarity is the weighted mean of the basic similarity and the state similarity. Considering the stability of the primary fan equipment operation and the frequency of similarity changes, the current health of the primary fan is calculated. The health is the weighted mean of the overall similarity and the similarity fluctuation degree;

[0025] S7: The memory matrix is ​​updated regularly based on K nearest neighbors.

[0026] As a further improvement of the present invention, the steady-state data screening in step S1 is performed based on the autocorrelation function, and the calculation formula is:

[0027]

[0028] Among them, X t is the observed value of the time series, is the mean of the observed values, and k is the lag order.

[0029] As a further improvement of the present invention, the dimensionality reduction process in step S1 is implemented based on the incremental PCA algorithm. When faced with large-scale data sets, the traditional PCA has a huge amount of calculation because it needs to process the entire data set at one time. Incremental PCA is a method that can process streaming data or large-scale data. It only processes one or a small batch of data samples at a time and gradually updates the principal components. For example, when processing real-time primary fan sensor data, the data is generated continuously, and incremental PCA can perform dimensionality reduction processing on these new data in real time.

[0030] The steps include:

[0031] S101: Initialization

[0032] Standardize the first batch of data so that the mean of each feature is 0 and the variance is 1;

[0033] Calculate the covariance matrix C of the first batch of data. Suppose the matrix of the first batch of data is X1, the number of samples is n1, and the number of features is d, then the covariance matrix

[0034] Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1,λ2,…,λ d and the corresponding eigenvectors v1,v2,…,v d . Sort the eigenvalues ​​from large to small, and the corresponding eigenvectors are also sorted accordingly;

[0035] Select the eigenvectors corresponding to the first k largest eigenvalues ​​as the initial principal components, where k is the preset dimension after dimensionality reduction;

[0036] S102: Incremental Update

[0037] Let the new sample be x. When a new sample arrives, first calculate the new sample mean in is the mean of the previous n-1 samples, and updates the covariance moment Among them C old is the previous covariance matrix estimate;

[0038] S103: Update principal components

[0039] Perform eigenvalue decomposition on the updated covariance matrix, sort the new eigenvalues ​​from large to small, sort the corresponding eigenvectors accordingly, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the updated principal components;

[0040] S104: Repeated Update

[0041] As new data continues to arrive, the new data is continuously processed, steps S102 and S103 are repeated, and the covariance matrix and principal components are continuously updated, thereby realizing incremental principal component analysis of the data.

[0042] As a further improvement of the present invention, the probability density in step S3 adopts the formula of normal distribution:

[0043]

[0044] Here, μ is the mean and σ is the standard deviation.

[0045] The method of dynamically updating the memory matrix based on K nearest neighbors is as follows: for the selected observation vector as the distance to the sample to be classified, the Euclidean distance between the selected observation vector and all vectors in the training set is calculated, the calculated Euclidean distance is sorted, and the k training samples closest to the sample to be classified are found. The categories of these k neighbors are counted, and the category of the sample to be classified is determined by voting, that is, it belongs to [L1, L2, ..., L i , …L M ] interval L i .

[0046] As a further improvement of the present invention, the detailed algorithm steps of the fuzzy clustering in step S4 are as follows:

[0047] S401: Initialization: Select the number of clusters C and initialize the fuzzy membership matrix U, where U ij Represents data point x j For the membership of cluster i, the membership value of each data point should be between [0,1], and the sum of all memberships for each data point j is 1:

[0048] S402: Calculate cluster centers: Calculate the center V of each cluster based on the current membership matrix U i :

[0049]

[0050] Where m is the fuzzy factor, and its value is between (1,2);

[0051] S403: Update the membership matrix: Calculate each data point x j The membership degree U of each cluster i ij ;

[0052]

[0053] S404: Determine the convergence, calculate the change of the membership matrix, and if the stop condition is met, such as the change is less than the set threshold, or the maximum number of iterations is reached, stop the iteration; otherwise, return to step S402;

[0054] S405: Output the results, output the final cluster center V and fuzzy membership matrix U.

[0055] As a further improvement of the present invention, the estimated vector X output by the model in step S6 est , obtained by the weight vector W, which is the similarity between the current observation vector and the historical data memory matrix;

[0056]

[0057] Finally, the estimated vector X est The expression is as follows:

[0058]

[0059] As a further improvement of the present invention, the overall similarity in step S6 is a weighted average of the basic similarity and the state similarity; the basic similarity refers to the observation vector X obs With the estimated vector X est The closeness is as follows:

[0060]

[0061] S(X o ,X e )∈(0,1]

[0062] Where α k represents the weight of the kth variable's contribution to fault warning, α k >0,S(X o ,X e )∈(0,1], the larger the S is, the closer the two are;

[0063] State similarity refers to the observation vector X obs The degree of closeness to the memory matrix is ​​as follows:

[0064]

[0065] S(X o ,D)∈(0,1], the larger the S is, the closer the two are;

[0066] The overall similarity is the weighted average of the basic similarity and the state similarity:

[0067] S=β1S(X o ,X e )+β2S(X o ,D)

[0068] β1∈(0,1),β2∈(0,1)

[0069] S∈(0,1],

[0070] The larger the S is, the higher the similarity between the primary fan and the historical normal operating state is.

[0071] Considering the stability of primary fan equipment operation and the frequency of similarity changes, the health degree is designed as follows:

[0072]

[0073] α1∈(0,1),α2∈(0,1)

[0074] H∈(0,1]

[0075] and Respectively represent the change in overall similarity at time t and the total overall similarity

[0076] For the real-time collected primary fan data, the sliding window method is used to eliminate the uncertainties and random interference in the equipment operation. The window width is set to N, and the overall similarity is calculated once per cycle. The overall health data [H1, H2, L, H N ], calculate the moving average of N consecutive health values ​​in the window:

[0077] The fault warning threshold Ht is determined based on the minimum average health value in the sliding window. When the device is operating normally, the minimum health value is calculated based on the minimum average similarity between the normal observation vector and the MSET model estimation vector as Ht. N , the fault warning threshold is: H AN =kH N , k is less than 1, which is the alarm threshold coefficient and is determined by the operator based on operating experience.

[0078] As a further improvement of the present invention, step S6 determines the current health of the primary wind turbine. If the early warning model determines that the primary wind turbine is faulty, the early warning signal will be promptly notified to the power station staff so that corresponding preventive measures can be taken. At the same time, the early warning information will be uploaded to the monitoring system of the power station to realize the integration of the early warning system and the monitoring system of the power station. The data uploading method includes but is not limited to HttpAPI interface, WebSocket interface, and IEC104 interface.

[0079] Beneficial effects: The present invention has the following advantages:

[0080] The present invention provides a method and system for evaluating the health of a primary fan of a coal-fired unit in a large thermal power plant based on improved multivariate state estimation. The primary fan needs a large amount of fault data for training, needs to divide the fault data into different faults, relies on the limitations of a single variable threshold for fault evaluation, and the online update problem of the fault diagnosis model. After data cleaning and steady-state screening of historical data, the population is sampled using a probability density equal-interval sampling method, fuzzy clustering is used to determine the state vector on each interval, typical state vectors are selected to form a memory matrix, and the k-nearest neighbor algorithm is used to update the memory matrix. The newer data of the equipment is used to replace the poor quality data in the historical training set, which improves the online adaptive ability of the model. The sliding window state similarity is used instead of the traditional residual threshold method as the evaluation standard for early warning. The present invention adopts an improved multivariate state estimation method, which has strong robustness to noise and can effectively improve the accuracy and reliability of primary fan state monitoring. By comprehensively analyzing multiple operating parameters and establishing a multivariate state estimation model, the improved multivariate state estimation technology can estimate and predict the state of the primary fan in real time, more accurately reflect the actual health status of the primary fan, reduce the risk of misjudgment and missed judgment, and provide operators with timely and accurate equipment status information. By setting the health score threshold, the system can issue an early warning before a fault occurs to ensure the safe operation and economic benefits of the power plant. The system has good adaptability and scalability, and can be adjusted and optimized according to different types of primary fans and operating environments. It is suitable for coal-fired units in various large thermal power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic diagram of the process of the primary fan fault early warning method in the present invention;

[0082] Figure 2 It is a schematic diagram of building a health assessment model in the present invention. DETAILED DESCRIPTION

[0083] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0084] A method for evaluating the health of a primary fan in a thermal power plant based on data fusion of the present invention comprises the following steps:

[0085] The memory matrix should cover a typical normal operating state space and minimize the size of the matrix. The probability density can reflect the probability of the variable value within a certain area. Selecting the state vector of the process memory matrix based on the probability density can effectively cover the common working space. The prediction accuracy of MSET essentially depends on historical data that is similar to the observed data. Therefore, effectively constructing a process memory matrix based on probability characteristics is conducive to improving prediction accuracy. Considering that the unit load can represent the working state of the equipment, probability density equal-interval sampling is performed based on the load. When selecting the state vectors that compose the improved process memory matrix D, the purpose is to remove redundant or similar state vectors; the probability density equal-interval sampling method is used to sample the population, and fuzzy clustering is used to determine the state vectors in each interval. Typical state vectors are selected to form a memory matrix. The flow chart is shown as follows. Figure 1 As shown, the specific steps are as follows:

[0086] S1: After dimensionality reduction, denoising, and data cleaning of historical data, steady-state data is screened to provide a high-quality data set for the model;

[0087] S2: Taking the maximum and minimum values ​​of the unit load as the upper and lower limits, the filtered data is divided into M equal intervals, recorded as [L1, L2, ..., L i , …L M ];

[0088] S3: Use the probability density equal interval sampling method to sample the population, integrate the probability density on each equal interval, and obtain the load value of the state vector in the load interval L i The probability P i , multiplied by the total number of state vectors K in the process memory matrix, to obtain the number of state vectors K in each load interval i , K i =P i ×K;

[0089] S4: At each load interval L i Fuzzy clustering is used to determine the state vector X on each interval i , set the load interval L i The optimal number of clusters is set as K i ;

[0090] S5: Each load interval L i The selected state vector X i Add to the process memory matrix D until the number of Xi reaches Ki;

[0091] S6: Collect the primary wind turbine data in real time, and calculate the estimated vector X based on the observed vector and the acquired memory matrix estThe sliding window method is used to calculate the overall similarity of the vector. The overall similarity is the weighted mean of the basic similarity and the state similarity. Considering the stability of the primary fan equipment operation and the frequency of similarity changes, the current health of the primary fan is calculated. The health is the weighted mean of the overall similarity and the similarity fluctuation degree.

[0092] S7: The memory matrix is ​​updated regularly based on K nearest neighbors.

[0093] The steady-state data screening is performed based on the autocorrelation function, and the calculation formula is:

[0094]

[0095] Among them, X t is the observed value of the time series, is the mean of the observed values, and k is the lag order.

[0096] The dimensionality reduction process is based on the PCA algorithm. When faced with large-scale data sets, traditional PCA has a huge amount of computation because it needs to process the entire data set at once. Incremental PCA is a method that can process streaming data or large-scale data. It only processes one or a small batch of data samples at a time and gradually updates the principal components. For example, when processing real-time primary fan sensor data, data is generated continuously, and incremental PCA can perform dimensionality reduction on these new data in real time.

[0097] The steps include:

[0098] S101: Initialization

[0099] Standardize the first batch of data so that the mean of each feature is 0 and the variance is 1;

[0100] Calculate the covariance matrix C of the first batch of data. Suppose the matrix of the first batch of data is X1, the number of samples is n1, and the number of features is d, then the covariance matrix

[0101] Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1,λ2,…,λ d and the corresponding eigenvectors v1,v2,…,v d . Sort the eigenvalues ​​from large to small, and the corresponding eigenvectors are also sorted accordingly;

[0102] Select the eigenvectors corresponding to the first k largest eigenvalues ​​as the initial principal components, where k is the preset dimension after dimensionality reduction;

[0103] S102: Incremental Update

[0104] Let the new sample be x. When a new sample arrives, first calculate the new sample mean in is the mean of the previous n-1 samples, and updates the covariance moment Among them C old is the previous covariance matrix estimate;

[0105] S103: Update principal components

[0106] Perform eigenvalue decomposition on the updated covariance matrix, sort the new eigenvalues ​​from large to small, sort the corresponding eigenvectors accordingly, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the updated principal components;

[0107] S104: Repeated Update

[0108] As new data continues to arrive, the new data is continuously processed, steps S102 and S103 are repeated, and the covariance matrix and principal components are continuously updated, thereby realizing incremental principal component analysis of the data.

[0109] The probability density in the multivariate state estimation step adopts the formula of normal distribution:

[0110]

[0111] Here, μ is the mean and σ is the standard deviation.

[0112] Health assessment model Figure 2 As shown,

[0113] The steps of periodically updating the memory matrix based on K nearest neighbors are as follows: for the actually collected primary wind turbine parameter data, after dimension reduction, denoising, and data cleaning, the steady-state data is screened, and the screened observation vector is the distance sample to be classified, and the Euclidean distance between the screened observation vector and all vectors in the training set is calculated, and the calculated Euclidean distance is sorted to find the k training samples closest to the sample to be classified, and the categories of these k neighbors are counted, and the category of the sample to be classified is determined by voting, that is, it belongs to [L1, L2, ..., L i , …L M ] interval L i Furthermore, the population is sampled using the probability density equal interval sampling method, the state vector on each interval is determined using fuzzy clustering, and the typical state vector is selected to update the memory matrix.

[0114] Furthermore, an estimation vector is calculated based on the memory matrix and the observation vector, and the overall similarity is calculated based on the estimation vector. The overall similarity is the weighted mean of the basic similarity and the state similarity. Considering the stability of the primary fan equipment operation and the frequency of similarity changes, the current health of the primary fan is calculated and determined. The health is the weighted mean of the overall similarity and the similarity fluctuation degree. When the health exceeds the threshold, an early warning is issued.

[0115] The detailed algorithm steps of the fuzzy clustering are:

[0116] S401: Initialization: Select the number of clusters C and initialize the fuzzy membership matrix U, where U ij Represents data point x j Membership to cluster i. The membership value of each data point should be between [0,1], and the sum of all memberships for each data point j is 1:

[0117] S402: Calculate cluster centers: Calculate the center V of each cluster based on the current membership matrix U i :

[0118]

[0119] Where m is the fuzzy factor, which usually takes a value between (1, 2).

[0120] S403: Update the membership matrix: Calculate each data point x j The membership degree U of each cluster i ij

[0121]

[0122] S404: Determine the convergence and calculate the change of the membership matrix. If the stopping condition is met (such as the change is less than the set threshold, or the maximum number of iterations is reached), stop the iteration; otherwise, return to S402.

[0123] S405: Output the results, output the final cluster center V and fuzzy membership matrix U.

[0124] The MSET algorithm first needs to build a health status historical data memory matrix, that is, select several typical valid observation vectors from the historical data to form a historical matrix. The matrix contains as few vectors as possible and can almost represent the normal operating status of the equipment under all working conditions. The health status historical data memory matrix needs to collect as much system health status historical data as possible so that it can cover all system dynamic operating conditions as much as possible. The specific implementation process of MSET is as follows:

[0125] The memory matrix D is a matrix of size n×m, where n represents the number of n features of the primary wind turbine equipment state and m represents the number of observation vectors. i represents the parameter of the ith primary fan, x ij Represents the value of the i-th parameter at time j.

[0126] X j =[x 1j ,x2j ,L,x nj ] T

[0127]

[0128] The observation data vector is denoted by X obs Indicates that the expression is: X obs =[x1,x2,L,x n ] T

[0129] For each observation vector X obs , the MEST model generates a weight vector W based on the memory matrix D, W = [w1,w2,L,w m ] T

[0130] The weight vector W is the similarity between the current observation vector and the historical data memory matrix. From this, we can get the estimated vector X output by the model est .

[0131]

[0132] That is, the observation vector X obs The prediction output of the MSET algorithm is a linear combination of each historical observation vector. The weight vector W will directly affect the estimation effect of the model. W is determined by the minimum residual ε between the observation vector and the estimation vector. The smaller the absolute value of the residual, the higher the prediction accuracy of the MSET algorithm model.

[0133] ε=|X est -X obs |=|DW-X obs |

[0134] ε 2 =(DW-X obs ) T (DW-X obs )

[0135] minε 2 =min[(DW-X obs ) T (DW-X obs )]

[0136] When ε satisfies the minimization constraint When the weight vector W is

[0137]

[0138] is a nonlinear operation symbol, and the Euclidean distance is calculated as follows

[0139]

[0140] Finally, the estimated vector X est The expression is as follows:

[0141]

[0142] Most fault warning systems only judge the fault state by the residual value between the normal estimated value and the actual observed value of a variable. However, this residual judgment method has limitations. Sometimes the fault information is not only contained in a single variable, but also requires multiple variables or a series of behavioral state trends to verify the fault information. Moreover, the residual value of a single parameter is easily affected by environmental noise, which can easily affect the accuracy of fault warning. The judgment criteria described by the similarity between the observed state and the normal estimated state can more fully mine the abnormal information of multiple variables. A similarity measurement index based on fault evaluation between states is used to solve the limitation of the threshold of multivariate state estimation and multivariate simultaneous monitoring. The sliding window state similarity is used instead of the traditional residual threshold method as the evaluation criterion for early warning, which can better evaluate multiple fault-related variables in the state.

[0143] Similarity is a measure of the degree of similarity between two samples, which can represent the difference between the two. Euclidean distance is the most common measure of vector space distance:

[0144]

[0145] When an abnormality occurs, the estimated values ​​of the remaining variables of the improved MSET model may also exceed the normal range. Therefore, if the residual is used as the sole judgment criterion for fault warning, not only will important abnormal information be missed, but it may also lead to misjudgment. Therefore, by comprehensively considering the similarity between the observation vector and the estimation vector, and the similarity between the observation vector and the historical health memory matrix, the difference between the two can be measured to characterize the fault information of the equipment. In addition, since the amount of fault information contained in each input variable of the primary fan in the model is different, the measurement accuracy of each variable is different, and the data reliability is also different, the contribution of each variable to fault warning is also different. On this basis, a contribution weight of each variable to fault warning is assigned.

[0146] In engineering applications, the design formula limits the range of similarity measured by Euclidean distance to the range [0,1].

[0147] The basic similarity refers to the observation vector X obs With the estimated vector X est The closeness is as follows:

[0148]

[0149] S(Xo ,X e )∈(0,1]

[0150] Where α k represents the weight of the kth variable's contribution to fault warning, α k >0,S(X o ,X e )∈(0,1], the larger S is, the closer the two are.

[0151] State similarity refers to the observation vector X obs The degree of closeness to the memory matrix is ​​as follows:

[0152]

[0153] S(X o ,D)∈(0,1], the larger the S is, the closer the two are.

[0154] The overall similarity is the weighted average of the basic similarity and the state similarity:

[0155] S=β1S(X o ,X e )+β2S(X o ,D)

[0156] β1∈(0,1),β2∈(0,1)

[0157] S∈(0,1], the larger the S is, the higher the similarity between the primary fan and the historical normal operation state is.

[0158] Considering the stability of primary fan equipment operation and the frequency of similarity changes, the health degree is designed as follows:

[0159]

[0160] α1∈(0,1),α2∈(0,1)

[0161] H∈(0,1]

[0162] and Respectively represent the change in overall similarity at time t and the total overall similarity

[0163] For the real-time collected primary fan data, the sliding window method is used to eliminate the uncertainties and random interference in the equipment operation. The window width is set to N, and the overall similarity is calculated once per cycle. The overall health data [H1, H2, L, H N ], calculate the moving average of N consecutive health values ​​in the window:

[0164] The fault warning threshold Ht is determined based on the minimum average health value in the sliding window. When the device is operating normally, the minimum health value is calculated based on the minimum average similarity between the normal observation vector and the MSET model estimation vector as Ht. N , the fault warning threshold is: H AN =kH N , k is less than 1, which is the alarm threshold coefficient and is determined by the operator based on operating experience.

[0165] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims

1. A method for evaluating the health of a primary fan in a thermal power plant based on data fusion, characterized in that: The steps include: S1: After dimensionality reduction, denoising and data cleaning of historical data, steady-state data is screened to provide a data set for the model; S2: Taking the maximum and minimum values ​​of the unit load as the upper and lower limits, the filtered data is divided into M equal intervals, recorded as [L1, L2, ..., L i , …L M ]; S3: Use the probability density equal interval sampling method to sample the population, integrate the probability density on each equal interval, and obtain the load value of the state vector in the load interval L i The probability P i , multiplied by the total number of state vectors K in the process memory matrix, to obtain the number of state vectors K in each load interval i , K i =P i ×K; S4: At each load interval L i Fuzzy clustering is used to determine the state vector X on each interval i , set the load interval L i The optimal number of clusters is set as K i ; S5: Each load interval L i The selected state vector X i Add to the process memory matrix D until the number of Xi reaches Ki; S6: Collect the primary wind turbine data in real time, and calculate the estimated vector X based on the observed vector and the acquired memory matrix est , the sliding window method is used to calculate the overall similarity of the vector. The overall similarity is the weighted mean of the basic similarity and the state similarity. Considering the stability of the primary fan equipment operation and the frequency of similarity changes, the current health of the primary fan is calculated. The health is the weighted mean of the overall similarity and the similarity fluctuation degree; S7: Adopt K nearest neighbor based dynamic regular update of memory matrix.

2. According to claim 1, a method for evaluating the health of a primary fan in a thermal power plant based on data fusion is characterized in that: In step S1, the steady-state data screening is performed based on the autocorrelation function, and the calculation formula is: Among them, X t is the observed value of the time series, is the mean of the observed values, and k is the lag order.

3. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The dimensionality reduction process in step S1 is implemented based on the incremental PCA algorithm, and includes the following steps: S101: Initialization Standardize the first batch of data so that the mean of each feature is 0 and the variance is 1; Calculate the covariance matrix C of the first batch of data. Suppose the matrix of the first batch of data is X1, the number of samples is n1, and the number of features is d, then the covariance matrix Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues ​​λ1,λ2,…,λ d and the corresponding eigenvectors v1,v2,…,v d . Sort the eigenvalues ​​from large to small, and the corresponding eigenvectors are also sorted accordingly; Select the eigenvectors corresponding to the first k largest eigenvalues ​​as the initial principal components, where k is the preset dimension after dimensionality reduction; S102: Incremental Update Let the new sample be x. When a new sample arrives, first calculate the new sample mean in is the mean of the previous n-1 samples, and updates the covariance moment Among them C old is the previous covariance matrix estimate; S103: Update principal components Perform eigenvalue decomposition on the updated covariance matrix, sort the new eigenvalues ​​from large to small, sort the corresponding eigenvectors accordingly, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the updated principal components; S104: Repeated Update As new data continues to arrive, the new data is continuously processed, steps S102 and S103 are repeated, and the covariance matrix and principal components are continuously updated, thereby realizing incremental principal component analysis of the data.

4. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The probability density in step S3 adopts the formula of normal distribution: Here, μ is the mean and σ is the standard deviation.

5. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The detailed algorithm steps of the fuzzy clustering in step S4 are as follows: S401: Initialization: Select the number of clusters C and initialize the fuzzy membership matrix U, where U ij Represents data point x j For the membership of cluster i, the membership value of each data point should be between [0,1], and the sum of all memberships for each data point j is 1: S402: Calculate cluster centers: Calculate the center V of each cluster based on the current membership matrix U i : Where m is the fuzzy factor, and its value is between (1,2); S403: Update the membership matrix: Calculate each data point x j The membership degree U of each cluster i ij ; Determine the convergence, calculate the change of the membership matrix, and if the stop condition is met, such as the change is less than the set threshold, or the maximum number of iterations is reached, stop the iteration; otherwise, return to step S402; S405: Output the results, output the final cluster center V and fuzzy membership matrix U.

6. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The estimated vector X output by the model in step S6 est , obtained by the weight vector W, which is the similarity between the current observation vector and the historical data memory matrix; Finally, the estimated vector X est The expression is as follows:

7. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The overall similarity in step S6 is a weighted average of the basic similarity and the state similarity; The basic similarity refers to the observation vector X obs With the estimated vector X est The closeness is as follows: S(X o ,X e )∈(0,1] Where α k represents the weight of the kth variable's contribution to fault warning, S(X o ,X e )∈(0,1], the larger the S is, the closer the two are; State similarity refers to the observation vector X obs The degree of closeness to the memory matrix is ​​as follows: S(X o ,D)∈(0,1], the larger the S is, the closer the two are; The overall similarity is the weighted average of the basic similarity and the state similarity: S=β1S(X o ,X e )+β2S(X o ,D) β1∈(0,1),β2∈(0,1) S∈(0,1]， The larger the S is, the higher the similarity between the primary fan and the historical normal operating state is; Considering the stability of primary fan equipment operation and the frequency of similarity changes, the health degree is designed as follows: α1∈(0,1),α2∈(0,1) H∈(0,1] and They represent the change in the overall similarity at time t and the total overall similarity respectively.

8. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The primary fan data collected in real time in step S6 is subjected to the sliding window method to eliminate uncertainties and random interference in the operation of the equipment. The width of the window is set to N, and the overall similarity is calculated once per cycle to obtain the overall health data [H1, H2, L, H N ], calculate the moving average of N consecutive health values ​​in the window: The fault warning threshold Ht is determined according to the average health value in the sliding window. When the equipment is operating normally, the minimum health value is calculated based on the minimum average similarity between the normal observation vector and the MSET model estimation vector as Ht. N ,, the fault warning threshold is: H AN =kH N , k is less than 1, which is the alarm threshold coefficient and is determined by the operator based on operating experience.

9. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1 is characterized in that: The step S6 determines the current health of the primary wind turbine. If the early warning model determines that the primary wind turbine is faulty, the early warning signal will be promptly notified to the power station staff so that corresponding preventive measures can be taken. At the same time, the early warning information will be uploaded to the monitoring system of the power station to integrate the early warning system with the monitoring system of the power station. The data uploading method includes but is not limited to HttpAPI interface, WebSocket interface, and IEC104 interface.

10. The method for evaluating the health of a primary fan in a thermal power plant based on data fusion according to claim 1, characterized in that: In step S7, the memory matrix is ​​updated regularly using K nearest neighbors, and the selected observation vector is taken as the distance to the sample to be classified, and the Euclidean distance between the selected observation vector and all vectors in the training set is calculated, and the calculated Euclidean distances are sorted to find the k training samples closest to the sample to be classified, and the categories of these k neighbors are counted, and the category of the sample to be classified is determined by voting, that is, it belongs to [L1, L2, ..., L i , …L M ] interval L i .

Images